Humanoid Daily News — 11/15/2026

Humanoid Daily News — 11/14/2026

Humanoid Daily News — 11/13/2026

Humanoid Daily News — 11/12/2026

Humanoid Daily News — 11/11/2026

Humanoid Daily News — 11/10/2026

Humanoid Daily News — 11/09/2026

Humanoid Daily News — November 08, 2026

11/08/2026

Astribot T1 Performs Lab Task Under Teleoperation

Cable-driven humanoid handles a science-related manipulation sequence in a laboratory setting.

Read full story →

11/08/2026

Digit 5 Moves Through Industrial Workflow

Battery charging to vehicle assembly and box transport in warehouse settings.

Read full story →

11/08/2026

MARK One Tests Case Packing and Palletizing

Greek industrial humanoid advances toward first customer deployment.

Read full story →

11/08/2026

Dexterity-BEV Aligns Perception and Action in Shared 3D Frame

Framework for cross-hardware skill transfer and standardized robotics data.

Read full story →

11/08/2026

Avatar Robotics Tests Systems in HQ-Turned Warehouse

Over 100 bin positions; robots have helped ship a million products this year.

Read full story →

Humanoid Daily News — 11/07/2026

Humanoid Daily News — 11/06/2026

Humanoid Press — Daily Brief · November 6, 2026

Agility Robotics Debuts on Nasdaq as AGLT in First Pure-Play Humanoid Listing

The company completed its SPAC merger, secured roughly $465 million in cash, and holds more than $300 million in committed multi-year orders for Digit 5 while expanding Digit deployments with GXO.

By humanoid.press · Editor: Humanoid Press Team · November 6, 2026

Nasdaq AGLT

Video credit: Agility Robotics · Product suite overview ▶ Via YouTube

Agility Robotics completed its SPAC merger and began trading on the Nasdaq under the ticker AGLT on November 5, 2026. The listing makes it the first pure-play humanoid robotics company to trade publicly in the United States. The transaction provided the company with roughly $465 million in cash. Agility has secured more than $300 million in committed multi-year orders for its next-generation Digit 5 humanoid from approximately 30 customers, subject to contractual milestones. The company is also expanding its existing partnership with GXO Logistics, adding Digit robots at facilities in North Carolina and Tennessee. Those units will perform repetitive material-handling tasks in e-commerce fulfillment. Prior Digit units have already logged more than 65,000 hours of operation at customer sites. The video shows CEO Peggy Johnson walking through the broader product suite, including the robot, software stack, Arc fleet management, and service support. The listing and order figures are company-reported. Independent confirmation of the full cash balance after transaction costs has not been published separately.

The Nasdaq debut follows years of commercial deployments of earlier Digit generations at sites including Schaeffler, GXO, and Toyota Motor Manufacturing Canada. Digit 5 is designed for cooperatively safe work in closer proximity to people and is targeted for early-access customer units in the first half of 2027, with general availability by the end of that year. The $300 million-plus order book is contingent on the robot meeting undisclosed performance milestones in throughput, uptime, and placement accuracy. The GXO expansion continues a multi-year relationship in which Digit has handled tote and parcel movement inside logistics facilities. The additional North Carolina and Tennessee deployments focus on e-commerce fulfillment work that is repetitive and high-volume. No public breakdown of the exact number of new robots or the timeline for those specific sites has been released. The video provides an overview of the humanoid, the AI software stack, Arc, and the service team rather than a live demonstration of the new GXO sites or the listing itself. The result is a company-reported milestone in both capital markets and commercial expansion.

Agility has stated that the listing supplies capital to scale Digit 5 production and to support the transition from pilot deployments to larger fleets. The robot’s safety architecture, developed in part with FORT Robotics, is intended to allow operation outside traditional work cells. NVIDIA supplies onboard compute for Digit 5. The company has not yet shown production Digit 5 units publicly; prototypes are expected at CES 2027. The GXO expansion adds robots to existing logistics sites rather than introducing an entirely new application. E-commerce fulfillment involves repeated pick-and-place and tote-handling cycles under varying package sizes and weights. Prior Digits have performed similar work. No independent audit of the $465 million cash figure or of the $300 million order book has been published. Both numbers appear in company announcements surrounding the listing. The video remains a product-suite overview. It does not include financial details or footage from the North Carolina or Tennessee facilities. The listing is a completed transaction. The additional GXO deployments are described as underway or imminent.

The pure-play humanoid listing is distinct from earlier public robotics companies that combine humanoids with other product lines. Agility’s revenue to date has come from Digit deployments and related services. The company has reported substantial losses while scaling, consistent with the capital-intensive nature of humanoid development and manufacturing. The $465 million cash position is intended to fund production ramp and continued customer support. The $300 million Digit 5 order book provides a forward revenue indicator, subject to milestone achievement. GXO’s decision to add more Digits in North Carolina and Tennessee is presented as evidence of ongoing operational use in live facilities. No third-party verification of the exact cash proceeds after fees or of the contractual terms of the Digit 5 orders has been released. The video shows the CEO discussing the robot and supporting systems. Financial and deployment details are drawn from the listing announcements. The work is commercial for existing Digits and pre-production for Digit 5.

Agility Robotics began trading on the Nasdaq as AGLT after completing its SPAC merger. The company received roughly $465 million in cash and holds more than $300 million in committed multi-year orders for Digit 5. It is expanding Digit deployments with GXO Logistics at facilities in North Carolina and Tennessee for e-commerce fulfillment tasks. Prior Digits have logged more than 65,000 hours at customer sites. The video is CEO Peggy Johnson’s product-suite overview. The listing and order figures are company-reported. Independent confirmation of the post-transaction cash balance has not been published separately. The result is a completed public listing and an announced expansion of existing commercial deployments.

The Nasdaq listing under ticker AGLT marks the first pure-play humanoid company to trade in the United States. Agility has more than $300 million in Digit 5 orders subject to milestones and is adding robots with GXO in North Carolina and Tennessee. The video covers the product suite rather than the financial transaction or the specific new sites. All capital and order figures are company-reported. The additional GXO deployments continue work already performed by earlier Digits in logistics environments. The listing itself is complete.

Watch Product Suite →

Also Today

Standard Bots Expands US Production of AI-Native Industrial Robot Arms

Standard Bots, based in Glen Cove, New York, designs and assembles AI-native industrial robot arms in the United States. CEO Evan Beard has argued that the United States installs far fewer industrial robots than China and that domestic manufacturing of robots is needed to rebuild manufacturing capacity. The company makes its own circuit boards and many of its own components in-house. It is expanding its facility toward 70,000 square feet and aims to manufacture essentially all of its robots domestically by 2027. Standard Bots offers six-axis arms including the Core (18 kg payload) and Thor (30 kg payload). The robots are programmed primarily through demonstration rather than traditional coding. The company reports customers ranging from small manufacturers to larger organizations. It took seven years to sell the first 100 robots. The video shows the factory floor, the room where circuit boards are made, and Beard discussing the company’s history and goals. No independent verification of production volumes or market share has been published. The work is commercial-stage for the robot arms.

Standard Bots raised $200 million in a Series C round in 2026, reaching a reported $1 billion valuation. The funding is intended to expand US manufacturing capacity. Beard has testified before Congress on the need for a national strategy for robotics and American manufacturing. The company states that its robots can learn new tasks through demonstration in minutes and that the software stack is AI-native. The factory tour in the video includes the surface-mount technology area where boards are produced and the assembly floor. No specific unit-production numbers or current annual capacity figures appear in the public video. The company has said it is growing rapidly and that robots are already in use at customer sites. Independent confirmation of those deployments is limited to company statements and the video itself. The demonstration is a company documentary rather than a third-party audit of manufacturing output. The result is consistent with Standard Bots’ public positioning as a US-based producer of industrial robot arms.

Beard has said that foreign robots dominate the US market and that many manufacturers found existing systems too expensive or too difficult to program for a wide range of tasks. Standard Bots’ approach combines vertically integrated hardware production with demonstration-based learning so that operators can teach the robot a task by showing it. The video shows robots on the factory floor and Beard describing the company’s progress after years of limited sales. The company designs many of its own parts, including actuators, and assembles the final products in New York. By 2027 it plans to bring still more of the manufacturing process in-house. No third-party data on the number of robots currently in customer facilities or on average deployment times have been released. The video is an interview and factory tour. It does not include side-by-side comparisons with imported arms or quantitative productivity metrics. The work remains a company-reported expansion of US robot production.

The United States has installed a fraction of the industrial robots that China has installed in recent years, according to industry statistics cited by Beard. Standard Bots is one of a small number of companies attempting to manufacture AI-native robot arms domestically at scale. The video walks through the production process from circuit-board fabrication to final assembly and shows the company training its models on data from robots already in the field. Beard has spoken about the difficulty of the early years, including an intervention by friends who urged him to quit. The company continued and later raised substantial funding to expand capacity. No audited production figures or customer-deployment statistics are provided in the video. The demonstration is commercial. Standard Bots sells its arms for applications that include machine tending, welding, palletizing, and inspection. The footage is consistent with the company’s published manufacturing and product claims. Independent verification of output volume has not been released.

Standard Bots is expanding US production of AI-native industrial robot arms in Glen Cove, New York. The company designs and assembles its robots domestically and is increasing facility size with the goal of manufacturing essentially everything in the United States by 2027. CEO Evan Beard has highlighted the gap between US and Chinese robot installations and has testified on the need for domestic robotics manufacturing. The video shows the factory, including in-house circuit-board production, and Beard discussing the company’s history. No independent production or deployment statistics have been published. The work is a commercial expansion of US-based robot manufacturing. The demonstration is company-reported.

By humanoid.press · Editor: Humanoid Press Team · November 6, 2026

Standard Bots US Manufacturing

US-made

Arms

AI-native

Software

Glen Cove

Factory

Watch Factory Tour →

Nori Robotics Ships $1,688 A3 Bimanual Mobile Manipulator

Nori Robotics has launched the Nori A3, a 19-DoF bimanual mobile manipulator that ships assembled for $1,688. The robot has a 45 by 45 cm footprint, weighs 20.4 kg, and uses a three-stage telescoping column that raises the head and arms from 69 cm to 145 cm. Each arm has 7+1 degrees of freedom, a 55 cm reach, and a 1.5 kg payload. Perception includes four 720p RGB cameras and a lidar with 12 m range. Onboard compute is a Raspberry Pi 5; heavier policies run off-board. The robot has been shipping to customers since July 2026. Units are assembled in San Francisco. The company has opened limited additional batches for fall 2026. The A3 is positioned as an accessible platform for home tasks and for developers rather than a fully autonomous household appliance. The video is the launch demonstration showing the robot in household settings. No independent performance data or long-term reliability figures have been published. The work is commercial for the hardware.

The A3 uses commodity serial-bus servos protected by a multi-layer architecture that clamps goal positions, detects stalls, and applies a two-tier thermal interlock. A sensorless grip-force estimate is derived from the servo current register. The wheeled base is differential drive with passive casters. Battery life is listed at 6–8 hours. The robot includes a speaker and microphone for spoken commands. Software support includes a desktop app, teleoperation, an open SDK, and tools for recording demonstrations. Owners are expected to teach skills that can later be shared. The launch video shows examples such as fetching items, operating a coffee machine, and other household sequences. These are demonstration runs under controlled conditions. No success rates across varied home environments have been published. The price is a retail figure that includes assembly. Comparable research platforms have higher parts costs. The demonstration is company-reported. Full autonomy for arbitrary household tasks is not claimed. The result is a low-cost bimanual mobile manipulator that has entered customer shipments.

Nori Robotics is a San Francisco startup. The A3 is the design successor to an earlier model. The company has described the robot as recovering capabilities that are usually lost at this price point through software rather than more expensive hardware. Protection layers are intended to prevent thermal burnout of the commodity servos. The video shows the robot moving in a kitchen environment and performing short manipulation sequences. No multi-hour continuous operation or recovery-from-failure sequences appear in the clip. Shipping began in July 2026 with additional batches planned for fall. The company has stated that complete documentation continues to expand. The robot is sold as a full-price purchase rather than a deposit. Independent verification of payload, reach, or battery-life claims under real household conditions has not been published. The demonstration is a product launch video. The work remains early-stage commercial. Broader claims about fully autonomous home operation are not made.

The A3’s telescoping column provides floor-to-counter reach while keeping the overall footprint small enough for home use. Each arm can handle light objects. The four cameras are mounted on the grippers, head, and neck. Lidar supports navigation. The launch video illustrates the robot performing household-style tasks. These are presented as examples rather than exhaustive capability claims. Owners use the provided tools to record demonstrations and train additional skills. No public leaderboard or third-party evaluation of skill success rates has been released. The price of $1,688 is significantly lower than most bimanual research platforms. The company has been transparent that the robot is a platform for further development as well as a consumer product. The video is the official launch footage. It does not include long-duration reliability testing. The result is consistent with Nori’s published specifications and shipping timeline.

Nori Robotics has begun shipping the Nori A3, a $1,688 bimanual mobile manipulator with 19 degrees of freedom, a telescoping column, dual 1.5 kg arms, four cameras, and lidar. The robot has been shipping since July 2026 and is assembled in San Francisco. It is positioned as an accessible platform for home tasks and developers. The launch video shows household demonstration sequences. No independent performance metrics have been published. The work is commercial for the hardware. Full autonomy for arbitrary tasks is not claimed. The demonstration is company-reported.

By humanoid.press · Editor: Humanoid Press Team · November 6, 2026

Nori A3

$1,688

Price

19 DoF

Total

1.5 kg

Per arm

Watch A3 Launch →

Humanoid Press — Daily Brief · November 5, 2026

Daimon Robotics Shows Canvas Heat-Transfer Printing with Tactile Feedback at IROS 2026

The seven-step demo with three tool switches uses tactile sensing to confirm grip and iron pressure, since vision alone cannot reliably judge when the transfer is complete.

By humanoid.press · Editor: Humanoid Press Team · November 5, 2026

IROS 2026 Tactile

Video credit: Daimon Robotics · IROS 2026 ▶ Via YouTube

At IROS 2026, Daimon Robotics demonstrated a canvas heat-transfer printing task that requires long-horizon manipulation with multiple tool changes. The sequence consists of seven sequential steps and three tool switches. The robot picks and places a sticker, handles an iron, irons the transfer, returns the tool, and peels off the film. Approximately 70 percent of the task time is spent ironing. Vision alone cannot reliably determine whether the heat transfer is complete. The system uses tactile feedback to sense whether the sticker is securely held and whether the iron is applying sufficient pressure. The robot continuously adjusts its actions based on that tactile information. The video shows the full sequence. The demonstration is research- and conference-stage. No quantitative success rates or independent verification have been published with the clip. The work illustrates the use of tactile sensing for a multi-step process where visual confirmation is insufficient. No commercial product is claimed.

Heat-transfer printing on canvas requires sustained contact under controlled pressure and temperature. The robot must switch tools mid-task, maintain a stable grasp on the sticker and the iron, and decide when the transfer is finished. Daimon Robotics states that vision cannot reliably judge completion. Tactile sensing provides the missing signal for grip security and applied pressure. The video shows the robot adjusting throughout the ironing phase. The seven-step sequence includes tool changes, which adds complexity for long-horizon execution. No data on failure modes, average cycle time beyond the 70 percent ironing figure, or performance across multiple trials appear in the public description. The demonstration is a controlled showcase at IROS 2026. It does not include endurance runs or unstructured environments. The result is consistent with the company’s description of tactile feedback enabling greater adaptability on this task. The work remains experimental.

The canvas heat-transfer demo is one example of long-horizon manipulation that combines tool use, sustained contact, and a completion criterion that is difficult to observe visually. The robot picks the sticker, places it, retrieves the iron, irons, returns the iron, and peels the film. Tactile sensors inform whether the sticker remains secure and whether iron pressure is adequate. The video shows continuous adjustment during the ironing phase. No specific tactile sensor model or resolution is detailed in the public materials. The demonstration does not claim that the same approach generalizes to arbitrary multi-step tasks. It is presented as a case where tactile feedback makes a measurable difference for this process. Independent replication of the sequence has not been published. The footage is a conference demonstration. No field deployment or product release is announced. The result adds a concrete example of tactile sensing supporting long-horizon tool use to the set of IROS 2026 showcases.

Tool switching three times within a seven-step sequence requires the robot to maintain state across stages and re-grasp different objects. The ironing phase occupies most of the task duration. Continuous tactile monitoring allows the robot to adjust pressure and contact without relying solely on vision for completion. The video illustrates the full workflow from sticker placement through film peeling. No success-rate figures or comparison with a vision-only baseline are provided in the public description. The demonstration is research-stage. Daimon Robotics has framed the work as moving from seeing to feeling for complex tasks. The footage is consistent with that description. Broader claims about general-purpose long-horizon manipulation are not made. The work remains a conference demonstration rather than a deployed system.

Daimon Robotics showed a canvas heat-transfer printing demo at IROS 2026. The seven-step sequence with three tool switches uses tactile feedback to confirm sticker grip and iron pressure, because vision alone cannot reliably judge when the transfer is complete. Approximately 70 percent of the task time is spent ironing. The robot continuously adjusts based on tactile input. The video shows the full process. The demonstration is research- and conference-stage. No quantitative success rates have been published. Independent verification has not been released. The work illustrates tactile sensing for a multi-step task where visual confirmation is insufficient. No commercial product is claimed.

A Daimon Robotics demonstration at IROS 2026 performs canvas heat-transfer printing through seven sequential steps and three tool switches. Tactile sensing confirms that the sticker is held securely and that the iron applies sufficient pressure. Vision is described as insufficient for judging completion. The video captures the continuous adjustment during ironing and the subsequent film peeling. The demonstration is experimental. No field deployment is claimed. The result is a conference showcase of tactile feedback supporting long-horizon tool use and sustained contact. Further quantitative evaluation has not been released with the footage.

Watch IROS Demo →

Also Today

FAIRINO Shows Vision-Guided Parcel Sorting with Cobots in South Korea

FAIRINO has released a video of collaborative robots performing smart parcel sorting in South Korea. The demonstration shows a robotic arm identifying and sorting parcels with vision guidance. FAIRINO is a Chinese collaborative-robot manufacturer that supplies cobots for pick-and-place, sorting, palletizing, and other industrial tasks. The company has an established presence in South Korea through local partners and has previously shown vision-based sorting and pick-and-place applications in the country. The video is a product and application showcase. No throughput figures, accuracy rates, or cycle times are published with the clip. The demonstration is commercial-stage for the cobot hardware. Specific system-integration details for the Korean installation are not provided in the public video. Independent performance data have not been released. The footage illustrates vision-guided parcel handling with a FAIRINO arm. The work is an application demonstration rather than a research result.

Parcel sorting is a common logistics application for collaborative robots. Vision systems identify parcel size, orientation, or destination markings, and the arm picks and places items into the correct location. FAIRINO cobots are used for similar pick-and-place and sorting tasks in multiple countries. The South Korea video shows the arm operating in a sorting context. No information on the vision software vendor, sensor model, or integration partner appears in the video description. FAIRINO markets its robots for logistics, manufacturing, and service applications. The company reports growing export volumes and has local representation in Korea. The video does not include multi-hour continuous operation or error-recovery sequences. It is a short application clip. No comparison with manual sorting or with other cobot brands is provided. The demonstration is consistent with FAIRINO’s published application portfolio. Commercial availability of the cobots is established. System-level performance for this specific installation remains company-reported.

FAIRINO positions its collaborative robots for tasks that include bin picking, pick-and-place, and sorting. The South Korea parcel-sorting video is one example of that positioning. The arm is shown identifying and relocating parcels under vision guidance. No payload, reach, or repeatability figures for the specific robot model are stated in the video. Those specifications are available on the company’s product pages for its FR series cobots. The video does not show dual-arm coordination or mobile-base integration. It is a fixed-arm sorting demonstration. Independent verification of sorting accuracy or throughput for the Korean site has not been published. The footage is a marketing and application video. It does not claim novel research results. The demonstration establishes that FAIRINO cobots are being shown in Korean logistics sorting contexts. Further operational data have not been released with the clip.

Vision-guided sorting reduces the need for fixed mechanical guides when parcel sizes or orientations vary. The FAIRINO video shows the arm responding to visual input to pick and place parcels. No details on lighting conditions, camera placement, or handling of damaged or non-standard parcels appear in the public description. The company has previously posted similar vision-sorting clips from Korean sites. The current video continues that series of application demonstrations. No failure cases or recovery behaviors are shown. The demonstration is commercial. FAIRINO cobots are sold through distributors in multiple countries, including South Korea. The video is an illustration of one deployed or demonstrated use case. Quantitative performance claims are not made in the clip. The result is consistent with the company’s logistics application marketing.

FAIRINO has shown collaborative robots performing vision-guided parcel sorting in South Korea. The video depicts a robotic arm identifying and sorting parcels. FAIRINO supplies cobots for pick-and-place and sorting applications and has a presence in the Korean market. No throughput or accuracy figures are published with the clip. The demonstration is a commercial application showcase. Independent performance data have not been released. The footage is consistent with FAIRINO’s published logistics examples. The work is an application demonstration rather than a research result. Commercial availability of the cobot hardware is established. System-level details for the specific installation are not provided.

By humanoid.press · Editor: Humanoid Press Team · November 5, 2026

FAIRINO Sorting

Vision

Guidance

Korea

Location

Cobot

Platform

Watch Sorting Demo →

Riemann Dynamics Shows Dual-Arm Capsule Coffee Making

Riemann Dynamics has released a video of its robot performing a multi-step capsule coffee task. The sequence includes placing a cup, opening the lever, picking up a capsule, aligning and inserting it, closing the lever, pressing the button, waiting for brewing, and then carefully lifting and placing the filled cup. The demonstration requires precise placement of a small capsule, coordinated dual-arm operation for the lever and button, and stable transport of a full cup without spilling. The video is titled as an example of a World Action Model capability. Riemann-1.0 is the company’s embodied World Action Model, described in an arXiv paper as a causal autoregressive model trained on large-scale interaction data. The coffee video is a real-world long-horizon manipulation example. No success rates for this specific coffee task are published with the clip. The demonstration is research- and company-showcase stage. Independent verification has not been released. The footage shows the full sequence from capsule insertion to cup placement.

Capsule coffee making combines several manipulation challenges: handling a small, lightweight capsule, operating a mechanical lever and button, coordinating two arms, and carrying a liquid-filled cup. The video shows the robot completing these steps in sequence. Riemann Dynamics has published results for Riemann-1.0 on long-horizon real-world tasks, reporting an 85 percent success rate and 94.4 percent progress success rate across a set of household-style tasks. The coffee demonstration is presented separately as a concrete example. No trial count or success rate is given for the coffee sequence itself. The model is described as unifying policy execution and action-conditioned world simulation. The video does not include failure cases or recovery. It is a successful run under the conditions shown. The work remains experimental. No commercial coffee-making product is claimed. The demonstration is consistent with the company’s focus on long-horizon dual-arm manipulation.

The coffee sequence requires both precision and coordination. The capsule must be aligned with the machine opening. The lever and button require appropriate force and timing. Dual-arm use allows one arm to stabilize or assist while the other performs the primary action. The final cup transfer must avoid spilling. The video shows the robot completing these stages without visible human intervention during the run. No details on the specific robot hardware or the exact policy used for this demonstration appear in the public video description. Riemann-1.0 has been evaluated on simulation benchmarks including LIBERO, RoboTwin 2.0, and RoboCasa-365, as well as real-world long-horizon tasks. The coffee video is an additional real-world illustration. Independent replication of the coffee sequence has not been published. The demonstration is company-reported. The footage is a showcase of multi-step dual-arm manipulation for a familiar household task.

World Action Models jointly predict actions and future visual states. Riemann-1.0 is presented as one such model that can serve as both an executable policy and a world simulator. The coffee video shows the policy executing a multi-step task rather than the simulation mode. The sequence is short enough to fit in a single demonstration but still contains multiple precise contacts and tool-like interactions with the machine. No quantitative force or placement-accuracy data are provided with the clip. The demonstration does not claim that the same policy generalizes to arbitrary coffee machines or cup types. It is a controlled example. The company has invited suggestions for future tasks. The video remains a research and marketing demonstration. No product timeline is announced. The result is consistent with published descriptions of Riemann-1.0’s long-horizon capabilities.

Riemann Dynamics has shown a dual-arm robot making capsule coffee through a multi-step sequence that includes capsule insertion, lever and button operation, and careful cup placement. The video is presented as an example of World Action Model capability. Riemann-1.0 is the company’s causal World Action Model. No success rate for the coffee task itself is published with the clip. The demonstration is research- and company-showcase stage. Independent verification has not been released. The footage shows the full sequence. The work illustrates long-horizon dual-arm manipulation on a familiar task. No commercial product is claimed. The result is consistent with the company’s published focus on multi-step real-world manipulation.

By humanoid.press · Editor: Humanoid Press Team · November 5, 2026

Riemann Coffee

Dual-arm

Coordination

Capsule

Task

WAM

Model

Watch Coffee Demo →

Humanoid Press — Daily Brief · November 4, 2026

PAKE Uses Partial Kinematic Embeddings for Whole-Body Loco-Manipulation

Researchers report a hierarchical framework that generates partial kinematic references and trains high- and low-level controllers, achieving 4.5 cm end-effector tracking on a quadruped with a six-DoF arm.

By humanoid.press · Editor: Humanoid Press Team · November 4, 2026

PAKE Loco-Manipulation

Video credit: Research demonstration · arXiv:2607.11041 ▶ Via YouTube

Researchers have introduced PAKE, a whole-body control framework for loco-manipulation that decomposes the problem into partial kinematic reference generation and low-level imitation control. The method uses a Kinematic Normalizing Flow model trained on a large-scale kinematic dataset to generate diverse partial references. A high-level controller navigates the latent space of that model to exploit kinematic redundancy, while a low-level controller produces dynamically consistent motion. The approach was validated on a quadrupedal robot equipped with a six-degree-of-freedom arm. In hardware tests across 24 episodes and eight mobile loco-manipulation tasks, the system reported end-effector pose-tracking errors of 4.5 cm and 0.14 rad, with locomotion tracking errors of 0.1 m/s linear and 0.01 rad/s angular. The paper was submitted in July 2026 and revised in October. The video shows the system performing coordinated locomotion and manipulation. The work is research-stage. No commercial product is claimed. Independent replication of the reported tracking errors has not been published. The demonstration establishes the framework’s performance under the tested conditions.

Loco-manipulation requires simultaneous control of a mobile base and a manipulator while exploiting the redundancy of a high-DoF system. PAKE addresses the high-dimensional action space by separating reference generation from execution. The Kinematic Normalizing Flow produces feasible partial kinematic trajectories. The high-level policy selects among redundant solutions in the latent space. The low-level policy then tracks the resulting motion under dynamics. Hardware evaluation covered multiple mobile tasks. The reported end-effector accuracy is 4.5 cm position and 0.14 rad orientation. Base velocity tracking remained within 0.1 m/s and 0.01 rad/s. Simulation results on a 4,000-target benchmark showed improved tracking accuracy and workspace coverage relative to evaluated baselines. The video illustrates the coordinated whole-body behavior. No data on maximum payload, continuous operation duration, or performance on untested terrains appear in the abstract. The framework remains experimental. The authors describe it as a practical solution for accurate whole-body loco-manipulation in high-DoF systems. Further downstream applications are noted as potential rather than demonstrated.

The quadruped-plus-arm platform used for validation allows omnidirectional legged mobility while the arm performs manipulation. PAKE generates partial torso-arm references and uses low-level imitation to preserve base mobility. The hierarchical design is intended to expand the reachable workspace without sacrificing locomotion accuracy. Hardware episodes spanned eight different mobile loco-manipulation tasks. Tracking errors are reported as means across those episodes. The video shows examples of the system moving while the arm tracks targets. No comparison tables or full ablation studies appear in the video itself. Those details are in the paper. The method does not claim zero-shot transfer to arbitrary new robots or environments. It is validated on the specific quadruped-arm hardware. The work is academic. No open-source release of the trained models or code is mentioned in the abstract. The demonstration provides a concrete example of hierarchical whole-body control that explicitly models kinematic redundancy. Scaling to full humanoids or longer horizons is not claimed.

End-effector tracking of 4.5 cm and 0.14 rad was achieved while the base continued to move. The paper states that the system maintained accurate locomotion tracking throughout the hardware episodes. The video shows the robot executing coordinated motions rather than stopping to manipulate. The Kinematic Normalizing Flow was trained on a large-scale kinematic dataset to produce diverse feasible partial references. The high-level controller learns to select among those references. The low-level controller handles dynamic consistency. The approach is hierarchical rather than a single end-to-end policy. Independent verification of the 24-episode hardware results has not been published. The authors report outperformance relative to competitive baselines in simulation. The video is a research demonstration of the framework. It does not include failure cases or long-duration runs. The result is consistent with the abstract’s description of accurate coordinated tracking. The work remains at the research stage.

PAKE is a whole-body loco-manipulation framework that uses partial kinematic embeddings generated by a Kinematic Normalizing Flow. A high-level controller selects redundant solutions; a low-level controller executes the motion. On a quadruped with a six-DoF arm, hardware tests across 24 episodes reported 4.5 cm and 0.14 rad end-effector tracking errors while maintaining base velocity tracking. The video shows the coordinated behavior. The paper is available on arXiv. The work is research-stage. No commercial product is claimed. Independent replication has not been published. The demonstration establishes the reported tracking performance under the tested conditions. The framework is presented as a practical approach for high-DoF whole-body control.

Researchers have shown PAKE performing whole-body loco-manipulation on a quadrupedal robot with an arm. The method decomposes the problem into partial kinematic reference generation and low-level imitation. Hardware results include 4.5 cm position and 0.14 rad orientation end-effector tracking across multiple mobile tasks. The video illustrates the system in operation. The work is experimental. No field deployment or product release is claimed. The demonstration is consistent with the published abstract. Further details are in the arXiv paper. The result adds a hierarchical approach that explicitly models kinematic redundancy to the set of whole-body loco-manipulation methods under active research.

Paper arXiv →

Also Today

Demonstration Shows Sustained Multi-Step Manipulation with Precise Contact

A research video demonstrates sustained multi-step robotic execution that requires precise contact, including inner-pot placement and button presses. The work is framed as progress on whole-body intelligence for long-horizon tasks. The sequences show the robot continuing through multiple stages rather than resetting between actions. Inner-pot placement demands accurate positioning inside a container. Button presses require controlled force and location. The demonstration is experimental. No success rates, trial counts, or quantitative precision measurements are published with the video. Independent verification has not been released. The video is a research illustration of multi-step performance under the tested conditions. It does not claim a commercial system. Broader generalization to arbitrary long-horizon tasks is not asserted. The footage shows continuous coordinated motion for the depicted actions. The result remains at the laboratory stage.

Long-horizon manipulation requires sequencing actions, maintaining state, and recovering from small errors. The video shows the robot completing multi-stage sequences that include placing an object inside a pot and pressing buttons. Precise contact is emphasized for both the insertion and the button interaction. The demonstration does not specify the robot platform or the exact control architecture in the public title and description. Whole-body intelligence approaches in the broader literature aim to coordinate locomotion, balance, and manipulation rather than treating them as separate modules. The video focuses on the manipulation outcomes. No data on failure modes, recovery time, or performance under disturbance appear in the public clip. The result is a controlled demonstration of sustained multi-step execution. It remains experimental. No deployment timeline is announced. The footage is consistent with ongoing research on long-horizon tasks that require both coordination and contact precision.

Inner-pot placement and button presses are examples of tasks where small positioning or force errors can cause failure. The video shows the robot completing these actions as part of longer sequences. The demonstration does not include force-sensor traces or millimeter-level accuracy figures. For researchers working on whole-body control, the relevant question is how high-level task goals are grounded in low-level coordinated motion over many steps. The video provides a visual example of such execution. It does not detail the underlying policy, training data, or perception stack. Independent replication of the shown sequences has not been published. The work is experimental. No commercial product is associated with the demonstration. The footage illustrates sustained multi-step performance with precise contact under the tested conditions. Further technical results have not been released with the video.

Sustained execution over multiple steps is distinct from single-action demonstrations. The video shows the robot continuing through a sequence rather than resetting between actions. Precise contact is required for the pot placement and the button presses. The demonstration does not report how many times the full sequence was attempted or the success rate. Whole-body approaches aim to keep the entire robot coordinated so that arm motions do not destabilize balance or previous placements. The video focuses on the end-effector and object outcomes. No full-body kinematics or balance metrics are provided in the public description. The result is a research showcase of long-horizon precise manipulation. It remains laboratory-stage. No field deployment is claimed. The footage is an illustration of the multi-step capability under the conditions shown.

A research video demonstrates sustained multi-step robotic manipulation with precise contact, including inner-pot placement and button presses. The work is framed as progress on whole-body intelligence for long-horizon tasks. The sequences involve coordinated motion over multiple stages. No success rates or quantitative precision data are published with the video. The demonstration is experimental. Independent verification has not been released. The footage shows continuous task execution under the tested conditions. No commercial product is claimed. The video is a research illustration of multi-step precise manipulation. Broader generalization is not asserted. The work remains at the laboratory stage.

By humanoid.press · Editor: Humanoid Press Team · November 4, 2026

Long-horizon Manipulation

Multi-step

Execution

Precise

Contact

Research

Stage

Watch Demo →

Generalist’s GEN-1.5 Shows One-Shot Physical Learning from a Single Demo

Generalist AI has released GEN-1.5, a robot foundation model that can learn a new physical task from a single short demonstration without gradient updates. The company describes the model as a large multimodal system that processes video (with a 30-second context window), other sensors, language, and proprioception, and outputs action trajectories at 100 Hz. A 3- to 12-second demonstration is placed in the context window as a “physical prompt.” The model then attempts the task. Across ten short-horizon manipulation tasks, Generalist reported an average one-shot success rate of 59 percent (±10 percentage points). With ten gradient steps on about five minutes of data per task, the average rose to 83 percent. The tasks shown are simple and short. All results come from the company. Independent verification has not been published. The video demonstrates the one-shot capability on several examples. No public model weights or API have been released. The work is research-stage. The company frames the result as an early step toward general intelligence for the physical world.

GEN-1.5 was introduced in an August 2026 company blog post and accompanying video. The one-shot behavior emerged from more than eight months of pretraining on physical interaction data; the company states that no architectural changes or meta-learning objectives were added specifically to encourage in-context learning. The model can chain short prompts into longer sequences and can use demonstrations recorded in simulation. Zero-shot generalization examples, including novel tool use, are also shown. The reported 59 percent average is across ten internal tasks such as opening a jar or retrieving an item from a wallet. The standard deviation is 10 percentage points. Fine-tuning with limited additional data improves the figure. The video shows the model receiving a physical prompt and then executing the task. No long-horizon or multi-object household tasks are included in the published results. The demonstration remains company-reported. Broader claims about general-purpose autonomy are not made. The result is an illustration of in-context physical learning under the tested conditions.

Physical prompting places a short sensorimotor demonstration directly into the model’s context window. The model then generates actions without updating its weights. Generalist reports that this capability appeared during large-scale pretraining rather than through explicit training for few-shot learning. The video includes examples of the robot performing tasks after seeing a single demonstration. Success rates are averages across the ten-task suite. Individual task results are not broken out in the public summary. The model also supports a separate fine-tuning mode that uses one to ten gradient steps on one to five minutes of data. The company has not released the model for external evaluation. No peer-reviewed paper accompanies the announcement. The video is a research demonstration of the one-shot behavior. It does not include failure analysis or comparisons with other foundation models on the same tasks. The work remains proprietary and experimental. Independent replication is not possible from the published materials.

The tasks used for the 59 percent figure are short-horizon manipulation actions. The company notes that the results are the first it knows of in which one-shot and few-shot learning of physical skills have emerged at this scale from pretraining alone. The video shows both the prompt and the subsequent model rollout. No data on continuous multi-hour operation or on performance outside the tested objects and environments have been released. GEN-1.5 follows the earlier GEN-1 model. The company describes the one-shot capability as an unexpected emergence rather than an engineered feature. The demonstration is consistent with the blog post. All quantitative claims are attributed to Generalist. No third-party evaluation has been published. The video is a company showcase of the reported capability. The result is research-stage. No product or API release is announced.

Generalist AI’s GEN-1.5 is a robot foundation model that learns new short-horizon physical tasks from a single 3- to 12-second demonstration without gradient updates. The company reported a 59 percent average success rate across ten tasks in one-shot mode and 83 percent after limited fine-tuning. The model processes video, sensors, language, and proprioception and outputs actions at 100 Hz. The video shows the physical-prompting process and subsequent execution. All results are company-reported. Independent verification has not been published. No public weights or API exist. The work is research-stage. The demonstration illustrates one-shot physical learning under the tested conditions. Broader general-purpose claims are not made.

By humanoid.press · Editor: Humanoid Press Team · November 4, 2026

GEN-1.5 One-shot

59%

One-shot avg

3–12 s

Prompt length

100 Hz

Actions

Read GEN-1.5 Blog →

Humanoid Press — Daily Brief · November 3, 2026

HEXACERCLE M11 Glove Teleoperates BrainCo Revo3 for Fine Tasks

The demonstration pairs a data glove with the 21-DoF Revo3 hand for finger opposition, bean picking with tweezers, and stable grasping under low-latency control.

By humanoid.press · Editor: Humanoid Press Team · November 3, 2026

Teleoperation Revo3

Video credit: HEXACERCLE / BrainCo demonstration ▶ Via YouTube

A demonstration video shows the HEXACERCLE M11 data glove teleoperating BrainCo’s Revo3 dexterous hand on fine-motor tasks. The glove captures human hand motion and maps it to the robotic hand in real time. The video includes finger opposition, bean picking with tweezers, stable grasping tests, and low-latency control. The bean-picking sequence also appears in materials associated with the 2nd World Humanoid Robot Games. Revo3 is a 21-degree-of-freedom hand with independent motors in each joint, designed for research, teleoperation, and humanoid integration. HEXACERCLE describes the M11 as a high-precision motion-capture glove for collecting human manipulation data. The video is a teleoperation showcase rather than an autonomous performance. No quantitative latency figures, success rates, or endurance data are published with the clip. The demonstration establishes that the glove-to-hand mapping can support precise finger motions and tool use under operator control. Independent verification of the mapping accuracy has not been released. The work remains at the demonstration stage.

BrainCo’s Revo3 features 21 fully active degrees of freedom, with five in the thumb and four in each of the other fingers. The joints are direct-drive and back-drivable. The hand supports multiple interfaces and has an open SDK with ROS 2 and simulation support. Teleoperation integrations with data gloves, including MANUS systems, have been shown by the company. The HEXACERCLE video specifically pairs the M11 glove with Revo3 for the listed tasks. Finger opposition and tweezer-based bean picking require coordinated multi-finger control and precise tip placement. The video shows the operator’s hand movements mirrored on the robot hand. No haptic feedback loop is described in the public clip. The demonstration is operator-driven. The mapping is not claimed to be fully autonomous. For data-collection workflows, the relevant capability is low-latency retargeting of human hand motion to the 21-DoF hand. The video illustrates that capability on the shown tasks. Broader claims about unsupervised fine manipulation are not made.

The bean-picking task with tweezers is presented as an example of fine motor control. The operator uses the glove to guide the Revo3 hand in selecting and moving individual beans. The video does not report error rates or average cycle times. Stable grasping tests are also shown, in which the hand maintains holds on objects under operator command. Low-latency teleoperation is listed as a feature of the demonstration. No millisecond figures are provided in the video description. HEXACERCLE has positioned its gloves for humanoid robotics, imitation learning, and large-scale data collection. The M11 Pro variant adds tactile sensing points, though the video focuses on motion capture. The demonstration is consistent with existing glove-to-dexterous-hand teleoperation practice. It does not introduce a new autonomous policy. The footage is a controlled showcase of the paired hardware. Commercial availability of the specific integration is not detailed beyond the demonstration itself.

Revo3 is intended for industrial automation, scientific research, and advanced robotic manipulation. Its 21 independent motors enable opposition, lateral motion, and multi-finger coordination. The HEXACERCLE demonstration uses that range of motion under human control. The video shows continuous teleoperation rather than scripted sequences. No external motion-capture system beyond the glove is described. The operator remains in the loop for all shown actions. For researchers collecting demonstration data, the setup provides a direct path from human hand motion to robot joint commands. The video does not include autonomous execution after data collection. It remains a teleoperation example. Independent testing of the system’s latency or tracking precision under varied conditions has not been published with the clip. The demonstration is research- and product-showcase stage. No field deployment is claimed.

The HEXACERCLE M11 data glove is shown teleoperating BrainCo’s Revo3 21-DoF dexterous hand. The video includes finger opposition, bean picking with tweezers, stable grasping, and low-latency control. The bean-picking task has also appeared in World Humanoid Robot Games materials. Revo3 uses independent direct-drive motors and supports open software interfaces. The demonstration is operator-controlled. No quantitative success rates or latency numbers are published with the video. The footage illustrates precise teleoperated fine-motor tasks. Independent verification has not been released. The work is a hardware and mapping demonstration rather than an autonomous system result.

A data glove from HEXACERCLE is paired with BrainCo’s Revo3 hand for real-time teleoperation of fine tasks. The video shows coordinated finger motions, tool use with tweezers, and stable grasps under operator control. The system is not presented as autonomous. The demonstration establishes that human hand motion can be retargeted to the 21-DoF hand with sufficient fidelity for the shown actions. Broader performance claims are not made. The footage is a controlled teleoperation showcase. The integration remains at the demonstration stage. No commercial product timeline beyond the shown hardware is announced in the video. The result is consistent with ongoing work on glove-based data collection for dexterous robotic hands.

Watch M11 + Revo3 →

Also Today

Open Duck Mini V2 Is an Open-Source 42 cm Biped That Walks with RL

Open Duck Mini V2 is an open-source miniature bipedal robot inspired by Disney’s BDX droid. It stands about 42 cm tall with legs extended and is designed to be built for under $400 using 3D-printed parts and off-the-shelf components. The platform uses 14 Feetech STS3215 serial bus servos, a Raspberry Pi Zero 2 W, a BNO055 IMU, and foot contact switches. Walking policies are trained with reinforcement learning in MuJoCo and deployed as ONNX models on the onboard computer. The project, led by Antoine Pirrone of Pollen Robotics, releases hardware designs, firmware, and trained policies under the Apache 2.0 license. A short video by Masato Kobayashi shows the assembly of one unit in roughly 60 seconds. The robot is a research and hobby platform rather than a commercial product. No quantitative walking reliability figures or long-distance endurance data are published with the build video. The design is community-supported. Independent builds have been documented by makers. The video is a time-lapse assembly demonstration of the open hardware.

Each leg has five degrees of freedom (hip yaw, roll, pitch, knee, and ankle). The head and neck add four more actuated joints. Policies are trained in simulation using the Open Duck Playground framework and transferred to the physical robot. The runtime runs on the Raspberry Pi Zero 2 W. The video by Kobayashi is a condensed build sequence rather than a walking demonstration. Public materials show the robot taking steps after assembly when a trained policy is loaded. The project provides assembly guides, a bill of materials, and ONNX policy files. Community members have reported successful walks as well as occasional falls during early testing. No formal success-rate benchmarks across multiple builders have been published. The platform is intended for experimentation with sim-to-real locomotion and low-cost bipedal hardware. It is not presented as a production robot. The open-source release allows others to replicate and modify the design. The video illustrates one builder’s assembly process.

The Open Duck Mini project began as a smaller v1 and evolved into the taller, more robust V2. Sponsorship and community support have come from sources including Hugging Face and Pollen Robotics. The mechanical design is largely finalized, with ongoing work on policy robustness. The Kobayashi video focuses on the physical build rather than software setup or walking. Builders typically 3D-print the structural parts, install heat-set inserts, wire the servos on a serial bus, and flash the runtime. The full bill of materials is published. Cost estimates remain under $400 for the core components, excluding tools and a printer. The video is a personal build log. It does not claim to represent every possible assembly outcome. For researchers and makers interested in affordable bipedal platforms, the relevant point is that a walking policy can be transferred from simulation to a low-cost, open hardware robot. The demonstration is consistent with the project’s public materials. Further policy improvements continue in the community.

Walking on Open Duck Mini V2 is policy-driven rather than hand-tuned. The ONNX models run in real time on the Raspberry Pi. The IMU and foot switches provide feedback. The video does not show the finished robot walking; it shows the construction. Other public clips and project documentation confirm that trained policies produce bipedal steps. Failures such as forward falls have been reported by some builders during initial runs. The project maintains a Discord community and GitHub repository with guides. No commercial support contract is offered. The platform is explicitly open-source. The Kobayashi video is one example of a completed build. It does not include performance metrics. The result is a documented path from CAD files and a trained policy to a physical walking robot under the stated cost target. Scaling or industrial use is not claimed. The work remains in the open-hardware and research domain.

Open Duck Mini V2 is an open-source 42 cm bipedal robot that walks using reinforcement-learning policies trained in MuJoCo. It uses 14 Feetech servos, a Raspberry Pi Zero 2 W, and 3D-printed parts, with a target build cost under $400. The project is led by Antoine Pirrone and released under Apache 2.0. A video by Masato Kobayashi shows one assembly in about 60 seconds. The platform is a hobby and research tool. No formal reliability benchmarks are published with the build video. Independent builds exist. The video is a time-lapse of construction. Walking capability is confirmed in other project materials. The design remains open and community-supported. No commercial product is claimed.

By humanoid.press · Editor: Humanoid Press Team · November 3, 2026

Open Source Biped

42 cm

Height

14

Servos

<$400

Target cost

Watch Build Video →

Researchers Show Sustained Multi-Step Manipulation with Precise Contact

A research demonstration shows a robot performing sustained multi-step tasks that require precise contact, including inner-pot placement and button presses. The video is titled as work on building whole-body intelligence for long-horizon tasks and precise manipulation. The sequences involve coordinated arm, hand, and body motion over multiple stages rather than isolated single actions. Inner-pot placement requires accurate positioning inside a container. Button presses require controlled force and location. The demonstration is experimental. No success rates, trial counts, or comparison baselines are published with the video. The work addresses the challenge of maintaining precision and recovery across extended task sequences. Independent verification has not been released. The video is a research illustration of multi-step execution. It does not claim a commercial system or a general-purpose product. The footage shows continuous task performance under the conditions tested. Broader generalization claims are not made in the public description.

Long-horizon manipulation requires the robot to sequence actions, maintain state across steps, and recover from small errors without resetting. The video shows the robot completing multi-stage sequences that include placing an object inside a pot and pressing buttons. Precise contact is emphasized for both the insertion and the button interaction. The demonstration does not specify the robot platform or the exact control method in the title and description provided. Whole-body intelligence approaches in the broader literature coordinate locomotion, balance, and manipulation rather than treating them as separate modules. The video focuses on the manipulation outcomes. No data on failure modes, recovery time, or performance under disturbance appear in the public clip. The result is a controlled demonstration of sustained multi-step execution. It remains at the research stage. No deployment timeline is announced. The footage is consistent with ongoing work on long-horizon robotic tasks that require both coordination and contact precision.

Inner-pot placement and button presses are examples of tasks where small positioning or force errors can cause failure. The video shows the robot completing these actions as part of longer sequences. The demonstration does not include quantitative precision measurements or force-sensor traces. For researchers working on whole-body control, the relevant question is how high-level task goals are grounded in low-level coordinated motion over many steps. The video provides a visual example of such execution. It does not detail the underlying policy architecture, training data, or perception stack. Independent replication of the shown sequences has not been published. The work is experimental. No commercial product is associated with the demonstration. The footage illustrates sustained multi-step performance with precise contact under the tested conditions. Further technical results have not been released with the video.

Sustained execution over multiple steps is distinct from single-action demonstrations. The video shows the robot continuing through a sequence rather than resetting between actions. Precise contact is required for the pot placement and the button presses. The demonstration does not report how many times the full sequence was attempted or the success rate. Whole-body approaches aim to keep the entire robot coordinated so that arm motions do not destabilize balance or previous placements. The video focuses on the end-effector and object outcomes. No full-body kinematics or balance metrics are provided in the public description. The result is a research showcase of long-horizon precise manipulation. It remains laboratory-stage. No field deployment is claimed. The footage is an illustration of the multi-step capability under the conditions shown.

A research video demonstrates sustained multi-step robotic manipulation with precise contact, including inner-pot placement and button presses. The work is framed as progress on whole-body intelligence for long-horizon tasks. The sequences involve coordinated motion over multiple stages. No success rates or quantitative precision data are published with the video. The demonstration is experimental. Independent verification has not been released. The footage shows continuous task execution under the tested conditions. No commercial product is claimed. The video is a research illustration of multi-step precise manipulation. Broader generalization is not asserted. The work remains at the laboratory stage.

By humanoid.press · Editor: Humanoid Press Team · November 3, 2026

Long-horizon Manipulation

Multi-step

Execution

Precise

Contact

Research

Stage

Watch Demo →

Humanoid Press — Daily Brief · November 2, 2026

ETH Zurich Researchers Show a Robotic Hand That Walks on Its Fingers

The Soft Robotics Lab turned a commercial five-fingered hand into a self-contained crawler that supports its own weight, recovers from falls, and performs simple manipulation.

By humanoid.press · Editor: Humanoid Press Team · November 2, 2026

ETH Zurich Walking Hand

Video credit: Nicole Davidson / ETH Zürich ▶ Via YouTube

Researchers at ETH Zurich’s Soft Robotics Lab have demonstrated a robotic hand that walks on its own fingers without an attached arm or body. The platform is based on a commercial five-fingered hand from Wuji Technology, fitted with a small onboard backpack that carries a battery, an inertial measurement unit, and a Raspberry Pi Zero. The fingers serve both as legs and as manipulators. The hand crawls across surfaces, steers, recovers from falls, and can press keys or push objects while supporting its own weight. The video shows the hand moving through an indoor setting decorated for Halloween. The work was led by Ph.D. candidate Amirhossein Kazemipour. The team trained separate task-specific policies in simulation using reinforcement learning, then transferred them to the physical hand. All inference runs onboard. The hand has been tested on 14 surfaces, including carpet, grass, gravel, and metal grating. Average crawling speed reached about 9 centimeters per second. Fall recovery succeeded in 21 of 25 trials in one reported set of tests. The demonstration is research-stage. No commercial product or field deployment is claimed.

The design choice was deliberate. Rather than building a custom lightweight hand, the researchers used an existing commercial hand so that manipulation capability would not be sacrificed for locomotion. The onboard computer runs multiple policies and switches among them as needed. Keyboard pressing was demonstrated without visual feedback, while object pushing used an overhead camera. The hand can right itself after falling and continue moving. The video illustrates the crawling motion in a staged indoor environment. Independent measurements of long-duration reliability, energy use, or performance on untested surfaces have not been published. The researchers have noted potential uses in confined spaces where a full arm cannot reach, such as certain inspection or maintenance tasks. Those applications remain prospective. The current system is a laboratory prototype. The Halloween-themed video is a public demonstration of the locomotion capability. Technical details appear in a preprint. Peer-reviewed publication has not yet been announced.

Walking with five unequal fingers requires coordinated control so that the hand remains balanced while individual fingers lift and plant. The team used simulation-based reinforcement learning rather than hand-crafted gait sequences. The resulting policies allow the hand to crawl, turn, and recover. The video shows continuous motion on a flat indoor surface. Outdoor tests on grass and gravel are described in reporting about the work. No data on maximum payload while crawling or on simultaneous high-force grasping during locomotion have been released. The backpack adds mass and power but keeps the system untethered. The researchers have compared the appearance to the disembodied hand from The Addams Family. The functional claim is that a hand can use its fingers for both support and interaction. The demonstration establishes that such locomotion is possible with commercial hardware and learned policies. Scaling, robustness under disturbance, and integration with a larger robot remain open. The video is a research showcase rather than a product announcement.

ETH Zurich released the Halloween-themed video in late October 2026. The Soft Robotics Lab described the hand as having learned to use its fingers as legs through machine learning in simulation. The same fingers can still interact with objects. The video was filmed at a location provided by ROOM Escape Room. The researchers have stated that the larger question is how robots can become more adaptable in the ways they use their bodies. The hand is not presented as a replacement for conventional mobile robots. It is a study in alternative locomotion for an anthropomorphic hand. No timeline for further development or open-sourcing of the policies has been announced. The footage shows the hand crawling, resting, and moving through a decorated space. It does not include endurance runs or multi-hour operation. For researchers working on modular or detachable end-effectors, the result is a concrete example of a hand that can locomote independently. Commercial transfer has not been claimed. The work remains academic.

A robotic hand from ETH Zurich’s Soft Robotics Lab walks on its fingers without a body. The platform uses a commercial five-fingered hand with an onboard computer and battery. It crawls, steers, recovers from falls, and can press keys or push objects. Policies were trained in simulation and run onboard. The video shows the hand in a Halloween setting. Tests covered 14 surfaces. Fall recovery and keyboard interaction have been reported. No commercial product is claimed. The demonstration is research-stage. Potential uses in confined spaces are noted by the researchers. Independent long-duration data have not been published. The video is a public illustration of the locomotion capability.

The ETH Zurich walking hand is a self-contained research platform that uses its fingers for both locomotion and interaction. The video captures the crawling motion in an indoor Halloween setting. The system is based on commercial hardware augmented with onboard power and compute. Learned policies handle crawling, steering, and recovery. The work is experimental. No field deployment or product release has been announced. The researchers have highlighted the adaptability question: how a robot can find new ways to use its existing body. The footage is consistent with the published description of the project. Further technical results appear in a preprint. The demonstration adds a concrete example of independent hand locomotion to ongoing research on modular robotic systems.

Watch Walking Hand →

Also Today

RoCogMan Lab Shows Sensorless In-Hand Ball Rotation Across Sizes

Researchers at the Robotic Cognition & Manipulation Lab (RoCogMan Lab) at Hanyang University ERICA have demonstrated continuous in-hand rotation of balls whose diameters are not provided to the controller. A five-fingered robotic hand rotates balls ranging from 55 mm to 100 mm in diameter while maintaining the grasp. The system uses only joint position feedback. No tactile sensors or force sensors are required for the demonstrated behavior. The hand can reverse the direction of rotation without releasing the object. The video shows the hand performing the rotations on balls of different sizes. The lab is led by Professor Ji-Hun Bae. The work is part of the lab’s ongoing research on robotic grasping and dexterous manipulation. No quantitative success rates, cycle counts, or robustness metrics under disturbance have been published with the video. The demonstration is experimental. The hand adapts its finger motions based on joint feedback alone. Independent verification of the results has not been published. The video is a research demonstration of the sensorless approach.

In-hand manipulation without prior knowledge of object size is a practical challenge because grasp geometry changes with diameter. The RoCogMan demonstration shows the hand adjusting its finger positions and motions to keep rotating balls of four different diameters. The controller does not receive an explicit size measurement. Adaptation occurs through joint position feedback. The video illustrates continuous rotation and a direction reversal while the grasp is maintained. No data on maximum rotation speed, torque limits, or performance with non-spherical objects are included in the public description. The lab has previously shown related in-hand manipulation work, including rotations under external disturbances and tactile-based repositioning. The current video focuses on the sensorless case across a size range. The approach is model-based rather than learned in the description provided with the clip. Commercial transfer or integration into a full humanoid system has not been announced. The demonstration remains at the laboratory stage. The footage is consistent with the stated capabilities of adaptive rotation without size input or tactile sensing.

RoCogMan Lab works on robotic cognition and manipulation, including dual-arm humanoid tasks and gripper design. The ball-rotation video is one example of the group’s in-hand manipulation research. The hand is a five-fingered platform. The diameters shown are 55 mm, 65 mm, 75 mm, and 100 mm. The video does not show the hand operating under significant external force or on a mobile base. No comparison with tactile-sensing methods on the same hardware is provided in the clip. For researchers studying sensor-minimal control, the relevant question is how far joint feedback alone can support stable continuous manipulation. The demonstration provides a concrete example within the tested size range. Broader generalization to arbitrary objects or longer sequences is not claimed. The lab has credited student Seonghyeon Chu for the work shown. The video is a research result rather than a product demonstration. Further quantitative evaluation has not been released with the footage.

The ability to reverse rotation direction while maintaining grasp is shown in the video. The hand continues to hold the ball as the rotation direction changes. No tactile or force sensing is used. The controller relies on joint position information. The video is short and focused on the manipulation itself. It does not include setup, calibration, or failure cases. Independent replication of the sensorless multi-size rotation has not been published. The lab’s public materials describe the work as adaptive in-hand rotation for unknown diameters. The demonstration establishes that the hand can perform the task under the stated conditions. Scaling to higher speeds, heavier objects, or more complex shapes remains future work. The footage is a laboratory illustration of the control approach. No commercial availability is claimed. The result adds to the lab’s body of in-hand manipulation demonstrations.

The RoCogMan Lab at Hanyang University ERICA has shown a five-fingered robotic hand continuously rotating balls of different diameters without being given the size in advance. The system uses only joint position feedback. No tactile or force sensors are required for the demonstrated behavior. The hand can reverse rotation direction while keeping the grasp. The video shows the rotations on balls from 55 mm to 100 mm. The work is experimental. No quantitative reliability figures have been published with the clip. The demonstration is a research result on sensorless adaptive in-hand manipulation. Independent verification has not been released. The video is consistent with the lab’s description of the capability.

By humanoid.press · Editor: Humanoid Press Team · November 2, 2026

RoCogMan In-Hand

55–100 mm

Ball size

Joint only

Feedback

5-finger

Hand

Watch Ball Rotation →

AthenaZero Hands Combine Speed, Strength, and Tactile Sensing

The RAI Institute has detailed the hands of its AthenaZero bimanual robot in a Science Robotics paper and accompanying video. Each hand has three fingers driven by tendons. Motors located in the torso transmit force through flexible cables so the hand itself remains light. The fingers can interdigitate, allowing the hand to hold objects of different sizes such as an egg or a hammer. Closing time is about 80 milliseconds. The design is intended to provide both high force and the ability to handle delicate objects. Barometric pressure sensors in the fingertips and palm provide tactile information. Each palm contains eight sensors and each fingertip contains three. The sensors detect light contact and indicate location. Local computation for the sensors is housed in the back of the hand to reduce latency. The video shows the hands handling objects. AthenaZero is a fixed-base bimanual platform designed for dynamic manipulation. The paper reports throwing speeds above 30 m/s and catching or batting above 14 m/s in controlled tests. The hands are one element of the low-inertia design. No commercial product is claimed. The work is research-stage.

AthenaZero places most of the arm mass closer to the base through quasi-direct-drive actuators and transmission remotization. The hands follow the same principle. Finger motors sit in the torso and drive the fingers through approximately 1.8-meter Bowden cables. The resulting hand is light enough for rapid motion while still delivering usable force. The tactile sensors are embedded in flexible fingertip and palm pads. They operate at 200 Hz according to the paper. The video illustrates contact sensing during object handling. The platform is not a full humanoid; it has no legs and stands on a fixed base. The paper focuses on dynamic tasks such as throwing, catching, and batting. The hands are shown supporting both forceful and delicate interactions. Independent replication of the reported speeds or sensor performance has not been published. The RAI Institute has released the paper and the video as research results. Deployment outside the laboratory is not claimed. The demonstration is consistent with the described hand design.

The combination of speed and delicacy in a single hand design is the stated goal. The tendon-driven fingers close quickly while the remote motors keep endpoint mass low. The tactile sensors add contact information without adding significant mass at the fingertip. The video shows the hands interacting with objects of different properties. No quantitative data on maximum grasp force, sensor resolution under dynamic conditions, or long-term cable wear appear in the public video description. The paper provides additional technical detail on the actuation and sensing. The hands are underactuated three-finger designs optimized in part for objects such as a baseball and a bat handle. The interdigitating fingers allow power grasps on a range of sizes. The video is a hardware spotlight on the tactile sensing and manipulation capability. It does not show the full throwing or catching sequences. Those appear in other materials associated with the paper. The work remains academic research. No product timeline has been announced.

Local compute in the back of the hand processes the barometric sensor data to reduce latency. The sensors indicate both the presence of contact and its approximate location. The video demonstrates the hands during object interaction. The design trades some transmission efficiency for lower endpoint inertia. The paper discusses this engineering choice in the context of dynamic manipulation. AthenaZero has been shown performing robot-to-robot and human-to-robot catching and batting in controlled settings. The hands are part of that system. No data on continuous multi-hour operation or on performance with arbitrary household objects have been released. The RAI Institute published the Science Robotics paper in September 2026. The video was released later as a focused look at the hands. The demonstration establishes the tactile and dynamic capabilities under the tested conditions. Broader claims about general-purpose dexterity are not made. The footage is a research illustration of the hand design.

The AthenaZero hands from the RAI Institute are three-fingered, tendon-driven designs with motors in the torso. They close in about 80 milliseconds and can handle both forceful and delicate objects. Barometric sensors in the palms and fingertips provide tactile feedback. Local compute sits in the back of each hand. The video shows the hands in use. The platform is a fixed-base bimanual research robot. The work was published in Science Robotics. No commercial product is claimed. The hands are part of a low-inertia design intended for dynamic manipulation. The video is a research demonstration of the tactile sensing and hand capability. Independent verification has not been published. The demonstration remains at the laboratory stage.

By humanoid.press · Editor: Humanoid Press Team · November 2, 2026

AthenaZero Tactile

~80 ms

Close time

3-finger

Hand

Tactile

Sensors

Paper Science Robotics →

Humanoid Press — Daily Brief · November 1, 2026

PHYMI Unveils Sagi, a Robotic Centaur With Cargo Box and Free Arms

The wheeled-legged platform carries loads on its back so its humanoid upper body can handle objects without balancing a payload in its hands.

By humanoid.press · Editor: Humanoid Press Team · November 1, 2026

Sagi Centaur

Video credit: PHYMI ▶ Via YouTube

PHYMI has unveiled Sagi, a robotic centaur that combines a humanoid upper body with four wheeled legs and a cargo box mounted on its back. The design keeps the arms free for manipulation while the box carries loads between stops. The video shows the platform moving and handling objects without the arms supporting the primary payload. No specifications for payload capacity, speed, runtime, or production timeline have been published. The platform is presented as a way to separate transport from manipulation. Independent performance data are not available. The video is a company demonstration of the concept. Commercial availability has not been announced. The wheeled-legged configuration is intended to improve mobility on suitable surfaces while retaining the ability to handle items with the upper body. The cargo box is the central practical feature shown. Further details on control, battery life, or intended environments remain unreleased. The demonstration focuses on the basic operating idea rather than on long-duration or complex tasks.

Most humanoid platforms must carry objects in their hands or on their body while walking or balancing. Sagi moves the load to a dedicated box so the arms can remain available for other work. The four wheeled legs are shown supporting stable movement. The video illustrates the robot reaching for and placing items while the box holds cargo. No measurements of energy use, maximum speed on flat ground, or stability on uneven surfaces appear in the release. The company has not stated whether the platform is intended for indoor logistics, outdoor work, or both. The humanoid upper body provides the manipulation capability. The wheeled base is the mobility solution. The combination is the design claim. The video does not show autonomous navigation over long distances or interaction with people. It is a hardware concept demonstration. Reliability under repeated loading and unloading has not been quantified. The platform remains at the unveiling stage.

Separating transport and manipulation can reduce the control problem for each subsystem. The arms do not need to stabilize a heavy object while the base moves. The base does not need to remain perfectly balanced for the arms to work. The video shows this separation in action. No data on cycle times, docking accuracy, or recovery from faults have been released. PHYMI has not published a timeline for pilots or customer trials. The cargo box is visible in the footage as the load-carrying element. The wheeled legs are the mobility element. The upper body is the manipulation element. The video presents all three together. Independent evaluation of the design has not been published. The demonstration is consistent with the company’s description of free arms and a load-carrying box. Broader operational claims are not made. The platform is shown as a working prototype for the concept.

Wheeled-legged robots have appeared in other designs that seek higher speed on prepared surfaces while retaining some ability to handle steps or rough ground. Sagi applies the idea to a platform that also carries a cargo box and a humanoid torso. The video focuses on the resulting freedom for the arms. No comparison with pure biped humanoids or with conventional mobile manipulators is provided. The company has not released joint counts, actuator types, or sensor details. The footage shows basic motion and object handling. It does not include endurance runs or multi-hour operation. For operators considering logistics or material handling, the open questions include how the box is loaded and unloaded, how the system navigates, and how it behaves when the box is full. Those details are not in the current video. The unveiling establishes the form factor. Performance numbers remain future information.

PHYMI’s Sagi places a cargo box on a four-wheeled-leg base so a humanoid upper body can handle objects without carrying the primary load. The video demonstrates the concept of free arms and separate transport. No payload, speed, runtime, or availability figures have been published. The platform is at the unveiling stage. Independent data are not available. The design aims to simplify the combination of mobility and manipulation. The video shows the robot moving and working with items while the box holds cargo. Further testing and specification releases have not been announced. The demonstration is a company presentation of the robotic centaur form. Real-world utilization details remain unreleased.

The Sagi platform is shown with a humanoid torso, four wheeled legs, and a rear cargo box. The arms remain available for handling while the box carries loads between stops. The video illustrates the operating idea. No commercial timeline or performance metrics have been released. The company presents the design as a practical separation of transport and manipulation. The footage is consistent with that description. Broader claims about autonomy, safety, or scale are not made. The platform is a prototype demonstration. Independent verification of stability, payload, or cycle time has not been published. The unveiling adds a wheeled-legged cargo variant to the range of humanoid-derived platforms under development.

Watch Sagi →

Also Today

Insurers Begin Covering Humanoid Robots as They Enter Shared Spaces

Insurers are starting to write policies specifically for humanoid robots as the machines move from laboratories into homes, schools, and public settings. In late October 2026, Adrian Flux and Tokio Marine HCC International announced what they described as the United Kingdom’s first household insurance product for domestic humanoid robots. The cover is available through Adrian Flux’s home insurance and is underwritten by Tokio Marine HCC International. It addresses damage to or theft of the robot itself, sudden damage the robot causes to the home or belongings, and personal liability if the robot injures a person or damages property belonging to someone else. It also includes provision for emergency recovery or safe disablement after a sudden or unforeseen malfunction. The companies said the product is intended to clarify responsibility as the technology develops. Domestic humanoids are expected to reach UK homes from 2027, while early models have already been offered for pre-order in the United States. The video shows a compact educational humanoid interacting with children on a soccer field, illustrating one setting where shared use raises practical questions about liability.

The policy treats the humanoid both as a high-value asset and as a potential source of risk. Coverage for the machine itself protects against loss or damage to an expensive device. Coverage for harm it causes protects the owner from claims that might otherwise fall outside ordinary home policies. The emergency recovery clause addresses the possibility that a malfunction leaves the robot in a state that requires intervention before it can be safely moved or powered down. Adrian Flux has compared the approach to earlier policies developed for driverless vehicles, noting that the terms will be updated as technology and risk experience evolve. No premium amounts or specific underwriting criteria have been published. The product is currently limited to UK customers purchasing home insurance through the broker. Independent data on claims frequency or severity for humanoid robots do not yet exist because few units are in private hands. The announcement is a market response to anticipated deployment rather than a response to a large existing book of business.

Similar discussions are under way in other markets. In Japan, Tokio Marine & Nichido Fire Insurance has said it will introduce AI robot insurance in November 2026 for wheeled, four-legged, and humanoid platforms. That product is described as covering repairs or replacement after collision or natural disaster, third-party injury and property damage, and the cost of substitute workers or rental robots while a unit is out of service. In China, China Pacific Insurance has offered humanoid coverage sold by the day, week, or month to match short-term rental use. These products differ in structure from the UK household policy but share the recognition that standard property and liability forms do not automatically address robots that move, interact with people, and can cause physical harm. The video of a humanoid playing soccer with children is one example of interaction in an uncontrolled setting. Insurance products are one mechanism for allocating the resulting risk between owners, manufacturers, and software providers. Allocation questions remain open in most jurisdictions.

Liability for an autonomous or semi-autonomous robot is not settled by existing product-liability or premises-liability rules in most places. An owner may be responsible for supervision. A manufacturer may be responsible for design or software defects. A software provider may share responsibility if the model was updated after sale. The new UK policy covers the policyholder’s personal liability, which is the conventional approach in household insurance. It does not resolve questions about manufacturer or software liability. The video shows a small humanoid in close proximity to children. Educational and recreational platforms are among the earliest units expected to leave controlled environments. Insurers are writing coverage before large numbers of claims have occurred. The companies involved have stated that the terms will change as experience accumulates. No claim has been made that the policies eliminate risk or that humanoids are ready for unsupervised use around children or in public. The products are early commercial responses to an emerging technology.

Insurers in the United Kingdom, Japan, and China have begun offering specialized coverage for humanoid robots. The UK product from Adrian Flux and Tokio Marine HCC International covers damage to the robot, damage it causes, personal liability, and emergency recovery. Other markets are introducing industrial or rental-oriented forms. The video illustrates a humanoid interacting with children, one of the settings that raises liability questions as the machines leave laboratories. No claims experience yet exists at scale. The policies are being written ahead of widespread household or public deployment. Responsibility among owners, manufacturers, and software providers remains a developing area. The announcements mark the start of a commercial insurance market for humanoid robots rather than a mature one.

By humanoid.press · Editor: Humanoid Press Team · November 1, 2026

Insurance Liability

UK

Household

2027

Expected homes

Liability

Covered

Watch Master Mini →

110 Humanoid Robots Set Guinness Record for Simultaneous Dance in Jinan

A performance in Jinan, China, featuring 110 humanoid robots dancing simultaneously for more than five minutes has been recognized as a Guinness World Record for the most humanoid robots dancing at once. The video shows the group executing a coordinated routine without real-time human control during the performance. No details on the specific robot models, the control system used for the choreography, or the preparation time have been published in connection with the record. The event is presented as a demonstration of synchronized motion at scale. Independent verification beyond the Guinness recognition is not detailed in the available materials. The footage captures the group dance. Technical specifications for the robots or the coordination method are not included. The record establishes a quantitative milestone for simultaneous humanoid performance. Broader claims about autonomy or practical utility are not made. The video is a record of the event.

Large-scale synchronized robot performances have been used to showcase coordination and reliability. The Jinan event used 110 humanoid platforms for a continuous dance lasting over five minutes. The video shows the robots moving together without visible real-time teleoperation. No information on failure rates during rehearsal or during the recorded performance has been released. The company or organizers behind the specific robots have not been identified in the public description of the record. The Guinness recognition confirms the count and the simultaneous nature of the activity. It does not address the underlying software, power systems, or recovery mechanisms. The footage is consistent with a pre-programmed or centrally coordinated routine. Individual robot capability outside the choreographed sequence is not shown. The event is a scale demonstration. Practical applications beyond performance are not claimed.

Guinness World Records for robot activities typically require documented counts, timing, and verification of the conditions. The Jinan dance meets the published criteria for the simultaneous humanoid category as recognized. The video shows the robots in formation and executing movements for the stated duration. No breakdown of how many robots completed the full routine without interruption is provided beyond the record itself. The absence of real-time human control during the performance is stated in the description. The control architecture that enabled the coordination is not described. For researchers studying multi-robot systems, the relevant open questions include communication method, clock synchronization, and behavior when individual units fall out of sync. Those details are not in the current video. The record is a verified count of simultaneous dancing humanoids. The technical implementation remains undescribed in the public materials.

Humanoid platforms have been used in group performances before, but the Jinan event sets a new documented high for the number dancing at the same time. The video captures the scale of the formation. No comparison with prior records or with non-humanoid robot groups is included. The performance lasted more than five minutes. The footage does not show setup, teardown, or the robots performing other tasks before or after the dance. Organizers have not released information on power management, thermal limits, or the software stack used. The demonstration establishes that 110 humanoid robots can execute a shared routine for the recorded duration under the conditions verified by Guinness. Independent engineering analysis of the system has not been published. The video is a record of the achievement. Further technical disclosure has not been announced.

110 humanoid robots danced simultaneously for more than five minutes in Jinan, setting a Guinness World Record for the category. The video shows the coordinated performance without real-time human control. No technical details on the robots, control system, or preparation have been released. The event is a verified scale demonstration of synchronized motion. Independent analysis beyond the record recognition is not available. The footage captures the group routine. Broader claims about autonomy or application are not made. The record confirms the count and duration under the stated conditions. The demonstration remains a performance milestone.

By humanoid.press · Editor: Humanoid Press Team · November 1, 2026

Guinness Dance

110

Robots

5+ min

Duration

Record

Guinness

Watch Dance →

Humanoid Press — Daily Brief · October 31, 2026

Figure AI Says Figure 04 Is in Design Lock With Parts Shipping

CEO Brett Adcock describes the next generation as the company’s largest hardware leap, with in-house actuators, a new battery, and redesigned hands now in testing.

By humanoid.press · Editor: Humanoid Press Team · October 31, 2026

Figure 04 Hardware

Video credit: Figure AI ▶ Via YouTube

Figure AI has reached full design lock on its fourth-generation humanoid, Figure 04. In a public all-hands recording released in late October 2026, CEO Brett Adcock said the robot’s specifications are frozen, parts have begun shipping, and component testing is underway. He called the step from Figure 03 to Figure 04 the largest generational leap in the company’s history. The video shows Adcock describing the status of actuators, batteries, and hands. Design lock means engineering drawings are finalized for production preparation. It is not a product launch. Figure has not released weight, payload, runtime, cost, or a public reveal date. Adcock has previously said the company built its 1,000th Figure 03 over the summer and reduced costs by more than 90 percent between Figure 02 and Figure 03. The video is a company status update on the next hardware generation. No assembled Figure 04 robot is shown walking or performing tasks.

The redesign focuses on three major subsystems. Figure is bringing up a new set of in-house actuators in its test lab, with the goal of fully characterizing every unique actuator. Adcock said the team had brought up ten complete battery packs the week before the all-hands and that load and thermal testing would follow. The hands are described as the most heavily reworked component. Adcock called a fourth iteration of the Figure 04 hand an engineering masterpiece, noting that it contains more actuators than the rest of the robot’s body combined. The team has spent months on millimeter-level fingertip tracking against human references. The video includes discussion of these hardware elements. No independent verification of actuator performance, battery capacity, or hand dexterity has been published. The claims are internal company statements. Figure has framed Figure 04 as the platform intended to move beyond demonstrations, though Adcock has not claimed the robot is ready for unsupervised home use.

Figure’s development pattern has been roughly annual generations. Figure 02 was shown in 2024 and Figure 03 in October 2025. Matching that pace would place a public Figure 04 appearance in late 2026 or early 2027, but the company has not confirmed a timeline. The all-hands also covered data collection and compute plans, including large-scale human video work and further GPU capacity. Those efforts run alongside the hardware program. The video focuses on the hardware status rather than on AI model results. No customer deliveries, pilot sites, or production volume figures have been announced for Figure 04. The design lock milestone allows parts to ship and testing to proceed. Assembly of complete robots and long-duration reliability testing remain future steps. The video is consistent with Adcock’s public statements from the all-hands. It shows the engineering discussion rather than a finished robot.

Adcock has described Figure 04 as potentially the iPhone 1 of humanoids, indicating a major capability step. The company has not published specifications that would allow external comparison of strength, speed, energy use, or durability. The video does not include side-by-side testing against Figure 03 or other platforms. Independent observers have noted that design lock is an important internal gate but does not guarantee the finished robot will meet external expectations for home or industrial use. Figure continues to invest in vertical integration, including in-house actuators. The battery work is intended to support longer operation. The hand redesign aims at higher dexterity. None of these subsystems have been shown operating together on a complete Figure 04. The video remains a status briefing. Commercial availability and performance data are not yet available.

Figure AI says Figure 04 is in full design lock with parts shipping. The company describes it as the largest generational leap, with new in-house actuators, a redesigned battery, and advanced hands under test. The video presents the hardware update from the all-hands. No specifications, price, or reveal date have been released. Design lock is an engineering milestone, not a product launch. Figure previously reported building 1,000 Figure 03 units and large cost reductions from the prior generation. The video is a company overview of the next hardware stage. Assembled robots performing tasks are not shown. Further testing of actuators, batteries, and hands is underway. Public demonstrations remain future work.

The Figure 04 program has moved from design into component testing and parts shipping. Adcock has stated that the robot represents the biggest hardware step the company has taken between generations. The video covers the status of the actuators, battery packs, and hands. No independent measurements of performance have been published. The company has not set a date for a public reveal or customer availability. The all-hands also addressed data and compute scaling. Those programs support the broader goal of more capable robots. The video focuses on the hardware milestone. Real-world capability will depend on how the integrated system performs once complete robots are assembled and tested. Those results have not been shown.

Watch Figure 04 →

Also Today

Generalist’s GEN-1.5 Shows One-Shot Physical Learning

Generalist AI says its GEN-1.5 model can learn a new physical task in seconds from a single demonstration. The company places a 3- to 12-second sensorimotor example—recorded either by a human with handheld grippers or by the robot itself—into the model’s 30-second context window. The robot then attempts the task immediately, without gradient updates or fine-tuning. Generalist calls the technique physical prompting and says the capability emerged from more than eight months of continuous pretraining rather than from any architecture built specifically for in-context learning. Across ten short-horizon manipulation tasks that include opening jars, unzipping pouches, and retrieving objects, the company reports 59 percent average success (±10 percentage points) with pure one-shot prompting. Ten gradient steps on roughly five minutes of additional data per task raise that figure to 83 percent (±9). The video shows a robot performing a cleaning or pot-related task consistent with the demonstration style. All numbers are company-reported. No independent verification has been published. The tasks are simple and short-horizon.

GEN-1.5 is a large multimodal model that processes video, other sensor data, language, and proprioceptive inputs and outputs action trajectories at 100 Hz. It holds roughly 30 seconds of context. A demonstration is inserted into that window as a physical prompt. Once the prompt is present, the model generates actions for the robot without any weight updates. The company reports limited compositional behavior: two separate demonstrations placed in context can be chained, with the model producing intermediate motions that appear in neither prompt. Zero-shot transfer from simulation prompts and some human-to-robot imitation are also described. The video illustrates the robot executing a physical skill after a short demonstration. Skills learned only in-context remain more brittle than fine-tuned policies. The company presents the result as evidence that one-shot and few-shot learning of physical skills can emerge at scale, not as a claim that the system is ready for production use. The announcement continues Generalist’s line of foundation models trained from physical interaction data.

Earlier Generalist models focused on scaling laws and high-success-rate post-training. GEN-1.5 is positioned as the first model the company knows that exhibits broad one-shot and few-shot physical skill learning without task-specific engineering. The video shows the robot handling objects in a manner consistent with the reported one-shot capability. No long-horizon tasks or multi-step industrial sequences are demonstrated. Independent researchers have noted that similar in-context learning has appeared in narrower settings before, but Generalist claims broader coverage across closed-loop dexterous skills. All evaluation data come from the company. The 59 percent and 83 percent figures have not been replicated by outside groups. The model is not open source. The video is a demonstration of the physical prompting approach. Deployment readiness is not claimed. The result is presented as a step toward more general physical intelligence.

The one-shot approach relies on placing a short demonstration into the context window so the model can infer the goal and generate actions. The video shows this process in action on a practical object-handling task. Success rates vary across tasks, with the company reporting a standard deviation of about 10 percentage points on the one-shot average. Few-shot fine-tuning improves performance but still leaves a gap to full reliability. The company notes that the capability was not explicitly trained for and emerged during extended pretraining. No architectural changes for meta-learning or in-context adaptation were added. The video is consistent with the written description of physical prompting. It does not show failure cases or long-duration operation. For developers, the practical questions include how well the approach scales to longer tasks and how brittle the in-context skills remain under variation. Those details are not fully quantified in the current release.

Generalist AI’s GEN-1.5 can attempt new short-horizon tasks after a single 3- to 12-second demonstration with no gradient updates. The company reports 59 percent average one-shot success across ten tasks and 83 percent after limited fine-tuning. The video shows the robot performing a physical task in the one-shot style. All results are company-reported. The tasks are simple. Skills learned only in context are more brittle than fine-tuned policies. The model outputs 100 Hz trajectories from a 30-second context window. The announcement frames the result as emergent one-shot and few-shot physical learning at scale. No independent verification has been published. The video is a demonstration of the capability. Production use is not claimed.

By humanoid.press · Editor: Humanoid Press Team · October 31, 2026

GEN-1.5 One-Shot

59%

One-shot

3–12 s

Demo

0

Gradient steps

Read GEN-1.5 →

HIER Lab Explores VLA and Whole-Body Control for Dual-Arm Tasks

The Hybrid Intelligent Experimental Robotics (HIER) Lab at North Carolina State University is working on dual-arm manipulation that pairs vision-language-action models with whole-body control. The lab’s stated goal is to give robots both high-level task autonomy and the low-level stability and precision needed for longer sequences of real-world work. Researchers are examining how the two layers can be coordinated so that a robot can plan and execute multi-step manipulation without the high-level policy fighting the low-level controller. Vision-language-action models supply the semantic understanding and task decomposition. Whole-body control supplies the continuous coordination of arms, base, and balance. The video shows coordinated dual-arm motions under the combined control approach. The work is still experimental. No commercial product or large-scale deployment is claimed. The lab is affiliated with the Department of Mechanical and Aerospace Engineering at NC State.

HIER Lab’s broader program includes legged humanoid and animaloid platforms, loco-manipulation, and robot intelligence for environments shared with people. The dual-arm VLA-plus-WBC work sits inside that larger effort. The video illustrates the kind of integration the group is studying rather than a finished capability ready for independent operation. Academic labs are exploring many variants of hierarchical VLA and whole-body control. HIER Lab’s contribution is the specific combination of the two on dual-arm hardware and the focus on long-horizon tasks that demand both levels of control at once. Results remain at the research stage. No success rates, task lengths, or reliability figures have been published for the demonstrated system. The video is a research demonstration of the control integration. Independent replication or commercial transfer has not been announced. The lab continues to develop platforms for agile loco-manipulation in complex environments.

Combining high-level vision-language-action reasoning with low-level whole-body control is a common research direction. The challenge is keeping the high-level plan stable while the low-level controller handles contact, balance, and precise motion. The video shows dual-arm coordination consistent with this approach. No long-horizon multi-step sequences with error recovery are quantified in the public materials. The lab’s public site describes work on intelligent robots capable of agile loco-manipulation with full autonomy. The dual-arm effort is part of that program. The video does not show mobile base motion or human interaction. It focuses on the manipulation hardware and control. For researchers, the relevant questions include how the hierarchy is implemented, what latency exists between layers, and how the system handles unexpected contact. Those details are not fully described in the current video. The work remains academic research.

HIER Lab has acquired manipulators and continues to expand its facilities. The dual-arm demonstration is one example of the group’s interest in harmonizing high-level autonomy and low-level control. The video shows the robots performing coordinated actions. No claims of human-level dexterity or reliability are made. The approach is presented as an exploration of how the two control layers can work together for longer tasks. Broader results on safety, efficiency, and real-world reliability have not been published. The video is consistent with the lab’s stated research questions. It does not represent a completed system ready for deployment. Other groups are pursuing similar hierarchical methods on different hardware. HIER Lab’s contribution is the specific dual-arm implementation and the focus on long-horizon manipulation. The demonstration remains at the research stage.

Researchers at the HIER Lab at NC State are testing dual-arm manipulation that combines vision-language-action models for high-level decisions with whole-body control for low-level precision and stability. The goal is longer-horizon tasks that require both layers. The video shows coordinated dual-arm motions under the combined approach. The work is experimental. No commercial product or large-scale results are claimed. The lab works on loco-manipulation and robot intelligence for shared environments. The video is a research demonstration of the control integration. Independent verification has not been published. Results remain at the academic stage. The demonstration illustrates the research direction rather than a finished capability.

By humanoid.press · Editor: Humanoid Press Team · October 31, 2026

HIER Lab Dual-Arm

VLA

High-level

WBC

Low-level

Dual-arm

Platform

Lab HIER Lab →

Humanoid Press — Daily Brief · October 30, 2026

Agility Robotics Targets More Than 20 Hours of Work a Day for Digit 5

Nyobolt’s ultra-fast charging battery is intended to give Digit 5 a 10:1 run-to-charge ratio, with early access planned for the first half of 2027.

By humanoid.press · Editor: Humanoid Press Team · October 30, 2026

Digit 5 Nyobolt

Video credit: Agility Robotics / Nyobolt ▶ Via YouTube

Agility Robotics has selected Nyobolt as the battery technology partner for its Digit 5 humanoid. The companies say the system is designed to deliver 90 minutes of runtime and recharge in 9 minutes, producing a 10:1 run-to-charge ratio. That is an improvement over the 2:1 ratio reported for Digit 4. Agility states that the higher ratio enables more than 20 hours of productive work in a 24-hour day, enough to cover two full shifts. Early access units are planned for the first half of 2027. The video presents the partnership and the claimed duty cycle. No independent field measurements of the 90-minute runtime or the 9-minute charge have been published. The figures are company targets based on internal testing of pre-production hardware. Digit 5 is described as Agility’s first humanoid designed for cooperatively safe work at scale. The battery upgrade is presented as a central part of the commercial case. No price or production volume has been announced.

For warehouse and logistics operators, robot uptime is a primary economic variable. A 2:1 ratio means a robot spends a substantial fraction of the day charging. A 10:1 ratio reduces that fraction if the robot can return to a dock on schedule. Agility’s Chief Business Officer Daniel Diez has said that Digit’s uptime is the business case for customers. The video illustrates the intended cycle of work and brief charging stops. It does not show a multi-hour continuous operation or a full fleet schedule. Independent calculations based on the stated 90-minute run and 9-minute charge produce an idealized ceiling of roughly 21.8 hours of work in 24 hours, before accounting for travel to the charger, queueing, maintenance, or faults. Those real-world factors are not quantified in the announcement. The partnership was announced on October 15, 2026. Nyobolt is a Cambridge, UK company focused on high-power, ultra-fast charging. The video is a company explanation of the battery upgrade and its intended effect on daily runtime.

Nyobolt’s technology was selected because Digit 5 requires more power from a smaller, lighter pack that can withstand the vibration and impact of continuous warehouse and manufacturing work. The company has previously raised funding and is valued in the billion-dollar range. Agility has not published cycle-life figures for the Nyobolt packs under the intended duty cycle. The video focuses on the runtime and charge-time claims rather than on battery chemistry or long-term durability data. Early access is targeted for the first half of 2027, with broader availability expected to follow. No customer sites or pilot results are shown. The 10:1 ratio is a company claim, not a measured fleet result. For operators evaluating humanoid deployment, the relevant open questions are the actual energy consumption under typical tasks, the reliability of the dock interface, and how the system behaves when multiple robots need to charge. Those details are not in the current release. The video remains a promotional overview of the intended operating window.

Digit 5 is positioned as an evolution of Agility’s existing humanoid platform, with improved safety features for work near people. The battery system is one of the upgrades intended to support longer continuous operation. The video shows the robot and references the Nyobolt partnership. It does not include side-by-side comparisons with Digit 4 or with other humanoid platforms that use battery swapping. Other manufacturers have pursued hot-swappable batteries as an alternative to ultra-fast charging. Agility’s approach centers on shortening the charge stop. No independent verification of the 9-minute full charge under load has been published. The “more than 20 hours” figure assumes the robot can repeatedly complete the 90-minute work / 9-minute charge cycle without interruption. The announcement does not address dock availability, power infrastructure, or recovery from faults. The video is consistent with the written joint release. It illustrates the claimed duty cycle. Commercial deployment results remain future work.

Agility Robotics has named Nyobolt as the battery partner for Digit 5. The companies claim a 90-minute runtime and a 9-minute recharge, for a 10:1 run-to-charge ratio that Agility says enables more than 20 hours of work in a 24-hour day. Early access is planned for the first half of 2027. The video presents the partnership and the intended operating window. No independent field data have been published. The figures are company targets. The upgrade is positioned as a way to improve fleet economics for warehouse and manufacturing customers. Broader availability timelines and pricing have not been announced. The video is a company overview of the battery system. Real-world utilization will depend on dock access, task schedules, and reliability. Those factors are not quantified in the current materials.

The Digit 5 battery partnership with Nyobolt is intended to raise the run-to-charge ratio from 2:1 on Digit 4 to 10:1. Agility states that this supports more than two full shifts of productive work per robot per day. The video explains the 90-minute runtime and 9-minute charge targets. No measured customer results are available. Early access units are expected in the first half of 2027. The announcement does not include energy-consumption data under representative warehouse tasks or details on the charging infrastructure required. The video remains a promotional presentation of the claimed duty cycle. Independent verification of the runtime and charge time has not been published. The partnership addresses one of the practical constraints on humanoid fleet utilization. Deployment economics will depend on how closely real operations match the idealized cycle.

Read Agility →

Also Today

Pollen Robotics Adds Hoverboard, Mobile App, and NFC Hats for Reachy Mini

Pollen Robotics has announced new accessories and software for Reachy Mini, the open-source desktop robot developed with Hugging Face. The Hoverboard is a two-wheeled base that turns the Wireless version into a mobile platform. Users place the robot on the base, lock it, and control both the wheels and the head from the Reachy Mini mobile app, which provides dual joysticks and a live camera feed. The app is available on Android and iOS and handles setup, Wi-Fi connection, and control without a computer for the Wireless model. NFC hats—chef, pirate, and crown—contain tags that switch the robot’s personality or unlock different interactions when snapped on. Users can also create their own NFC-triggered behaviors. The video demonstrates the Hoverboard in motion, the mobile app interface, and the hat system. First Hoverboard orders are expected to ship at the end of January 2027. The Hoverboard is priced at approximately €135 and is compatible only with the Wireless version. Reachy Mini itself is sold separately. The video is a product announcement of the new mobility and customization options.

Reachy Mini is a small open-source companion robot with a 6-DoF head, camera, microphones, and speaker. The Wireless version includes onboard compute and can be controlled from a phone. The new Hoverboard adds two motorized wheels with soft tires, a built-in battery, and anti-tip supports so the robot can sit when stopped. Control is through the mobile app, which also supports voice interaction, community apps, and remote viewing. The NFC accessory pack includes three themed hats, an NFC module that installs in the head, tags, and magnets. Snapping on a hat changes the robot’s behavior according to the tag. Users can attach their own tags to custom objects. The video shows the robot driving on the Hoverboard and responding to different hats. No independent battery-life or speed figures for the Hoverboard have been published. The system is intended for exploration and play rather than industrial tasks. The app and the robot’s software stack remain open source. The video is a promotional overview of the new accessories and the phone-based control experience.

The Reachy Mini ecosystem already includes a growing catalog of community apps hosted on Hugging Face. The mobile app lets users browse and launch those apps, talk to the robot with AI backends, and control the head manually. Adding the Hoverboard extends the same control model to locomotion. The video illustrates both the driving interface and the NFC personality switching. No data on maximum speed, turning radius, or runtime on the Hoverboard appear in the release. The base is described as arriving fully assembled. Shipping is targeted for late January 2027. The NFC hats are available for both Wireless and Lite versions. The video does not show long-duration autonomous navigation or obstacle avoidance. It focuses on direct user control and the accessory ecosystem. For owners of existing Reachy Mini units, the Hoverboard and hats are optional upgrades. The announcement expands the ways the robot can be used without requiring additional coding. The footage is consistent with the product pages on the Pollen Robotics store.

Pollen Robotics and Hugging Face have positioned Reachy Mini as an accessible platform for physical AI experimentation. The new Hoverboard and mobile app lower the barrier to moving the robot around a room and interacting with it from a phone. NFC hats add a simple physical interface for switching behaviors. The video shows all three features in use. No pricing beyond the Hoverboard and NFC pack has been highlighted in the release, and the robot itself is sold separately. Community developers can still build custom apps that respond to NFC tags or to the mobile platform. The open-source repositories remain available on GitHub. The video is a product launch for the accessories rather than a research demonstration. Independent reviews of the Hoverboard’s stability or the app’s reliability have not been published. The features are aimed at owners who want more mobility and customization. Shipping timelines are company estimates. The announcement continues the pattern of iterative accessory releases for the Reachy Mini platform.

Pollen Robotics has introduced a Hoverboard base, a mobile app, and NFC hats for Reachy Mini. The Hoverboard adds two-wheeled mobility controlled from the phone app, while NFC hats switch personalities via tags. The video demonstrates the new features. Hoverboard orders are expected to ship at the end of January 2027 and are compatible only with the Wireless version. No independent performance data have been published. The accessories expand the ways users can control and customize the open-source robot. The mobile app handles setup and control without a computer for the Wireless model. The video is a product announcement. Broader autonomous navigation capabilities are not shown. The features are optional upgrades for existing and new Reachy Mini owners. The open-source software stack remains available for further development.

By humanoid.press · Editor: Humanoid Press Team · October 30, 2026

Reachy Mini Accessories

Hoverboard

Base

NFC

Hats

App

Control

Store Hoverboard →

AthenaZero’s Tendon-Driven Hands Close in About 80 Milliseconds

Researchers at the RAI Institute have detailed the hands of AthenaZero, a low-inertia bimanual robot designed for dynamic manipulation. Each hand has three underactuated fingers driven by tendons. Motors are located in the torso and transmit force through flexible cables approximately 1.8 meters long. The fingers can interdigitate to grasp objects of different sizes, from an egg to a hammer. Closing time is about 80 milliseconds. Barometric pressure sensors in the fingertips and palm provide tactile feedback. Eight sensors are in each palm and three in each fingertip, with local compute in the back of the hand to reduce latency. The design prioritizes low end-effector mass while retaining speed and force. The video shows the hands closing and contacting objects. AthenaZero was the cover article in the September 2026 issue of Science Robotics. The paper reports throwing speeds above 30 m/s and catching and batting above 14 m/s. No independent commercial availability or pricing has been announced. The robot is a research platform on a fixed base rather than a full humanoid.

Conventional robot hands often place actuators in the fingers or wrist, adding mass that reduces acceleration. AthenaZero remotes the hand motors to the torso and uses Bowden cable transmissions. The underactuated three-finger design allows the fingers to wrap around objects of varying diameter. The barometric sensors detect contact location and force at approximately 0.1 N resolution and 200 Hz. Local computation in the hand reduces the latency of the tactile loop. The video illustrates the combination of speed and gentleness. The Science Robotics paper focuses on the overall low-inertia arm design and demonstrates baseball-inspired tasks: throwing, catching, and batting. Robot-to-robot and human-to-robot catch games are described. The hands are a component of that system. No long-duration reliability data for the cable transmissions or sensor calibration drift have been published. The design is optimized for dynamic interaction rather than continuous static payload. The video is a hardware demonstration of the hand mechanism. Quantitative force and speed figures come from the published paper.

AthenaZero stands on a fixed base with a twisting torso and two 7-DoF arms. The low effective end-point mass—comparable to a human arm and roughly an order of magnitude below many conventional manipulators—is achieved through quasi-direct-drive actuators and transmission remotization. The hands contribute to that goal by keeping finger actuators out of the moving mass. The paper reports that the robot can throw, catch, and bat on human-comparable timescales. The video focuses on the hand hardware and tactile sensing. It does not show the full throwing or catching sequences. For researchers studying contact-rich manipulation, the relevant details are the cable friction losses, the peak fingertip force, and the sensor resolution under dynamic impact. Those measurements are in the Science Robotics paper. The platform is a laboratory system. No commercial product based on AthenaZero has been announced. The video remains a visual explanation of the tendon-driven, sensorized hand design.

Tactile sensing in robot hands is often limited by latency or by the mass of the sensing hardware. AthenaZero places barometric sensors in flexible fingertip and palm materials and processes them locally. The eight palm sensors and three per fingertip give contact location at roughly 3 mm resolution. The 80-millisecond closing time is a company and paper figure for the unloaded hand. Loaded closing times and maximum grasp force under the Bowden transmission are reported in the paper. The video shows the fingers interdigitating around different objects and the sensors responding to light contact. It is consistent with the published description. Independent replication of the 80-millisecond figure has not been published outside the RAI Institute work. The design trades some transmission efficiency for reduced end-effector inertia. The result is a hand that can move quickly and still sense contact. The video is a research demonstration rather than a product launch. Broader deployment of the same hand design has not been announced.

AthenaZero’s three-fingered, tendon-driven hands close in about 80 milliseconds and can grasp both delicate and forceful objects. Motors are remoted to the torso via cables. Barometric sensors in the fingertips and palm provide contact feedback with local compute. The design is part of a low-inertia bimanual platform published in Science Robotics. The video shows the hand mechanism and tactile response. No commercial availability has been announced. Throwing, catching, and batting results are reported in the paper. The hands prioritize speed and low mass. Independent long-term reliability data for the transmissions have not been published. The platform is a research system on a fixed base. The video is a hardware overview of the sensorized hands. Quantitative performance figures come from the published study. Broader adoption of the design remains unreleased.

By humanoid.press · Editor: Humanoid Press Team · October 30, 2026

AthenaZero Tactile

80 ms

Close

3

Fingers

Tactile

Sensors

Paper Science Robotics →

Humanoid Press — Daily Brief · October 29, 2026

Reward AI’s OM-1 Learns Manipulation Directly from Human Demonstrations

The Omnibody Model 1 policy is trained only on data from a wearable hand, with no teleoperation or on-robot experience, and is shown running across arms and humanoids.

By humanoid.press · Editor: Humanoid Press Team · October 29, 2026

OM-1 Foundation Model

Video credit: Reward AI ▶ Via YouTube

Reward AI has released OM-1, a general-purpose robot manipulation policy that learns exclusively from human demonstrations collected through a wearable device. The company describes the system under the principle “One Model, One Data Interface, Any Body.” People wear an Omnibody Hand—a seven-degree-of-freedom mocap device—and perform tasks at normal speed. The One Data Interface records images, tactile signals, inter-finger proximity, hand pose, and force. OM-1 then maps those streams to robot actions. No teleoperation data and no on-robot experience are used in training. The video shows the resulting policy performing package picking on stairs, mopping, sock picking, fridge opening, and other contact-rich tasks on different robot bodies, including industrial arms and humanoids. Reward AI states that a new task can be learned from less than 30 minutes of human demonstration. No weights, code, or public API have been released. The video is a company demonstration of the claimed zero-shot transfer across embodiments.

Most robot foundation models still rely on teleoperation or on-robot data collection. Reward AI argues that human-level manipulation requires redesigning the data, learning, and control pipeline together rather than simply scaling existing robot data. The Omnibody Hand captures demonstrations at human speed. Electromagnetic tracking is reported to reduce mean overshoot error compared with visual-inertial methods. A separate reinforcement-learning control layer, trained in simulation, turns the policy’s intended motions into joint commands for whatever body is being used. The video presents several tasks at real-time speed and shows the same policy running on different morphologies. Independent success rates, cycle times, or failure modes have not been published. The demonstrations illustrate the claimed transfer from human motion to robot execution. No third-party evaluation of the “less than 30 minutes” figure or of cross-body generalization has been released. The system remains an in-house model. The video is consistent with the company’s blog description of OM-1.

Reward AI’s prior work includes DexCap and related wearable capture systems. OM-1 builds on that approach by treating the captured human data as the sole training source for a general policy. The company states that the policy supports contact-rich tasks and improves as the volume and diversity of human data increase. The video includes examples of last-mile delivery style picking, stair navigation with objects, mopping, and fridge interaction. These sequences are presented as evidence that one model can drive multiple bodies. No quantitative comparison against teleoperated baselines or against other foundation models appears in the release. The absence of released weights means external groups cannot yet reproduce or extend the results. For developers evaluating the approach, the relevant open questions are the exact sensor rates, the latency of the control layer, and the conditions under which the policy fails. Those details are summarized in the company’s technical blog. The video serves as the primary public demonstration of the claimed capabilities.

The Omnibody stack separates capture, policy learning, and high-frequency execution. Because every demonstration arrives through the same data interface, Reward AI says there is no need to re-collect data when the target robot changes. The video shows industrial arms and humanoid platforms executing tasks that were demonstrated by people wearing the glove. The sequences are labeled as real-time. No long-duration reliability data or multi-site deployment results have been published. The model is described as general-purpose within the contact-rich manipulation domain shown. It is not presented as a fully autonomous system capable of open-ended reasoning. The video is a curated set of successful runs. Failure cases and recovery behaviors are not detailed in the public materials. The approach differs from methods that fine-tune large vision-language-action models on robot data. Independent verification of the cross-embodiment claim is limited to the released footage. The system is not yet available for external use.

Reward AI’s OM-1 is a robot policy trained only on human demonstration data from a wearable hand. The company states that no teleoperation or on-robot data are used, and that the same policy runs on industrial arms and humanoids. The video shows package handling, mopping, sock picking, and fridge opening at claimed human speed. A new task is said to be learnable from under 30 minutes of demonstration. No weights or public evaluation results have been released. The video is a company demonstration of the Omnibody approach. Independent success rates and generalization benchmarks remain unpublished. The model is positioned as an alternative to robot-data-heavy foundation models. Broader external testing has not been announced. The footage illustrates the claimed transfer from human motion to robot execution across different bodies.

OM-1 learns manipulation directly from people wearing a sensorized glove rather than from teleoperated robot trajectories. Reward AI reports that the policy generalizes across tabletop arms, industrial arms, and humanoids without robot-specific fine-tuning. The video presents several contact-rich tasks performed by different platforms. No quantitative throughput or reliability figures have been published beyond the company’s statements. The “less than 30 minutes” learning claim is company-reported. The absence of released code or weights limits independent verification. The video remains the main public evidence of the claimed capabilities. The system is an in-house research model. Deployment details and commercial availability have not been announced. The approach emphasizes capturing human physical intelligence at natural speed and deploying the resulting policy on multiple robot bodies.

Technical details OM-1 →

Also Today

Boston Dynamics Shows Atlas Built in About a Day, Targets One Hour

Boston Dynamics has released video from its Waltham, Massachusetts prototype shop showing how Atlas humanoids are built. The company states that a complete robot can now be assembled in roughly one day, down from multiple days a year earlier. Staff describe an ambition to reduce that time to about one hour by 2027 and eventually to minutes. Machining, assembly, testing, and repair all occur under one roof. Precision work including gear cutting is performed in-house so engineers and machinists can revise parts the same day or the next. Newly machined components move directly into assembly. Robotic arms assist with some sub-assembly tasks. The video shows the shop floor, in-house machine tools, and collaboration between design and manufacturing. No independent confirmation of daily output or commercial delivery rates has been published. The one-day figure is a company statement about prototype operations. The target of one hour remains an ambition. The footage is a behind-the-scenes look at the current manufacturing process for Atlas.

Keeping machining and assembly in the same building shortens the loop between design changes and physical testing. An engineer can discuss a tolerance or a geometry change with a machinist and receive a revised part quickly enough to install and test it the same week. Boston Dynamics staff note that some components have tolerances tight enough that outside suppliers decline the work. In-house capability therefore supports both speed and the ability to produce specialized parts. The video includes views of milling machines, assembly stations, and Atlas units on the floor. Sr. Technical Program Manager Katie Conner narrates the progression from multi-day builds to the current roughly one-day cycle. No data on yield, scrap rates, or the number of robots built per week appear in the release. The process is described as applying manufacturing discipline to research hardware. Commercial production volumes and customer shipment schedules are not stated. The video focuses on the prototype operations team rather than a high-volume factory line.

Atlas development has historically involved long cycles between hardware revisions. The current shop-floor approach aims to compress those cycles by co-locating design, machining, and assembly. Parts that previously waited on external vendors can now be iterated in days. The video shows robotic assistance for repetitive operations such as fastening, while human technicians handle fit and function checks. Repair capability is also on-site, so a failed actuator or joint can be replaced without shipping the robot elsewhere. No cost figures or headcount for the prototype operations team have been released. The one-hour target for 2027 is a stated goal rather than a demonstrated result. Independent observers have not published measurements of actual build times. The video is consistent with the company’s description of rapid iteration. It does not establish a production rate for commercial deliveries. The emphasis is on shortening the time from design change to tested hardware.

Boston Dynamics has previously shown Atlas performing dynamic motions and industrial tasks. The new video shifts attention to the manufacturing system that produces the robots used in those demonstrations. In-house gear cutting and tight-tolerance machining are highlighted as reasons the shop stays co-located with engineering. The footage includes both the machine tools and the assembly area. Staff describe a culture of real-time collaboration. No third-party audit of the one-day build claim has been published. The ambition to reach one hour is presented as a multi-year goal. The video does not show a complete robot being built from start to finish in a single continuous take. It is a montage of the processes involved. For readers tracking humanoid manufacturing, the relevant open questions are the actual cycle time under sustained operation, the supply chain for non-machined components, and when the process will support regular customer deliveries. Those details are not in the current release.

Boston Dynamics reports that Atlas can be assembled in roughly one day in its prototype shop, with a stated target of one hour by 2027. Machining, assembly, and repair occur under one roof so design changes can be tested quickly. The video shows in-house machine tools, assembly stations, and collaboration between engineers and machinists. No independent verification of daily output or of the one-hour target has been published. The footage is a company overview of the current prototype manufacturing process. Commercial production rates remain unstated. The approach is intended to accelerate hardware iteration. Broader factory output figures have not been released. The video illustrates the co-located machining and assembly workflow used for Atlas.

By humanoid.press · Editor: Humanoid Press Team · October 29, 2026

Atlas Manufacturing

~1 day

Build

1 hour

2027 Target

In-house

Machining

Watch Atlas Shop →

REVEL Explores KEYENCE Vision and Inspection for Robotic Manufacturing Cells

REVEL Robotics has released video of a visit by KEYENCE to its facility, focused on precision manufacturing and inspection technologies. The KEYENCE team demonstrated laser marking, 3D optical microscopy, machine vision, measurement systems, and sensors. The discussion explored how these tools could work alongside REVEL’s robotic manufacturing cells. One opportunity highlighted is closed-loop inspection: a robot or vision system identifies a part, verifies dimensions, tracks production history, and potentially feeds corrections back to a CNC machine. The video also covers part traceability, automated quality control that could reduce manual handling, and connected manufacturing workflows. No deployed system or production results are shown. The footage is a partner demonstration and discussion rather than a completed installation. REVEL is developing humanoid robots intended to preserve and multiply human skills in manufacturing. The visit illustrates potential integration of vision and sensing with those robotic cells. No commercial agreement or timeline for integration has been announced.

KEYENCE products shown include systems for laser marking, high-resolution optical measurement, and machine vision that can guide robots or inspect parts. In a robotic manufacturing cell, such sensors can confirm that a part is correctly oriented before a robot handles it, measure critical dimensions after machining, and log the results for traceability. The video walks through these capabilities on the REVEL floor. Staff discuss how automated inspection might reduce the need for operators to move parts between stations. No quantitative data on cycle-time reduction or defect detection rates appear in the release. The demonstration is exploratory. REVEL’s broader work includes plans to bring CNC machines in-house and to develop robots that can tend those machines. The KEYENCE visit fits that direction by showing sensing technologies that could support quality feedback. The video does not show a robot actively correcting a CNC program in real time. It is a hands-on overview of the available tools and the potential workflows.

Part traceability is a recurring requirement in precision manufacturing. Laser marking can apply permanent identifiers that vision systems later read. Combined with dimensional measurement, the same station can confirm both identity and geometry. The video includes examples of these processes and discussions of how they might integrate with robotic handling. Safety sensors for collaborative cells are also mentioned. No specific REVEL robot model is shown performing the inspection tasks in the footage. The conversation remains at the level of opportunity and technical fit. Independent case studies of KEYENCE systems in other robotic applications exist, but none are cited as results from the REVEL site. The video is useful as a record of the partner discussion. It does not constitute evidence of a working closed-loop manufacturing cell. For readers following REVEL’s manufacturing plans, the relevant next steps would be a pilot integration and measured results. Those have not been published.

REVEL Robotics is building humanoid platforms intended for industrial tasks that currently rely on skilled human labor. Integrating vision and measurement tools is one way to add quality control without adding manual steps. The KEYENCE demonstration covers both the sensing hardware and the software interfaces that could communicate with a robot controller. The video shows the team examining parts under optical systems and discussing data flow. No software architecture or communication protocol details are provided beyond the high-level discussion. The visit is framed as part of REVEL’s “Building REVEL” series. It follows earlier episodes that covered other manufacturing partners. No timeline for deploying the discussed capabilities has been stated. The footage is consistent with an early-stage technology exploration. Commercial deployment of a combined REVEL–KEYENCE cell remains unannounced. The video illustrates the inspection technologies under consideration for future robotic manufacturing workflows.

REVEL Robotics hosted KEYENCE for a demonstration of laser marking, 3D optical microscopy, machine vision, and sensing systems. The discussion explored how these tools could support robotic manufacturing cells, including closed-loop inspection that feeds corrections back to CNC machines. The video shows the technologies and the partner conversation. No production results or deployed system are presented. The footage is an exploratory overview rather than a completed integration. Part traceability and automated quality control are highlighted as potential benefits. No commercial agreement or deployment timeline has been announced. The visit fits REVEL’s broader interest in in-house manufacturing and robot-tended processes. The video remains a record of the technical discussion. Independent performance data from a combined system have not been published. The demonstration illustrates sensing options for future robotic cells.

By humanoid.press · Editor: Humanoid Press Team · October 29, 2026

REVEL Inspection

KEYENCE

Partner

Vision

Inspection

CNC

Feedback

Watch Building REVEL →

Humanoid Press — Daily Brief · October 28, 2026

Positronic and Nebius Open Physical AI Competition for Real-Hardware Evaluation

The October–December challenge invites teams to move manipulation models from simulation to real robots, with more than $20,000 in GPU and evaluation credits available.

By humanoid.press · Editor: Humanoid Press Team · October 28, 2026

PhAIL Competition

Video credit: Positronic Robotics ▶ Via YouTube

Positronic Robotics and Nebius have launched the Physical AI Competition, an open challenge running from October through December 2026. The competition is designed to move robotic manipulation models from simulation to real-world hardware. Teams can compete for more than $20,000 in GPU compute and evaluation credits. Positronic operates PhAIL, a leaderboard that evaluates vision-language-action models on real robots using commercial-style tasks such as bin-to-bin order picking. Published results on the platform show peak average completion rates around 70 percent on some benchmarks, with lower reliability and throughput on others. The video explains why the organizers consider current physical AI evaluation incomplete and how participants can submit models for hardware testing. No independent audit of the credit amounts or the number of participating teams has been published. The competition is an open call rather than a completed ranking. Registration and details are available through Positronic’s site.

Most published robot AI results come from carefully tuned laboratory setups or simulation. PhAIL measures throughput, reliability, and consistency over repeated runs on physical hardware. Inaugural evaluations compared models including OpenPI, GR00T, SmolVLA, and ACT against human and teleoperated baselines. The best AI models completed roughly half the items in timed episodes and required operator intervention every few minutes. The competition invites teams to improve on those numbers by submitting policies that are then evaluated on Positronic’s rigs. Nebius provides the compute credits. The video walks through the problem of evaluation and the structure of the challenge. It does not show a live leaderboard update or a winning entry. For developers working on manipulation models, the relevant questions are the exact task definitions, the hardware configuration used for scoring, and how credits are allocated. Those details are on the competition page. The video is an explanatory overview of the open challenge.

Positronic Robotics builds tools for connecting AI models to physical robots and evaluating them under repeatable conditions. PhAIL publishes synchronized video, telemetry, and scoring logs so results can be audited. Nebius joined as a founding compute partner earlier in 2026. The competition extends that collaboration by offering credits to teams that participate. The video highlights benchmark data showing current model limitations and the gap between simulation success and hardware performance. No list of registered teams or early submissions has been released. The challenge runs through December. Final results and credit awards will follow the evaluation period. The organizers frame the competition as a way to generate more comparable real-hardware data. Independent verification of the ~70 percent peak figure is limited to the published PhAIL tables. The video remains the primary public explanation of the competition rules and the evaluation gap it targets.

Physical AI evaluation has relied heavily on demonstration videos and single-task success rates. PhAIL was built to report units per hour and mean time between failures or assists on a commercially representative task. The competition uses that infrastructure to invite broader participation. Teams receive GPU credits for training or fine-tuning and evaluation credits for hardware runs. The video describes the process of moving a model from simulation to the real robot and the metrics that matter on the shop floor. It does not include a comparison of credit values against commercial cloud pricing. The $20,000-plus figure is the organizers’ stated total available for the challenge. No third-party confirmation of the prize pool has been published. The competition is open. Early participation numbers have not been disclosed. The video is a call for entries rather than a results announcement. It illustrates the evaluation methodology and the opportunity for teams to test models on real hardware.

Positronic Robotics and Nebius have opened a Physical AI Competition running from October through December 2026. Teams can submit manipulation models for evaluation on real robots and compete for more than $20,000 in GPU and evaluation credits. PhAIL data show peak average completion around 70 percent on some benchmarks, with lower reliability on commercial-style tasks. The video explains the evaluation gap and the competition structure. No independent audit of participation or credit distribution has been published. The challenge is an open call. Results will follow the evaluation window. The video is an explanatory overview of the hardware-evaluation opportunity. Broader industry adoption of the same metrics remains unreleased. The competition targets the gap between simulation performance and real-robot reliability. Registration is open through the Positronic site.

The Physical AI Competition from Positronic and Nebius invites teams to evaluate manipulation models on real hardware rather than simulation alone. Credits totaling more than $20,000 are available for GPU use and evaluation runs. PhAIL leaderboard results indicate current models reach roughly 70 percent peak average completion on some tasks while showing lower throughput and reliability on others. The video describes the problem and the challenge format. No list of entrants or early scores has been released. The competition runs through December. Final rankings and credit awards have not yet been announced. The video remains a promotional and explanatory overview. Independent verification of the prize total is limited to organizer statements. The challenge is designed to generate more comparable real-robot data. Participation details are on the competition page.

Competition PhAIL →

Also Today

NVIDIA Expands Isaac Platform; Agile ONE SN Reference Humanoid Launches

NVIDIA has advanced its Isaac robotics platform with new reference frameworks, simulation pipelines, and data tools. Agile Robots simultaneously launched Agile ONE SN, a compact humanoid reference platform built on NVIDIA Isaac GR00T and powered by NVIDIA Jetson Thor. The robot stands 1.40 meters tall, weighs approximately 40 kilograms, and has 31 degrees of freedom in the body plus dual WUJI Hand 2 five-finger hands with 20 degrees of freedom each. Onboard Jetson Thor provides 2,070 FP4 teraflops and 128 GB of unified memory. Skills are trained with Isaac Lab and Newton and evaluated with Isaac Lab-Arena, including on DGX Station. The platform includes preconfigured GR00T software, whole-body control, and a workflow from demonstration capture to deployment. Planned integration with the NVIDIA Halos safety stack is noted. The video shows the transition from simulation and data to physical skills on the Agile ONE. No independent production or pricing figures have been published. Availability is targeted for early 2027 for research and development use. The announcement was made around NVIDIA GTC.

Humanoid development has required teams to assemble separate tools for simulation, training, evaluation, and onboard deployment. Agile ONE SN is presented as a single system that reduces that integration work from weeks to days. The robot uses head and wrist cameras, lidar, and IMUs. The WUJI hands include tactile sensing. GR00T 1.7 is the open vision-language-action model referenced in the stack. The video illustrates frontier AI agents building training worlds, replaying real-world motion, and tuning physics with NVIDIA open libraries before skills move to the physical robot. No side-by-side comparison of setup time against prior workflows has been released. The reference platform is intended for researchers and developers rather than immediate commercial deployment. NVIDIA has previously announced other GR00T reference designs, including one based on a Unitree chassis. Agile ONE SN is a distinct compact platform from Agile Robots. The video is a product and platform overview. Quantitative performance data on specific skills remain company-reported.

NVIDIA Isaac GR00T provides open models, simulation tools, and deployment paths for humanoid policies. Jetson Thor is the onboard compute module used for real-time inference. The Agile ONE SN announcement pairs the hardware body with that software stack so developers can train in simulation and transfer to the robot without building the full infrastructure themselves. The video shows the workflow from prompts and data through simulation to physical execution. It does not include long-duration reliability tests or energy-consumption figures. For research groups evaluating reference platforms, the relevant questions are the actual transfer success rate from Isaac Lab to the physical robot, the latency of the onboard loop, and the availability of the safety stack integration. Those details are not fully specified in the announcement. The platform is positioned as a way to accelerate skill development. No customer deployments of Agile ONE SN have been announced. The video remains the primary public demonstration of the combined system.

Agile Robots is a Munich-based company. Agile ONE SN is its compact humanoid designed for R&D and built to the GR00T reference specifications. The dual five-finger hands bring the total degrees of freedom across body and hands to 71. Tactile sensing on the hands supports manipulation. The video emphasizes the open NVIDIA libraries and models used for training and evaluation. No price has been published. Early 2027 availability is a company target. The announcement expands the set of reference platforms available to researchers. It does not replace commercial humanoid products already in pilot use. The video is consistent with the written NVIDIA and Agile Robots releases. It illustrates the simulation-to-real path on the new platform. Independent third-party testing of the reference system has not been published. The launch is a platform and hardware announcement. Operational results on customer sites remain unreleased. The system is intended for laboratory skill development and validation.

NVIDIA has expanded the Isaac platform with new frameworks and pipelines while Agile Robots launched Agile ONE SN, a 1.40-meter reference humanoid built on GR00T and powered by Jetson Thor. The robot includes dual five-finger tactile hands and is scheduled for early 2027 research availability. Skills are trained in Isaac Lab and evaluated before transfer to the physical platform. The video shows the workflow from simulation and data to real-robot execution. No independent pricing or production volumes have been published. The platform is a reference system for developers. Quantitative skill-transfer results remain company-reported. The announcement is concurrent with NVIDIA’s broader Isaac updates. Broader commercial deployments of Agile ONE SN have not been announced. The video is a product overview of the integrated reference humanoid. Independent benchmarks of the onboard loop have not been released. The system is positioned to reduce integration time for humanoid skill development.

By humanoid.press · Editor: Humanoid Press Team · October 28, 2026

Agile ONE GR00T

1.40 m

Height

31+40

DoF

Thor

Compute

Read NVIDIA →

AthenaZero Uses Tendon-Driven Hands for Forceful and Delicate Grasps

Researchers at the RAI Institute have detailed the hands of AthenaZero, a low-inertia bimanual robot designed for dynamic manipulation. Each hand has three underactuated fingers driven by tendons. Motors are located in the torso and transmit force through flexible cables approximately 1.8 meters long. The fingers can interdigitate to grasp objects of different sizes, from an egg to a hammer. Closing time is about 80 milliseconds. Barometric pressure sensors in the fingertips and palm provide tactile feedback. Eight sensors are in each palm and three in each fingertip, with local compute in the back of the hand to reduce latency. The design prioritizes low end-effector mass while retaining speed and force. The video shows the hands closing and contacting objects. AthenaZero was the cover article in the September 2026 issue of Science Robotics. The paper reports throwing speeds above 30 m/s and catching and batting above 14 m/s. No independent commercial availability or pricing has been announced. The robot is a research platform on a fixed base rather than a full humanoid.

Conventional robot hands often place actuators in the fingers or wrist, adding mass that reduces acceleration. AthenaZero remotes the hand motors to the torso and uses Bowden cable transmissions. The underactuated three-finger design allows the fingers to wrap around objects of varying diameter. The barometric sensors detect contact location and force at approximately 0.1 N resolution and 200 Hz. Local computation in the hand reduces the latency of the tactile loop. The video illustrates the combination of speed and gentleness. The Science Robotics paper focuses on the overall low-inertia arm design and demonstrates baseball-inspired tasks: throwing, catching, and batting. Robot-to-robot and human-to-robot catch games are described. The hands are a component of that system. No long-duration reliability data for the cable transmissions or sensor calibration drift have been published. The design is optimized for dynamic interaction rather than continuous static payload. The video is a hardware demonstration of the hand mechanism. Quantitative force and speed figures come from the published paper.

AthenaZero stands on a fixed base with a twisting torso and two 7-DoF arms. The low effective end-point mass—comparable to a human arm and roughly an order of magnitude below many conventional manipulators—is achieved through quasi-direct-drive actuators and transmission remotization. The hands contribute to that goal by keeping finger actuators out of the moving mass. The paper reports that the robot can throw, catch, and bat on human-comparable timescales. The video focuses on the hand hardware and tactile sensing. It does not show the full throwing or catching sequences. For researchers studying contact-rich manipulation, the relevant details are the cable friction losses, the peak fingertip force, and the sensor resolution under dynamic impact. Those measurements are in the Science Robotics paper. The platform is a laboratory system. No commercial product based on AthenaZero has been announced. The video remains a visual explanation of the tendon-driven, sensorized hand design.

Tactile sensing in robot hands is often limited by latency or by the mass of the sensing hardware. AthenaZero places barometric sensors in flexible fingertip and palm materials and processes them locally. The eight palm sensors and three per fingertip give contact location at roughly 3 mm resolution. The 80-millisecond closing time is a company and paper figure for the unloaded hand. Loaded closing times and maximum grasp force under the Bowden transmission are reported in the paper. The video shows the fingers interdigitating around different objects and the sensors responding to light contact. It is consistent with the published description. Independent replication of the 80-millisecond figure has not been published outside the RAI Institute work. The design trades some transmission efficiency for reduced end-effector inertia. The result is a hand that can move quickly and still sense contact. The video is a research demonstration rather than a product launch. Broader deployment of the same hand design has not been announced.

AthenaZero’s three-fingered, tendon-driven hands close in about 80 milliseconds and can grasp both delicate and forceful objects. Motors are remoted to the torso via cables. Barometric sensors in the fingertips and palm provide contact feedback with local compute. The design is part of a low-inertia bimanual platform published in Science Robotics. The video shows the hand mechanism and tactile response. No commercial availability has been announced. Throwing, catching, and batting results are reported in the paper. The hands prioritize speed and low mass. Independent long-term reliability data for the transmissions have not been published. The platform is a research system on a fixed base. The video is a hardware overview of the sensorized hands. Quantitative performance figures come from the published study. Broader adoption of the design remains unreleased.

By humanoid.press · Editor: Humanoid Press Team · October 28, 2026

AthenaZero Tactile

80 ms

Close

3

Fingers

Tactile

Sensors

Paper Science Robotics →

Humanoid Daily News — 10/27/2026

Humanoid Daily News - October 27th, 2026

10/27/2026

Primo: Open-Source Humanoid Shared Before It Is Built

3D-printed design with 30 RobStride actuators and Jetson AGX Thor — still in CAD.

Read full story →

10/27/2026

RoboParty Opens RP1 as a Full-Stack Humanoid

CAD, actuators, firmware, and training tools released; hardware already walks and recovers.

Read full story →

10/27/2026

Menlo Ships Asimov 1 Open-Source Humanoid Kits

From a broken commercial ankle to a self-built kit in one year.

Read full story →

10/27/2026

TUM RoboGym Opens; NEURA Shows 4NE1 Gen 4

Munich training center launches with a lighter humanoid running Neuron AI.

Read full story →

10/27/2026

RealMan’s ECO62 Fits Six Axes in a Small Space

3.3 kg body, 355 mm radius, ≤100 W — built for tight automation.

Read full story →

Humanoid Press — Daily Brief · October 26, 2026

Lunes Robotics Shows Parcel Sorting Taught by Demonstration and One Sentence

The system splits a video demonstration into known steps, adds label reading and routing from a short prompt, then executes the sort on its own.

By humanoid.press · Editor: Humanoid Press Team · October 26, 2026

Lunes Teaching

Video credit: Lunes Robotics ▶ Via YouTube

Lunes Robotics has released a video showing a robot learning to sort cartons by destination without conventional programming. The task routes boxes addressed to Augsburg to one side and all other cartons to the other. A human demonstrates the physical sorting on video. A single sentence then supplies the routing rule based on the postal code on the label. The system matches the demonstration to skills it already knows, adds the steps the video does not show—locating the carton, reading the label, and applying the rule—and then executes the sort. Demonstration, teaching, and execution segments were recorded separately and are presented in sequence. Parts of the execution are shown at double speed. No success rates, cycle times, or results across varied label formats have been published. The video is a teaching-method demonstration rather than a completed production deployment. Lunes Robotics develops hardware-agnostic software for humanoid and industrial robots intended for factory tasks.

Traditional robot programming for sorting typically requires explicit coding of grasp points, motion paths, and decision logic. The Lunes approach uses a video demonstration for the physical steps and a short text prompt for the rule that the video cannot convey. The system is described as recognizing the carton, decomposing the shown motion into known skills, and inserting the missing perception and decision skills. The video shows the robot reading labels and directing cartons accordingly after the teaching step. It does not show the internal skill library or the model used to match and compose those skills. For manufacturers evaluating reduced programming effort, the relevant open questions are how many prior skills must already exist, how the system handles label variations or occlusions, and what the failure rate is on a full shift. Those details are not included in the promotional video. The sequence illustrates demonstration-plus-prompt teaching under controlled conditions.

Lunes Robotics is based in Germany and focuses on industrial-grade software that can run on different humanoid platforms for repetitive factory work. Earlier material from the company has described vision-language models for workpiece detection and vision-language-action concepts for grasping, including teleoperated demonstrations. The parcel-sorting video isolates the teaching pipeline for a logistics-style task. The Augsburg routing rule is a simple binary condition. More complex routing tables or multi-destination sorts are not shown. The video notes that execution segments are sped up. No comparison of teaching time against conventional programming is provided. The demonstration is consistent with the company’s positioning of software that turns generic robots into workers for material handling. Independent trials in an operating warehouse have not been released. The clip remains the primary public evidence of the demonstration-and-prompt method.

Parcel sorting by destination is a common warehouse task that still involves manual labor in many facilities. Automating it with fixed systems usually requires structured conveyors and barcode scanners. The Lunes video shows a robot arm handling individual cartons after a short teaching sequence. Whether the same method scales to mixed carton sizes, damaged labels, or higher throughput is not addressed. The company states that the system adds skills the video never showed, including label reading and routing. No technical description of the label-reading model or its accuracy on real postal codes has been published. The video is labeled as teaching robots by showing. It is useful as a qualitative illustration of reduced coding for a defined sort. Broader claims of autonomous night-shift operation would require longer trials and reliability data that have not been released. The demonstration focuses on the teaching and execution sequence for one routing rule.

Lunes Robotics has shown a robot learning to sort cartons to Augsburg versus elsewhere through a video demonstration and a single-sentence prompt. The system matches known skills and adds label reading and routing. Execution is then shown at normal and double speed. No independent metrics of accuracy or throughput have been published. The video is a controlled teaching demonstration. The method is presented as an alternative to writing explicit code for the task. Operational results in a live warehouse remain unreleased. The sequence illustrates demonstration-based skill composition for a logistics sort. Final reliability under varied conditions has not been reported. The platform continues to be developed as industrial software for humanoid and other robots.

Lunes Robotics has demonstrated a teaching method in which a video of parcel sorting is combined with a short text prompt to produce autonomous execution. The robot reads labels and routes cartons according to the stated rule. Demonstration, teaching, and execution were recorded separately. Parts of the execution appear at double speed. No success-rate or cycle-time data have been published. The video is a capability illustration of reduced programming for a defined sort. Broader warehouse deployments of the same method have not been announced. The approach relies on an existing skill library that the system matches and extends. Independent verification of label-reading accuracy has not been released. The clip shows the teaching pipeline under controlled conditions.

Watch Demo Lunes →

Also Today

XPENG IRON Adjusts Mid-Gesture Using On-Device VLA 2.0

XPENG has released video of its IRON humanoid following changing hand gestures without spoken commands. In the sequence a tester offers a handshake, switches to a fist bump, and moves into a high-five. When the target shifts, the robot brings in its other hand. The company attributes the behavior to an upgraded VLA 2.0 model running on-device, with stronger temporal understanding and future-behavior prediction. XPENG states that the robot reassesses the visual scene approximately every 0.2 seconds and predicts action changes within about six seconds. The footage is labeled an R&D test-version demonstration. No success rates, end-to-end latency figures, or results across multiple users have been published. The video shows continuous observation and adjustment rather than a fixed routine. IRON is XPENG’s general-purpose humanoid that uses the company’s Physical AI foundation. The same model family is used in the company’s vehicles.

Fixed gesture recognition followed by a complete pre-planned motion fails when a person changes intention midway. Continuous re-evaluation while the arm is already moving requires the model to update its plan in real time. XPENG’s description of on-device VLA 2.0 inference matches the visual result in the clip. The 0.2-second interval and six-second horizon are company figures. Independent timing measurements have not been released. The demonstration is useful as a qualitative illustration of adaptive upper-body interaction. It does not show the same behavior in unstructured public environments or with people who move more erratically than the staged tester. The company notes that final product behavior may differ from the development footage. No contact-force data or safety margins for the handshake and high-five interactions have been published. The sequence isolates the gesture-following loop under controlled conditions.

IRON incorporates a spine-like core, bionic muscle layout, and soft skin covering. Earlier XPENG videos have shown the robot walking, tracking speakers, and holding multi-language conversations. The current clip focuses on upper-body coordination and the ability to switch limbs when the target moves. The on-device claim indicates that inference occurs on the robot’s compute rather than in the cloud. IRON carries multiple Turing chips. The video does not show the internal compute configuration or power draw during the interaction. Continuous adjustment during a short social gesture is a limited but visible test of the loop. Broader claims of general social competence would require longer trials with more varied participants. No such data have been released. The demonstration is consistent with XPENG’s positioning of IRON as a platform that combines physical motion with real-time AI. The video remains the primary public evidence of the adaptive interaction behavior.

VLA 2.0 is XPENG’s vision-language-action model. The company has deployed versions in its vehicles and is applying an upgraded version to IRON. The video shows the robot switching hands and modifying its motion mid-interaction when the high-five position changes or a palm closes into a fist. The footage is a controlled development demonstration. No independent latency or success-rate data have been published. The demonstration focuses on adaptive upper-body interaction. Broader operational results in unstructured settings remain unreleased. The platform continues to be developed as a general-purpose humanoid using XPENG’s Physical AI stack. The sequence demonstrates on-device temporal reasoning applied to social motion. Final product behavior may differ from the R&D test version shown. No long-duration or multi-user trials of the same capability have been released.

XPENG’s IRON humanoid has been shown adjusting in real time during handshakes, fist bumps, and high-fives without spoken commands. The company attributes the behavior to an upgraded on-device VLA 2.0 model with temporal understanding and prediction. The robot is said to reassess the scene roughly every 0.2 seconds. The video demonstrates continuous observation and motion adjustment under controlled conditions. No independent metrics have been published. The sequence is an R&D test-version demonstration. The clip illustrates adaptive social interaction on the IRON platform. No long-duration or unstructured public trials of the same capability have been released. The demonstration focuses on upper-body coordination. Broader claims of general social competence remain unreleased. The platform continues development as a general-purpose humanoid.

By humanoid.press · Editor: Humanoid Press Team · October 26, 2026

IRON VLA 2.0

0.2 s

Reassess

On-device

VLA

R&D

Demo

Watch IRON →

UBTECH Starts Liuzhou Factory; Unitree Opens UnifoLM-WLA-1.0

UBTECH has started production at a factory in Liuzhou, Guangxi, that the company describes as designed for 10,000 industrial humanoid robots a year. The 14,000-square-meter plant, built with Siemens Digital Industries Software, uses flexible work cells rather than fixed conveyors. Cruzr robots handle depalletizing, parts transport, and replenishment alongside AGVs and unmanned forklifts. The company states that the line is timed for one completed robot roughly every ten minutes. Primary models are the Walker S and Cruzr series. Finished units undergo multi-hour stress testing and visual inspection before storage in a compact high-bay warehouse. The video shows robots and automated logistics inside the facility. No independent production-volume figures or sustained takt-time measurements have been published. The ten-minute interval is a company design target. Actual output over a full operating year has not been independently verified. The plant is presented as the first of its scale for industrial humanoids.

Separately, Unitree has fully open-sourced UnifoLM-WLA-1.0, a 6-billion-parameter embodied foundation model. The company states that a single model coordinates 64 tasks spanning tabletop manipulation and full-body mobile work, and generalizes across two-finger grippers and multiple five-finger dexterous hands. Training used approximately 2,500 hours of real-robot data. The model includes an embodied reasoner component and an action expert. Code, weights, and related datasets are available on GitHub and Hugging Face under an open license. The release followed a staged open-source plan that began in mid-September. No independent benchmarks of the 64-task set have been published outside company evaluations. The model is intended for developers working with Unitree platforms, particularly the G1. The open-source release lowers the barrier for fine-tuning on local data. Performance on customer robots remains dependent on further adaptation. The video for this brief focuses on the UBTECH factory; the Unitree model is a concurrent software announcement.

UBTECH’s Liuzhou facility is intended to move the company from lower-volume assembly toward higher-throughput manufacturing. The use of its own Cruzr robots for material handling is presented as “robots building robots.” Siemens software provides the manufacturing operations management layer, including work-order scheduling and serial-number traceability. The plant covers both bipedal Walker S units and wheeled Cruzr platforms. No customer order backlog or shipment figures tied specifically to the new line have been released. The video shows hanging Walker units, mobile robots, and the overall layout. It does not include a continuous multi-hour production run or a full accounting of human versus robotic labor on the line. For manufacturers evaluating humanoid production capacity, the relevant questions are the actual annual output, the yield rate after testing, and the cost structure at the stated volume. Those details are company-reported. The facility is an industrial-scale claim rather than a fully audited production result.

Unitree’s UnifoLM-WLA-1.0 is described as unifying tabletop and whole-body manipulation in one weight set. The embodied reasoner is built on a vision-language backbone and is trained on large volumes of spatial and multimodal samples. The action expert is trained on the real-robot hours. The company has released model weights, training code for action experts, and fine-tuning tools. Datasets referenced include Unitree open collections and other real-robot sources. The open-source step is notable because Unitree is a high-volume humanoid maker. Developers can now fine-tune the model on their own data rather than relying solely on published claims. No side-by-side comparison against other open foundation models on the same 64 tasks has been released. The model is a software release concurrent with the UBTECH factory news. Both announcements address scaling—one on the manufacturing side, one on the model side. Independent validation of either claim remains limited to company statements and early open-source materials.

UBTECH has begun production at a Liuzhou factory designed for 10,000 industrial humanoids a year, with a stated takt of one robot roughly every ten minutes. Cruzr robots and automated logistics handle much of the material movement. Separately, Unitree has open-sourced UnifoLM-WLA-1.0, a 6B model that the company says handles 64 tasks across tabletop and full-body work with different end-effectors. The factory video shows the plant layout and robotic material handling. No independent volume or takt measurements have been published. The model release includes code and weights. Performance on the full task set remains company-reported. Both developments address different aspects of scaling humanoid deployment. Actual sustained output and independent model benchmarks have not been released. The announcements are concurrent rather than linked. The factory is an industrial production claim. The model is an open-source software release.

By humanoid.press · Editor: Humanoid Press Team · October 26, 2026

UBTECH Unitree

10k

Capacity

10 min

Takt

6B

Model

Watch UBTECH →

Humanoid Press — Daily Brief · October 25, 2026

LimX TRON 2 Pairs with TENNIIX for Tennis Court Coverage

Enhanced Robotics and LimX Dynamics combine the multi-form robot’s all-terrain mobility with TENNIIX PRO for fast movement, jumps, net stops, and real-time returns.

By humanoid.press · Editor: Humanoid Press Team · October 25, 2026

TRON 2 Tennis

Video credit: LimX Dynamics / Enhanced Robotics ▶ Via YouTube

Enhanced Robotics and LimX Dynamics have released video of the TRON 2 multi-form robot operating on a tennis court with the TENNIIX PRO system. The clip shows the robot covering the court, jumping, stopping near the net, and returning balls in real time. LimX describes TRON 2 as a modular platform that can switch between dual-arm, bipedal, and wheeled-leg configurations. The tennis demonstration uses the robot’s mobility for rapid lateral movement and recovery. No independent measurements of ball-return accuracy, court-coverage speed, or success rate across repeated rallies have been published. The video is a capability demonstration rather than a completed product announcement with technical specifications. TRON 2 is marketed for research, data collection, and industrial prototyping. The tennis application is presented as one example of physical AI deployment on the platform.

TRON 2 features 7-DoF arms, a dual-arm payload of up to 10 kg in standard configuration, and mobility that varies by form. In wheeled-leg mode the company lists speeds of roughly 3–5 m/s and a 30 kg ground payload. The bipedal form supports stair climbing with visual planning. The tennis video emphasizes all-terrain movement and dynamic stops rather than detailed manipulation of the racket. Enhanced Robotics supplies the TENNIIX PRO ball-return and practice system. The collaboration pairs the robot’s mobility with an existing tennis-practice device. No pricing for the combined system or availability timeline has been released. The demonstration shows the robot moving quickly across the court and recovering after jumps. It does not include long-duration continuous play or comparison against human practice partners. The footage is useful as a qualitative illustration of the platform’s mobility range.

LimX Dynamics is a Shenzhen-based company focused on embodied intelligence platforms. TRON 2 was introduced as a follow-on to the earlier TRON 1 biped and is sold in standard and EDU versions. The modular design allows the same base to be configured for different research tasks. The tennis demonstration is one of several application videos the company has released. No customer deployments of the TENNIIX combination have been announced. The video does not show the internal control architecture or the sensors used for ball tracking. For researchers evaluating mobile platforms, the relevant questions are the stability during rapid direction changes, the energy cost of repeated jumps, and the latency of the ball-return loop. Those details are not included in the promotional clip. The demonstration remains a visual example of dynamic mobility on a sports court.

Tennis practice systems have traditionally used fixed ball machines or human coaches. Pairing a mobile robot with an existing ball-return device is a new configuration. The video shows the robot positioning itself and executing returns without a human operator visible in the frame. Whether the motion is fully autonomous or assisted by teleoperation is not stated. LimX has previously published teleoperation latency figures for TRON 2 in the range of 100 ms. The tennis sequence does not include those measurements. The collaboration is framed as accelerating physical AI deployment. No revenue figures or order volumes tied to the tennis application have been disclosed. The video is the primary public evidence of the combined system in operation. Claims of fast coverage and precise net stops are company statements. Independent verification has not been published.

Enhanced Robotics and LimX Dynamics have shown the TRON 2 robot operating on a tennis court with the TENNIIX PRO system. The video depicts rapid court coverage, jumps, net stops, and real-time ball returns. TRON 2 is a modular multi-form platform with dual-arm and wheeled-leg options. No independent performance metrics have been released. The demonstration is a capability illustration rather than a finished commercial product. The collaboration pairs robot mobility with an existing tennis practice device. Operational results under continuous play remain unreleased. The footage shows the platform’s dynamic movement on a sports surface.

LimX Dynamics’ TRON 2 multi-form robot has been demonstrated with Enhanced Robotics’ TENNIIX PRO system on a tennis court. The video shows fast lateral movement, jumps, precise stops near the net, and ball returns. The company presents the sequence as an example of physical AI deployment. No quantitative accuracy or endurance data have been published. The demonstration is controlled. Broader availability of the combined system has not been announced. The clip illustrates the robot’s mobility range in a sports setting. Independent testing of the ball-return loop has not been released. The platform continues to be developed for research and application use.

Watch Demo TRON 2 →

Also Today

Pollen Robotics Microduck Draws Strong Pre-Order Interest

Pollen Robotics’ Microduck, a 25 cm biped priced at $399 before taxes and shipping, has drawn substantial pre-order volume since its August 27 launch. The robot stands under 800 g, has 15 motors, a front camera, an 8×8 depth LiDAR, two IMUs, microphones, a speaker, NFC, Wi-Fi, and Bluetooth. Out of the box it can walk, sit, crouch, roller-skate, pick objects with its beak, and recover from many common falls. The open-source software stack allows developers to train new behaviors in simulation and transfer them to the robot. First deliveries were targeted before Christmas 2026 in North America, Europe, and the UK. After the initial rush the company stated it could no longer guarantee Christmas delivery for new orders and now estimates four to six months. The video shows the robot walking, falling, recovering, and performing trained behaviors. Demand figures reported by the company and third-party coverage exceed 10,000 units in the first week and later surpassed 20,000 after a public demonstration.

The combination of a low price, a small form factor, and an open reinforcement-learning pipeline has generated interest among makers and researchers. Microduck is not positioned as a finished household appliance. It is sold as a platform on which users can run supplied policies or train new ones. The software, including the SDK and training tools, is open source. Hardware designs are not fully open in the same way. The robot can be controlled with a game controller, physical accessories, and NFC-tagged objects. Multiple units can be used together for simple multi-robot experiments. The video demonstrates the default behaviors and the recovery sequence. No independent long-term reliability data from early customers have been published because the first units have not yet shipped in volume. The pre-order response indicates strong interest in an affordable, trainable biped. Delivery timelines have lengthened as order volume grew.

Pollen Robotics, part of Hugging Face, previously released Reachy Mini. Microduck adds legs and learned locomotion. The company has stated that default policies run locally on the robot. Developers can extend the stack and share new behaviors. The four colorways are Cream, Graphite, Lavender, and Sky. A game controller is included. The video shows the robot’s size relative to a desk and its ability to recover from falls. Search interest in the name rose sharply after the launch and again after a high-profile demonstration. The company has not published a final production capacity figure. Early reports placed first-week orders above 10,000 units. Later coverage indicated the total had moved past 20,000. These numbers are company and media reports rather than audited figures. The product remains in the pre-order phase. First shipments have not yet reached customers at scale.

Small, affordable robots that support on-device learned policies have been limited. Most research platforms are larger and more expensive. Microduck’s price point and open software stack lower the barrier for experimenting with reinforcement learning on physical hardware. The robot’s articulated beak can pick small objects. Its sensors support basic interaction and recovery. The video illustrates the walking gait, fall recovery, and object interaction. No formal safety certification for unsupervised use around children has been published. The company positions the robot for developers and enthusiasts rather than as a fully autonomous home companion. Demand has exceeded the initial delivery window. The company has adjusted its shipping estimate accordingly. The video remains the primary public demonstration of the robot’s out-of-box behaviors.

Pollen Robotics’ Microduck has attracted strong pre-order interest at a $399 price point. The 25 cm biped includes 15 motors, multiple sensors, and an open-source software stack for training new behaviors. Default actions include walking, recovery, and object picking. First deliveries were targeted before Christmas 2026; new orders are now estimated at four to six months. The video shows the robot’s size, gait, and recovery. Demand figures reported publicly exceed 10,000 units early and later moved higher. The robot is a maker and research platform rather than a finished consumer appliance. No large-scale customer reliability data have been published because shipments are still pending. The combination of price and open training tools has driven the response.

By humanoid.press · Editor: Humanoid Press Team · October 25, 2026

Microduck $399

$399

Price

25 cm

Height

15

Motors

Pre-order Microduck →

NEURA Robotics Introduces Neuron AI Multimodal Model

NEURA Robotics has introduced Neuron AI, a multimodal foundation model for Physical AI developed and trained in-house in Metzingen, Germany. The model uses a Perception-Language-Action architecture that combines 3D vision, tactile and force signals, audio, robot state, and language into a shared representation updated over time. A memory layer is included so the robot can retain context when objects leave the camera view. Neuron AI is trained on data from robots operating in NEURA Gyms, from simulation, and from people wearing sensor-equipped suits. The company states that the same model runs across five of its robot embodiments and their grippers and hands. Customers can fine-tune the model for their own processes through the Neuraverse platform and deploy the results to fleets. The video features CEO David Reger and Business Unit Head Dr. Jens Buchner explaining the approach. No independent benchmarks of task success rates or latency have been published. The announcement was made in mid-October.

Vision-language-action models primarily use camera images. Neuron AI adds touch, force, and audio so the robot can respond to events that cameras alone cannot capture, such as the click of a connector or the resistance of a screw. The memory component is intended to support object permanence and causality. Training data come from three sources: real robot interactions in NEURA Gyms, simulated variation, and human demonstrations. The company plans to expand the Gyms globally to increase data volume and reduce the time required to train new skills. The model is available for customer fine-tuning via Neuraverse. The video is an explanatory interview rather than a task demonstration. NEURA has previously shown its 4NE1 humanoid performing longer sequences. No quantitative comparison against vision-only models has been released. The claims of multimodal fusion and memory are company statements. Independent verification of performance gains has not been published.

NEURA Robotics builds several robot platforms, including the 4NE1 humanoid and industrial arms. Neuron AI is positioned as a shared foundation across these embodiments. The company says the model encodes the different sensor streams in one space over time and produces a single stream of action. Fine-tuning by customers is offered so that site-specific processes can be added without retraining the entire model from scratch. The video shows the executives describing the architecture and the training approach. It does not include side-by-side task results or failure-mode analysis. For operators evaluating foundation models for physical systems, the relevant questions are the data volume required for a new skill, the latency of the closed loop when touch and audio are included, and the safety margins during contact-rich tasks. Those details are not provided in the announcement or the video. The model is an architecture and training update rather than a finished, independently benchmarked product.

Physical AI models that incorporate touch and audio remain less common than vision-centric systems. NEURA’s approach treats the additional sensors as first-class inputs rather than optional add-ons. The memory layer is described as necessary for tasks in which objects move out of view. The company has stated that Neuron AI is the only multimodal foundation model of its type developed in-house in Germany. No third-party confirmation of that uniqueness claim has been published. The video is consistent with the written announcement. It provides an accessible explanation of the PLA architecture and the three training data sources. Operational results on customer sites have not yet been reported. The Neuraverse platform is the stated channel for fine-tuning and fleet deployment. The announcement is a model and platform update. Performance data remain company-reported.

NEURA Robotics has introduced Neuron AI, a multimodal foundation model that combines 3D vision, touch, audio, robot state, and language with a memory layer. The model is trained on robot data from NEURA Gyms, simulation, and human demonstrations and runs across five robot embodiments. Customers can fine-tune it via Neuraverse. The video features company executives explaining the architecture. No independent benchmarks have been published. The model is positioned as a shared brain for the company’s robots. Claims of multimodal advantage and reduced skill-training time are company statements. The announcement is an architecture and training update. Operational results on deployed fleets remain unreleased. The video is an explanatory overview of the model.

By humanoid.press · Editor: Humanoid Press Team · October 25, 2026

Neuron AI PLA

PLA

Model

5

Embodiments

Memory

Layer

Learn More Neuron AI →

Humanoid Press — Daily Brief · October 24, 2026

XPENG IRON Follows Changing Gestures Without Spoken Commands

The humanoid adjusts mid-motion during handshakes, fist bumps, and high-fives using on-device VLA 2.0, reassessing the scene roughly every 0.2 seconds.

By humanoid.press · Editor: Humanoid Press Team · October 24, 2026

IRON VLA 2.0

Video credit: XPENG ▶ Via YouTube

XPENG has released video of its IRON humanoid following a person’s changing hand gestures without spoken commands. In the sequence a tester offers a handshake, switches to a fist bump, and then moves into a high-five. When the target shifts beyond the reach of the arm IRON first extends, the robot brings in its other hand. The company attributes the behavior to an upgraded VLA 2.0 model running on-device, with stronger temporal understanding and future-behavior prediction. XPENG states that the robot reassesses the visual scene approximately every 0.2 seconds and predicts action changes within about six seconds. The footage is labeled an R&D test-version demonstration. No success rates, end-to-end latency figures, or results across multiple users have been published. The video shows continuous observation and adjustment rather than a fixed routine completed before the next response. IRON is XPENG’s general-purpose humanoid that uses the company’s Physical AI foundation.

Fixed gesture recognition followed by a complete pre-planned motion is simpler to implement but fails when a person changes intention midway. Continuous re-evaluation while the arm is already moving requires the model to update its plan in real time. XPENG’s description of on-device VLA 2.0 inference matches the visual result in the clip. The same model family is used in the company’s vehicles. Transferring temporal prediction capability to the humanoid is a company claim. The 0.2-second interval and six-second horizon are XPENG figures. Independent timing measurements have not been released. The demonstration is useful as a qualitative illustration of adaptive upper-body interaction. It does not show the same behavior in unstructured public environments, under varying lighting, or with people who move more erratically than the staged tester. The company notes that final product behavior may differ from the development footage.

IRON incorporates a spine-like core, bionic muscle layout, and soft skin covering. The interaction test focuses on upper-body coordination and the ability to switch limbs when the target moves. Earlier XPENG videos have shown the robot walking, tracking speakers with a microphone array and lip reading, and holding multi-language conversations. The current clip isolates the gesture-following loop. No contact-force data or safety margins for the handshake and high-five interactions have been published. For researchers studying closed-loop human-robot interaction, the relevant open questions are the actual latency of the vision-to-motion cycle, the robustness of the prediction when intentions change rapidly, and the force limits during physical contact. Those details are not included in the promotional video. The sequence demonstrates on-device temporal reasoning applied to social motion under controlled conditions.

VLA 2.0 is XPENG’s vision-language-action model. The company has deployed versions in its vehicles and is applying an upgraded version to IRON. The on-device claim indicates that inference occurs on the robot’s compute rather than in the cloud. IRON carries multiple Turing chips. The video does not show the internal compute configuration or power draw during the interaction. Continuous adjustment during a short social gesture is a limited but visible test of the loop. Broader claims of general social competence would require longer trials with more varied participants and environments. No such data have been released. The demonstration is consistent with XPENG’s positioning of IRON as a platform that combines physical motion with real-time AI. The video remains the primary public evidence of the adaptive interaction behavior.

XPENG has shown its IRON humanoid adjusting in real time during handshakes, fist bumps, and high-fives without spoken commands. The company attributes the behavior to an upgraded on-device VLA 2.0 model with temporal understanding and prediction. The robot is said to reassess the scene roughly every 0.2 seconds. The video demonstrates continuous observation and motion adjustment under controlled conditions. No independent metrics have been published. The sequence is an R&D test-version demonstration. Final product behavior may differ. The clip illustrates adaptive social interaction on the IRON platform. No long-duration or unstructured public trials of the same capability have been released.

XPENG’s IRON humanoid follows changing hand gestures in real time without spoken commands. The company links the behavior to an upgraded on-device VLA 2.0 model that reassesses the scene approximately every 0.2 seconds. The video shows the robot switching hands and modifying its motion mid-interaction. The footage is a controlled development demonstration. No independent latency or success-rate data have been published. The demonstration focuses on adaptive upper-body interaction. Broader operational results in unstructured settings remain unreleased. The platform continues to be developed as a general-purpose humanoid using XPENG’s Physical AI stack.

Watch Demo IRON →

Also Today

Siipotec and KUKA Automate Wood Production at ET Listat

Finnish integrator Siipotec has delivered an automated production system for ET Listat that combines KUKA industrial robots, a 36-meter linear track, and autonomous mobile robots. The line sorts cut-to-length wood products and directs them to up to 18 stations based on length and quality. Two KUKA robots retrieve finished layers from sorting bins and perform handling operations that include 180-degree rotation around two axes. A third KUKA robot mounted on the 36-meter linear track assembles individual product layers into finished bundles. AMRs then transport the bundles. Siipotec describes the installation as an award-winning solution that improves productivity, consistency, ergonomics, and material flow while reducing waste. The video shows the robots, track, and AMR movement inside the ET Listat facility. No independent throughput measurements or before-and-after production figures have been published outside company statements.

Wood-processing lines often involve long, heavy, and variable-length pieces that are difficult to handle manually. The combination of fixed robots, a long linear track, and mobile robots allows the system to cover a large area without requiring workers to move between stations. The linear-track robot can reach multiple sorting and bundling positions. Servo-controlled guides and positioning systems are used to maintain accuracy. Siipotec specializes in automation for the mechanical wood industry and has previously delivered robotic cells and production lines for Finnish manufacturers. The ET Listat project is presented as a complete material-handling and internal-logistics solution. The company has not released cycle times, uptime percentages, or energy-consumption data. The video illustrates the physical layout and the movement of the robots and AMRs. It does not include quantitative performance metrics.

KUKA industrial robots are widely used for material handling, packaging, and assembly. Mounting one on a 36-meter track extends its working envelope for long products. The AMRs handle the final transport step, removing the need for fixed conveyors across the entire facility. Siipotec integrated the robots, track, vision or sensing systems, and AMR fleet control. The installation is one of several wood-industry projects the integrator has completed. No customer quote or specific award name is given in the publicly available description beyond the general “award-winning” characterization. The video is a customer-success overview rather than a technical deep dive. For manufacturers evaluating similar lines, the relevant questions are the actual throughput gain, the maintenance requirements of the long track, and the reliability of the AMR handoff. Those details are not provided in the promotional material.

ET Listat manufactures building mouldings, profiles, decoration panels, and wooden floors. The company produces large volumes of cut-to-length products that must be sorted, bundled, and moved. The Siipotec system automates the sorting, bundling, and internal transport steps. The 36-meter track robot and the AMR fleet are the two most visible elements in the video. The project demonstrates how industrial robot arms, linear motion, and mobile robots can be combined for variable-length material handling. No data on worker reassignment, injury reduction, or long-term reliability have been released. The video shows the system operating in the factory. Claims of higher productivity and improved ergonomics are company statements. The installation is a practical example of integrated industrial automation rather than a humanoid-robot application.

Siipotec has installed a KUKA-based automation system at ET Listat that uses industrial robots on a 36-meter linear track and AMRs to sort, bundle, and transport wood products. The solution is described as award-winning and is intended to improve productivity, quality, and ergonomics. The video shows the robots and mobile units in operation. No independent performance metrics have been published. The project is a conventional industrial-robot and AMR installation for the wood industry. It demonstrates integrated material handling across a large floor area. Quantitative results remain company-reported. The system is an example of production-line automation rather than humanoid deployment.

By humanoid.press · Editor: Humanoid Press Team · October 24, 2026

KUKA AMR

36 m

Track

18

Stations

AMRs

Transport

Watch Siipotec →

Luwu Dynamics Releases Open-Source XGO-Duck Based on Microduck

Luwu Dynamics has published the hardware, runtime, and training resources for XGO-Duck, a 3D-printable biped adapted from Pollen Robotics’ Microduck. The robot uses an Arduino UNO Q (Qualcomm processor running Linux plus STM32 microcontroller), 15 Feetech 1910 serial servos, a custom expansion board with a QMI8658A IMU, and printed structural parts. The supplied runtime includes walking, get-up, and pick policies that run on the UNO Q. The hardware repository contains STL files, PCB designs, a bill of materials, and an assembly guide. The project is explicitly credited to Pollen Robotics’ Microduck and microduck_rl work. The video introduces the hardware, assembly, on-robot software, and simulation workflow. No commercial production timeline or large-scale user base has been announced. The robot is a maker-oriented open-source platform rather than a finished consumer product.

Microduck is a small duck-shaped biped with 15 degrees of freedom that Pollen Robotics has shown walking and interacting. XGO-Duck retains the body shape, joint layout, and joint order but replaces the electronics and servos. The Arduino UNO Q splits compute: the Linux side runs policy inference and a web interface, while the STM32 handles real-time servo and sensor communication. The expansion board manages power, the servo bus, and the IMU. Luwu Dynamics has released ONNX policy files for walking, recovery, and picking so builders can start without training their own models. A separate training repository allows users to explore reinforcement learning. The video shows the assembly process and the robot executing learned motions. Independent long-duration reliability data from multiple builders have not been published. The project is an open hardware and software adaptation intended for makers who want to experiment with biped locomotion and on-device policies.

Open-source robot platforms lower the barrier for experimenting with learned motion policies. XGO-Duck provides printable parts, a documented electronics stack, and ready-to-run policies. Builders still need to source the Arduino UNO Q, the 15 servos, the expansion board, and batteries, then calibrate the joints. The runtime includes a web interface for control and monitoring. The training repository is separate and requires a suitable GPU machine for new policy development. The video walks through the hardware, assembly steps, and the transition from simulation to the physical robot. No formal safety certification or commercial support package has been announced. The project is suitable for research and hobby use. Claims of reliable walking or picking are limited to the supplied policies under the conditions shown. Broader performance depends on individual builds and calibration.

Pollen Robotics previously demonstrated Microduck and has made related software available. Luwu Dynamics adapted the design for the Arduino UNO Q and Feetech servos and published the full set of files. The expansion board is a custom design that integrates the IMU and servo interfaces. The video shows both the mechanical assembly and the software side, including policy execution. The project does not include a camera or advanced perception in the base configuration. Additional sensors can be added by builders. No large community deployment numbers have been reported. The repositories are public on GitHub. For makers and students, the combination of printable mechanics, dual-chip compute, and learned policies offers a complete starting point. The video is a project overview rather than a controlled benchmark.

Luwu Dynamics has released open-source hardware and software for XGO-Duck, a 3D-printable biped based on Pollen Robotics’ Microduck. The robot uses an Arduino UNO Q, 15 Feetech servos, and a custom expansion board. Supplied policies cover walking, get-up, and pick. The video shows hardware, assembly, and on-robot software. The project is a maker platform rather than a commercial product. No independent large-scale reliability data have been published. Builders can start with the supplied policies or explore the training repository. The adaptation is credited to the original Microduck work. The robot is intended for hands-on experimentation with learned motion.

By humanoid.press · Editor: Humanoid Press Team · October 24, 2026

XGO-Duck Open Source

15

Servos

UNO Q

Compute

3D Print

Structure

Watch XGO-Duck →

Humanoid Press — Daily Brief · October 23, 2026

FORT Robotics Raises About $20 Million as It Prepares Nasdaq Listing

Tiger Global leads new commitments with participation from Mark Cuban Companies; total capital tied to the proposed SPAC exceeds $50 million.

By humanoid.press · Editor: Humanoid Press Team · October 23, 2026

Funding Safety

Video credit: FORT Robotics ▶ Via YouTube

FORT Robotics announced on October 22 approximately $20 million in additional funding commitments from existing investors. Tiger Global is leading an approximately $10 million financing, with participation from Mark Cuban Companies, expected to close this week. Tiger Global has also committed an additional $10 million to a previously announced common equity PIPE connected to FORT’s proposed business combination with Newbury Street II Acquisition Corp. Together with earlier PIPE commitments, the total capital tied to the transaction now exceeds $50 million. Upon closing, the combined company is expected to be named FORT Robotics Holdings, Inc. and list on Nasdaq under the ticker FROB, subject to regulatory approvals. FORT develops the Trust Layer for Physical AI, a safety platform currently deployed across more than 19,500 units to over 600 customers, including Agility Robotics, DoorDash, Zoox, and Google DeepMind. The video shows the company’s Wireless E-Stop Pro wearable safety device.

The additional capital is intended to expand the team and operating capacity as customer demand grows across warehousing, defense, construction, and other markets adopting autonomous systems. FORT has positioned its products as foundational safety infrastructure that allows robot manufacturers and end users to certify safety and accelerate deployment. The company holds 25 patents and has been backed by Tiger Global since its Series B in 2022. The proposed SPAC combination values the business at roughly $500 million pre-financing, with an enterprise value near $557 million. No new product performance data or customer contract values were released with the funding announcement. The Wireless E-Stop Pro is a SIL 3-certified wearable device with 200-meter range and an IP65 enclosure, designed for operators to stop machines remotely. The video demonstrates that product rather than the funding itself.

Physical AI safety has drawn increasing attention at industry events. Speakers at RoboBusiness in Santa Clara this week repeatedly cited safety as a prerequisite for scaling autonomous systems. FORT’s Trust Layer is one commercial response to that requirement. The company says its solutions are used in warehousing, transportation, manufacturing, construction, agriculture, mining, energy, and defense. The funding expands resources for growth but does not include disclosed revenue figures or order backlogs. The Nasdaq listing, if completed, would make FORT the first publicly traded company focused principally on physical AI safety, according to the company. Regulatory approvals and closing conditions remain outstanding. The video illustrates one of FORT’s mobile safety products in use on industrial equipment.

FORT was founded in 2018 by Samuel Reeves. The company has expanded from hardware-centric safety devices to a broader software and connectivity platform. Recent milestones include an expanded partnership with Agility Robotics and the introduction of Connectivity Aggregation software. The new commitments from Tiger Global and Mark Cuban Companies are incremental to earlier financing. No independent verification of the 19,500-unit deployment figure has been published outside company statements. The Wireless E-Stop Pro video shows the device pairing with machines and providing status feedback. It is relevant as an example of the safety hardware the company sells. The funding announcement is the primary news; the video provides product context.

The proposed listing under the FROB ticker remains subject to regulatory review and other closing conditions. FORT intends to use the capital to support new and expanding customer opportunities. The company has not published updated financial results with the announcement. The Wireless E-Stop Pro is marketed for dense worksite environments where operators need mobile protection. The video demonstrates its range, status display, and rugged enclosure. For observers tracking capital flows into physical AI infrastructure, the $20 million commitment and the path toward a Nasdaq listing are the material developments. Deployment numbers and customer names are company-reported. The transaction has not yet closed.

FORT Robotics has secured approximately $20 million in additional funding commitments led by Tiger Global with participation from Mark Cuban Companies. Total capital connected to its proposed SPAC with Newbury Street II exceeds $50 million. The company expects to list on Nasdaq as FORT Robotics Holdings under the ticker FROB if the combination closes. Its Trust Layer is deployed across more than 19,500 units to over 600 customers. The video shows the Wireless E-Stop Pro safety device. The funding is intended to expand capacity as demand grows. The listing remains subject to approvals. The announcement is a capital and public-market update rather than a product launch.

Company Site FORT →

Also Today

NVIDIA GTC Berlin Highlights Agile ONE, Vera Rubin, and RTX Spark

NVIDIA CEO Jensen Huang delivered the GTC Berlin 2026 keynote on October 21. Announcements included the Agile ONE SN reference humanoid developed with Agile Robots, scheduled for early 2027; RTX Spark Windows laptops with up to 128 GB of memory for local AI agents; the Vera Rubin DSX Pod, a containerized AI factory design that NVIDIA says can deploy in as little as 14 days at a prepared site; and confirmation that Vera Rubin is in full production, with Huang claiming another roughly 30-times performance leap over Blackwell. OpenShell and Sentry security tools for autonomous AI agents were also introduced. The video is a third-party recap that also covers Unitree’s H2-X demonstration. All performance and availability figures are company claims. No independent benchmarks were released with the keynote.

The Agile ONE SN is a 31-DoF reference platform built on NVIDIA Isaac GR00T and Jetson Thor. NVIDIA says it ships preconfigured with compute and software to reduce researcher setup time from weeks to days. Agile Robots is targeting Q1 2027 availability. RTX Spark is a line of Windows PCs with RTX graphics and up to 1 petaflop of local AI performance. Vera Rubin NVL72 systems are liquid-cooled and connect 72 GPUs per rack. The DSX Pod packages these systems in modular containers for faster site deployment. OpenShell is an open-source runtime for isolating and monitoring AI agents. The video attributes every figure to the announcing company. Huang’s keynote also addressed European AI infrastructure and Hyperion vehicle platforms. The humanoid and compute announcements are the portions most relevant to robotics.

Reference platforms are intended to give developers a validated starting point rather than a finished commercial product. The Agile ONE SN is one such system. NVIDIA has not published pricing or production volumes. Vera Rubin’s “full production” status has been stated at multiple events; racks are reported to be running at customers. The 30-times claim is a company comparison under specific workloads. Independent measurements of end-to-end robot performance on the new stack have not been released. The video recap includes the Unitree H2-X high-impact demonstration alongside the GTC announcements. It does not add new technical data. For teams evaluating onboard compute or simulation tools, the Jetson Thor and Isaac GR00T updates are the practical takeaways. Availability timelines remain early 2027 for the reference humanoid and DSX Pods.

NVIDIA has expanded its robotics platform with GR00T models, Isaac tools, and Jetson Thor modules. Multiple companies, including Agile Robots, are building on the stack. The GTC Berlin keynote presented these elements together with data-center and personal-AI hardware. The video is a summary that also covers Unitree’s recent H2-X video. No new customer deployments or quantitative robot task success rates were announced. The reference humanoid is positioned for research and development. RTX Spark targets local agent workloads on Windows. Vera Rubin targets large-scale inference and training. All claims are attributed to NVIDIA or the partner companies. The keynote is a platform update rather than a product launch with immediate availability.

NVIDIA’s GTC Berlin keynote highlighted the Agile ONE SN reference humanoid, RTX Spark laptops, Vera Rubin in full production, and the DSX Pod containerized AI factory design. OpenShell agent security tools were also introduced. The video is a recap that includes Unitree’s H2-X demonstration. All specifications and timelines are company statements. The Agile ONE is scheduled for early 2027. No independent performance data were released. The announcements expand NVIDIA’s robotics and AI infrastructure offerings. Operational results on robots will depend on subsequent integrations. The keynote is a platform and hardware update.

By humanoid.press · Editor: Humanoid Press Team · October 23, 2026

GTC Berlin NVIDIA

Early 2027

Agile ONE

Vera Rubin

Production

RTX Spark

Local AI

XPENG IRON Adjusts in Real Time During Handshake and High-Five

XPENG has released video of its IRON humanoid adjusting mid-interaction during handshakes, fist bumps, and high-fives. When a palm closes into a fist or the high-five position changes, the robot switches hands or modifies its motion without completing a fixed routine first. The company attributes the behavior to an upgraded VLA 2.0 model with stronger temporal understanding and future-behavior prediction, running on-device. XPENG says the robot reassesses the scene approximately every 0.2 seconds and predicts action changes within about six seconds. The demonstration is a controlled interaction test. No independent latency measurements or success rates across varied users have been published. The video shows continuous observation and adjustment rather than a single pre-scripted sequence. IRON is XPENG’s general-purpose humanoid powered by the company’s Physical AI stack.

Real-time motion following requires the robot to update its plan while an action is already under way. Fixed gesture recognition followed by a complete motion is simpler but less flexible when a person changes intention. XPENG’s description of continuous observation and on-device VLA 2.0 inference is consistent with the visual result in the video. The model is also used in the company’s vehicles. Transferring the same foundation to the humanoid is a company claim. The 0.2-second reassessment interval and six-second prediction horizon are XPENG figures. No third-party timing data have been released. The video is useful as a qualitative demonstration of adaptive social interaction. It does not show the robot performing the same behavior in unstructured public settings or under varying lighting and crowd conditions.

IRON features a spine-like structure, bionic muscle layout, and soft skin. The interaction sequence tests upper-body coordination and the ability to switch limbs when the target moves. XPENG has previously shown IRON walking, tracking speakers, and holding multi-language conversations. The current video focuses on physical gesture following. The company notes that the footage is from development and testing and that the final product may differ. No force or contact-safety data for the handshake interactions have been published. For researchers studying human-robot interaction, the relevant questions are the latency of the closed loop, the robustness of the prediction model when intentions change rapidly, and the safety margins during contact. Those details are not included in the promotional clip. The demonstration illustrates on-device temporal reasoning applied to social motion.

VLA 2.0 is XPENG’s vision-language-action model. The company has deployed it in vehicles and is now applying an upgraded version to IRON. The on-device claim means inference occurs on the robot’s compute rather than in the cloud. IRON carries multiple Turing chips. The video does not show the internal compute configuration. Continuous adjustment during a handshake is a limited but visible test of the loop. Broader claims of general social competence would require longer, more varied trials. No such data have been released. The demonstration is consistent with XPENG’s positioning of IRON as a platform that combines physical motion with real-time AI. The video is the primary public evidence of the adaptive interaction behavior.

XPENG has shown its IRON humanoid adjusting in real time during handshakes, fist bumps, and high-fives. The company attributes the behavior to an upgraded on-device VLA 2.0 model with temporal understanding and prediction. The robot is said to reassess the scene every 0.2 seconds. The video demonstrates continuous observation and motion adjustment. No independent metrics have been published. The sequence is a controlled development demonstration. Final product behavior may differ. The clip illustrates adaptive social interaction on the IRON platform.

By humanoid.press · Editor: Humanoid Press Team · October 23, 2026

IRON VLA 2.0

0.2 s

Reassess

On-device

VLA 2.0

Real-time

Adjust

Watch Demo IRON →

Humanoid Press — Daily Brief · October 22, 2026

Mind Children Shows MC1 Codey Social Humanoid Prototype

The child-sized wheeled robot is built for conversation, facial expression, and social interaction in education, healthcare, and hospitality settings.

By humanoid.press · Editor: Humanoid Press Team · October 22, 2026

Codey Social

Video credit: Mind Children ▶ Via YouTube

Mind Children has released new footage of MC1, the social humanoid prototype also known as Codey. The robot is approximately three feet tall, rides on wheels, and is designed for conversation, expressive facial movement, and social engagement rather than industrial labor. The company identifies education, healthcare, and hospitality as potential application areas. Co-founders Chris Kudla and Ben Goertzel have described the robot as a low-cost interface for human connection. Target production price is under $10,000. A second unit named Joy has been built with Korean manufacturing partners. The Seattle prototype can hold eye contact, respond to emotional cues, and carry on autonomous conversations using onboard computing and AI. Physical capabilities remain limited; the robot cannot yet fetch objects or perform complex manipulation. In the video, Codey remains stationary and introduces itself through speech and facial expression.

Social humanoids differ from factory or warehouse platforms in both form and function. MC1 uses a child-sized body, a stylized gray face with servo-driven expressions, and low-torque motors intended to fail safely on contact. Mind Children has tested the robot’s responses to inappropriate prompts and has designed the architecture so the underlying language model can be replaced. Recent updates include integration of SingularityNET’s OmegaClaw agent framework, which the company says allows the robot to remember past interactions and adapt. The prototype has appeared at technology events and has been shown in controlled indoor settings. No large-scale public deployments have been announced. Pilot studies in museums and event spaces are planned for later this year. A second-generation model is targeted for 2027. The video focuses on the robot’s conversational introduction.

Affordability is a central design constraint. Most full-size humanoids cost tens of thousands of dollars. Mind Children argues that a sub-$10,000 price point is necessary for classrooms, care facilities, and hospitality venues. The modular 3D-printed construction of the current prototype is expected to transition to more conventional manufacturing as volumes increase. The company has raised crowdfunding and is backed by SingularityNET. Physical limits are acknowledged: the robot is not intended for heavy lifting or unsupervised outdoor operation. Near-term goals include teaching simple button-pushing and expanding memory for relationship building. The video shows Codey speaking and presenting itself without locomotion. Independent testing of long-term reliability and user acceptance has not been published.

Mind Children was founded in 2023 in the Seattle area. The team includes engineers with backgrounds in automotive, aerospace, and consumer products. Codey’s personality is described as curious and childlike, with the ability to make simple jokes and maintain conversational context. The company has said the robot is aware of its surroundings and can describe scenes in front of it. These capabilities depend on the quality of the language model and the sensor suite. The video shows the robot introducing itself through voice and expression while remaining in place. It does not demonstrate autonomous navigation or multi-day continuous operation. Those remain development targets. For observers tracking social robotics, the low target price and the explicit focus on education and care distinguish the project from industrial humanoid efforts.

The MC1 prototype is still in the laboratory and demonstration stage. Mind Children has stated that commercial units will be hardened for field use. Current limitations include restricted physical dexterity and the need for controlled environments. The company plans pilot deployments in public attractions by the end of 2026 and broader use in healthcare and education thereafter. The video provides a clear view of the robot’s face, speech, and introductory style while it stays stationary. No performance metrics on conversation success rate, battery life, or safety incidents have been released. The project is an early-stage social humanoid aimed at human interaction rather than labor replacement.

Mind Children has shown its MC1 Codey social humanoid prototype. The child-sized wheeled robot is designed for conversation, facial expression, and social engagement in education, healthcare, and hospitality. Target price is under $10,000. Physical capabilities are limited. In the video, Codey introduces itself through speech and expression while remaining stationary. Pilot deployments are planned for later this year. The platform remains a prototype. No large-scale operational data have been published. The project focuses on affordable social robotics rather than industrial tasks.

Learn More Codey →

Also Today

Unitree Releases H2-X Kung Fu Guardian Demonstration

Unitree has released a promotional video titled “H2-X Kung Fu Legend Guardian.” The clip shows the humanoid performing high-impact martial arts movements, jumping from a platform labeled four meters high, landing, and recovering. It also depicts staged protection and rescue scenes, including intervening with a knife-wielding figure and lifting the front of a car so a dummy can be pulled free. Unitree states the footage is real and has not been sped up. The company lists a maximum endurance of approximately eight hours while standing. No detailed hardware specifications, pricing, or availability for the H2-X have been published. The video is a dynamic capability demonstration rather than a product announcement with technical data. The standard Unitree H2 is approximately 182 cm tall with 31 degrees of freedom; the H2-X relationship to that platform is not specified beyond the name.

High-impact demonstrations test joint strength, whole-body coordination, and impact recovery. The H2-X sequence includes kicks, punches that damage a car door, aerial flips, and a high drop onto mats. The protection and car-lift scenes are staged. Unitree has not published force measurements, success rates across repeated trials, or energy consumption during the stunts. The eight-hour standing endurance figure is a company claim. Dynamic operation under load would reduce that figure. The video has circulated widely and generated discussion about the physical power of current humanoid platforms. It does not address safety systems, autonomous decision-making, or real-world deployment constraints. The company includes a note to use the robot safely and in a friendly manner. The demonstration is visual evidence of dynamic motion capability.

Unitree has previously shown its robots performing parkour, flips, and high-speed locomotion. The H2-X video continues that pattern with higher-impact actions and a guardian narrative. The car-lift sequence requires significant torque at the arms and torso. Whether the motion is fully autonomous or partially teleoperated is not stated. No independent verification of the eight-hour endurance or the authenticity of every frame has been released. For researchers, the relevant questions are the control policies that generate the motions, the actuator limits that allow the impacts, and the recovery behaviors after falls. Those details are not included in the promotional clip. The video is useful as a qualitative illustration of the platform’s dynamic range. It is not a technical datasheet.

Unitree is a Chinese company known for commercial quadruped and humanoid platforms. The H2 series has been marketed for research and industrial use. The H2-X video adds a security-themed presentation. No contracts or deployments for a guardian role have been announced. The demonstration does not show the robot operating unsupervised in public spaces or interacting with real people outside staged scenes. Battery life, compute architecture, and sensor suite for the H2-X remain undisclosed. The video is the primary public evidence of the platform’s high-impact capabilities. Claims of real, unsped footage are company statements. Quantitative performance data have not been provided.

Unitree has released a video of the H2-X performing martial arts, high drops, and staged protection and rescue actions. The company states the footage is real and lists approximately eight hours of standing endurance. No detailed specifications or independent verification have been published. The video is a dynamic capability demonstration. The guardian narrative is staged. The clip shows the physical range of the platform under controlled conditions. Real-world autonomous security use has not been demonstrated. The demonstration is visual evidence of high-impact motion.

By humanoid.press · Editor: Humanoid Press Team · October 22, 2026

H2-X Unitree

~8 h

Endurance

4 m

Drop

Staged

Demo

Watch Demo H2-X →

Humanoid Expands NVIDIA Collaboration for Onboard Compute

UK-based Humanoid has announced an expanded collaboration with NVIDIA to develop a high-performance robotic edge runtime for its next-generation robots. The companies are integrating NVIDIA Jetson Thor, NVIDIA Isaac OS, NVIDIA Holoscan Sensor Bridge, and the Holoscan SDK. Humanoid says the move to Jetson Thor has allowed it to consolidate two onboard computers into a single system while maintaining computing capacity, with potential additional headroom. Unified CPU and GPU memory on Jetson Thor is expected to reduce data movement and latency. Holoscan Sensor Bridge places camera data directly into GPU-accessible memory. Isaac OS provides a GPU-native runtime for coordinating perception, inference, and control. Multi-Gigabit Ethernet supplies up to 100 Gb of bandwidth for sensor and control signals. The company has used NVIDIA Isaac Lab and Isaac Sim since its early days for training and simulation. The announcement was made on October 21. The video is the company’s general platform overview.

Onboard compute is a constraint for humanoid robots that must process high-bandwidth sensor data and run AI models in real time. Moving from two computers to one reduces hardware complexity, power draw, and integration effort. Humanoid states that the new architecture simplifies how different workloads share resources. The collaboration does not include published benchmarks of inference latency, power consumption, or task success rates on the new stack. NVIDIA has released Jetson Thor modules and Isaac tools for general robotics development. Humanoid is one of several companies adopting the platform. The press release quotes CEO Artem Sokolov on the need for significant computing power and real-time integration of sensing, AI, and control. No deployment timeline or robot model name for the next generation is given. The work is framed as foundational architecture rather than a finished product release.

Humanoid positions itself as a UK AI and robotics company focused on commercially scalable and safe humanoid robots. Earlier work has included proofs of concept in industrial settings and the use of NVIDIA simulation tools. The expanded collaboration extends that relationship into the onboard runtime. Holoscan is intended to lower latency between sensors and models. Isaac OS is intended to coordinate multiple robotics workloads on the GPU. These are engineering choices rather than demonstrated performance gains. Independent measurements of the consolidated computer’s effect on robot behavior have not been released. The video shows the company’s robots and facilities but does not specifically demonstrate the new Jetson Thor configuration. The announcement is a technology partnership update. Operational results will depend on subsequent hardware and software integration.

NVIDIA has been expanding its robotics platform with Jetson Thor, Isaac GR00T, and related tools. Multiple European and global robotics companies are building on the stack. Humanoid’s announcement is consistent with that broader trend. The company says the technologies will make the platform easier to develop, test, and scale. No customer deployments or production volumes tied to the new compute architecture have been stated. The October 21 press release is the primary source. The video is relevant as company context but does not contain the technical details of the collaboration. For observers tracking compute choices in humanoid development, the consolidation to a single Jetson Thor system is the concrete engineering claim. Performance data remain unreleased.

Humanoid is expanding its collaboration with NVIDIA to build an onboard edge runtime using Jetson Thor, Isaac OS, Holoscan Sensor Bridge, and related tools. The company says Jetson Thor consolidates two computers into one while maintaining capacity. Unified memory and direct sensor-to-GPU paths are expected to reduce latency. The announcement was made on October 21. No independent benchmarks have been published. The video is a general company overview. The collaboration is an architecture and development partnership. Operational results have not yet been demonstrated. The work is intended to support the company’s next generation of robots.

By humanoid.press · Editor: Humanoid Press Team · October 22, 2026

NVIDIA Jetson Thor

Jetson Thor

Compute

1 system

From 2

Holoscan

Sensors

Company Site Humanoid →

Humanoid Press — Daily Brief · October 21, 2026

REVEL Partners with Mazak to Manufacture Robot Parts In-House

Three Mazak CNC machines will be installed at REVEL’s Prague headquarters so the company can prototype components and develop robots that tend the machines.

By humanoid.press · Editor: Humanoid Press Team · October 21, 2026

Mazak CNC

Video credit: REVEL Robotics ▶ Via YouTube

REVEL Robotics has announced a partnership with Mazak that will place three CNC machines at its Prague headquarters. The Mazak team visited the facility to discuss using the machines to manufacture robot components while REVEL develops its humanoid robots to tend those same machines. The company says the arrangement will give its engineers more control over prototyping, testing, and iterating parts in-house. REVEL builds humanoid robots intended to preserve and multiply skilled human work in manufacturing and other sectors facing labor shortages. The Prague site is the company’s European engineering and R&D base. The partnership is presented as an early step toward robots that help produce more robots. No timeline for the machines’ installation or for robot tending capability has been stated. The video is an episode of the company’s Building REVEL series showing the visit and the components under discussion.

Bringing CNC capacity inside the headquarters is a practical manufacturing decision. REVEL currently relies on external suppliers for many parts. Longer lead times and limited customization options have been cited by the company as constraints. Installing Mazak machines is expected to shorten the loop between design and physical test. At the same time, the company is working on the control and perception systems its robots would need to load, unload, and interact with CNC equipment. That dual goal—faster in-house prototyping plus robots that can operate the machines—is the stated ambition of the collaboration. The video shows the Mazak visitors and the Prague workspace. It does not include footage of robots already tending machines. The engineering work to reach that point remains ahead.

REVEL’s approach centers on capturing skilled human motion and judgment, then deploying that capability on humanoid platforms. The company has previously discussed targets of scaling production and has open roles across engineering in Prague. Adding CNC machines is consistent with a strategy of controlling more of the hardware stack. Mazak is a major machine-tool manufacturer with a presence in Central Europe. The partnership gives REVEL access to industrial CNC platforms while giving Mazak a potential future application in robot-assisted machine tending. Neither company has published quantitative milestones for the collaboration. The public evidence is the announcement of the visit and the plan to install the three machines. The video documents the meeting and the current state of the Prague facility.

Machine tending is a known application for industrial robots, but humanoid form factors remain uncommon in that role. Most CNC cells use fixed or mobile industrial arms with custom end effectors and fixturing. A humanoid would need reliable grasping of varied parts, safe interaction with machine doors and controls, and recovery from errors without constant human intervention. REVEL has not claimed those capabilities exist today. The partnership is framed as a development path: manufacture parts with Mazak machines now, and develop the robots that can support that work later. The video is useful as a record of the planning discussion. It does not demonstrate autonomous machine tending. For observers tracking how humanoid companies approach manufacturing, the in-house CNC step is a concrete operational move.

REVEL was founded in 2025 and maintains offices in Palo Alto and Prague. The company has described its robots as tools for skilled work that is difficult to staff. The Mazak collaboration is the latest public update on its manufacturing plans. Three CNC machines are planned for the Prague headquarters. The company intends both to use those machines for prototyping and to develop robots capable of tending them. The video shows the Mazak team’s visit and the components the company expects to produce. Installation timing and robot performance on the machines have not been detailed. The partnership is an early-stage manufacturing and development agreement rather than a completed production system.

REVEL Robotics is partnering with Mazak to install three CNC machines at its Prague headquarters. The machines will be used to manufacture robot components in-house. The company is also developing its humanoid robots to tend CNC equipment. The video records a visit by the Mazak team and outlines the current plans. No robots are shown operating machines. The collaboration is a manufacturing and R&D step. Further engineering is required before the robots can support the machines. The announcement is the public status of the effort.

Company Site REVEL →

Also Today

UniX AI Shows Panther on Autonomous Facility Patrol

UniX AI has released video of its Panther wheeled dual-arm robot performing facility patrol tasks. The robot is shown moving through low-light outdoor areas, office floors, and parking garages while using multimodal perception, adaptive manipulation, and anomaly detection. Panther is powered by the company’s UniTac, UniCortex, and UniFlex systems. UniX AI describes the platform as capable of 24/7 autonomous patrols in unstructured environments. The robot features mass-produced 8-DoF bionic arms, adaptive grippers, and an omnidirectional four-wheel steering and drive chassis. Earlier this year the company announced real-home deployments of Panther for household tasks. The facility-patrol video extends the same platform to security and inspection scenarios. No independent verification of continuous unsupervised operation has been published. The clip is a company demonstration of the robot in mixed indoor and outdoor settings.

Facility patrol requires navigation in spaces that change, detection of anomalies, and occasional physical interaction with equipment or barriers. UniX AI’s materials list fire-equipment inspection, route planning, and multi-floor coverage among the intended uses. The video shows the robot moving and performing inspection-style actions. It does not include long-duration logs, false-positive rates, or comparisons with human patrols. The company has previously claimed monthly production volumes above 100 units for its wheeled platforms and has positioned Panther for commercial service, home, and industrial roles. The patrol application is one of several the company markets. The video is relevant to that claim. Performance under real security contracts remains unstated in the public materials.

Wheeled dual-arm robots have advantages in indoor and mixed environments where stairs are limited. The omnidirectional chassis and vertical reach of the upper body allow Panther to access floor-level and elevated objects without repositioning the base for every task. UniX AI’s stack separates perception and manipulation (UniTac/UniTouch), long-horizon planning (UniCortex), and cross-scenario generalization (UniFlex). The company says these systems support the patrol behaviors. The video is the latest visual evidence. It does not replace quantitative testing of detection reliability or recovery from unexpected obstacles. For facility operators evaluating robotic patrol, the relevant questions are battery life, weather sealing, and integration with existing security systems. Those details are not fully specified in the current demonstration.

UniX AI is a Suzhou-based company founded in 2024. It has raised funding and begun deliveries of wheeled humanoid platforms. Panther is its third-generation full-size wheeled dual-arm robot. The facility-patrol video shows the robot in environments that include low light and mixed surfaces. The company presents the system as ready for 24/7 autonomous use. The evidence is the demonstration footage and the technical stack description. Independent long-term operational data have not been released. The video illustrates the platform’s mobility and inspection actions in the shown settings. Broader claims of continuous unsupervised security coverage remain company statements.

UniX AI has shown its Panther robot performing facility patrol in indoor and outdoor environments. The platform uses UniTac, UniCortex, and UniFlex and features 8-DoF arms on a wheeled chassis. The video demonstrates navigation, perception, and inspection-style tasks. No independent metrics of continuous autonomous operation have been published. The demonstration is consistent with the company’s security and inspection positioning. Real-world contract performance has not been detailed. The clip is a company demonstration of the patrol capability.

By humanoid.press · Editor: Humanoid Press Team · October 21, 2026

Panther Patrol

8-DoF

Arms

Wheeled

Base

24/7

Claim

Company Site UniX AI →

Mission Canvas Combines Navigation, Behavior Trees, and VLA

Mission Canvas is an open framework that lets developers define map-based, long-horizon robot tasks by combining LiDAR navigation, Behavior Trees, and Vision-Language-Action models. Users create a map, place waypoints, attach a Behavior Tree to each destination, and set the order of execution. Navigation moves the robot between waypoints. Behavior Trees organize the actions at each stop. VLA models supply adaptable manipulation where rule-based steps are insufficient. The framework is presented as a tool for experimentation and research rather than a finished commercial product. Code, documentation, and tutorial videos are available. The system builds on earlier Action Canvas work that defined single-location tasks. Mission Canvas extends that approach across multiple locations. The video demonstrates the workflow of mapping a space, designing missions, and watching the robot execute the sequence.

Long-horizon tasks require the robot to move, act, recover from local failures, and continue. Pure VLA policies can handle flexible manipulation but often lack reliable multi-location sequencing and safety checks. Pure rule-based systems are predictable but brittle when the environment changes. Mission Canvas places both under a Behavior Tree structure so that navigation, deterministic actions, and learned policies can be mixed at each waypoint. The developer can add new capabilities, such as pedestrian detection, by extending the tree. The video shows the interface and a robot carrying out a multi-location sequence. It does not include large-scale reliability statistics or comparisons against other orchestration tools. The value claimed is the ability to compose existing skills into repeatable multi-area missions without rewriting low-level code for every new workflow.

Behavior Trees are a standard way to structure robot tasks because they support sequencing, fallbacks, and parallel actions in a way that is easier to inspect than large neural policies. Adding VLA inference as a leaf node lets the tree call a learned policy when the situation requires adaptability and fall back to rule-based steps when precision or safety is required. Mission Canvas implements that pattern on top of LiDAR-based navigation. The open-source release includes the code needed to reproduce the workflow. Researchers can modify the trees, swap VLA models, or add new perception modules. The video is a demonstration of the current interface and execution loop. It does not claim that the framework solves general long-horizon autonomy. It shows one practical way to connect navigation, rules, and learned actions on a mapped space.

The framework is associated with ROBOTIS development tools and has been discussed in the Open Robotics community. Action Canvas handled single-workstation tasks. Mission Canvas adds the spatial layer. Both are positioned for developers who already have robot skills and want a way to deploy them across a building or workspace. The video walks through map creation, waypoint placement, Behavior Tree design, and execution. No production deployment numbers or customer case studies are attached to the tool. It remains a research and development framework. For teams building mobile manipulation systems, the open code and the clear separation between navigation, rules, and VLA are the practical contributions. The demonstration illustrates the intended workflow.

Mission Canvas is an open framework for map-based, long-horizon robot tasks. It combines LiDAR navigation, Behavior Trees, and VLA models so developers can define multi-location missions. Users place waypoints, attach task trees, and run the sequence. The video shows the mapping, design, and execution steps. Code and documentation are available. The tool is intended for research and experimentation. No large-scale reliability data have been published. The framework is a composition layer for existing robot capabilities rather than a complete autonomy solution.

By humanoid.press · Editor: Humanoid Press Team · October 21, 2026

Mission Canvas VLA

Open

Source

BT + VLA

Mix

Map

Based

View Code GitHub →

Humanoid Daily News — 10/20/2026

Humanoid Daily News - October 20th, 2026

10/20/2026

JR West Rolls Out New Railway Maintenance Robots

Single-arm heavy lifter and improved humanoid for overhead lines and tree cutting.

Read full story →

10/20/2026

UniX AI Shows Panther for Facility Patrol

Multimodal perception and adaptive manipulation for 24/7 unstructured sites.

Read full story →

10/20/2026

ultraArm P1 Conveyor Kit for Hands-On Automation

Dual-arm depalletizing, 3D vision picking, and long-run integration practice.

Read full story →

10/20/2026

DR02 Pro Imitation-Learning Tutorial

Full pipeline from motion retargeting to tracking policy and robot deployment.

Read full story →

10/20/2026

OmniHand 3 Lite: Mouse-Sized Dexterous Hand

11 DoF, 350 g, 0.5 s open/close, built for small humanoids and drop-tested.

Read full story →

Humanoid Press — Daily Brief · October 19, 2026

RoboParty Opens ROBOTO 01 as Full-Stack Humanoid Platform

RP1 is an open-source bipedal humanoid with mechanical designs, actuators, firmware, and control tools released for researchers and developers.

By humanoid.press · Editor: Humanoid Press Team · October 19, 2026

RP1 Open Source

Video credit: RoboParty ▶ Via YouTube

RoboParty has released RP1, also called ROBOTO 01, as a full-stack open-source bipedal humanoid platform. The company unveiled the robot at IROS 2026 in late September and has since published mechanical CAD, bill of materials, assembly documentation, actuator designs, firmware, low-level drivers, simulation models, and SDKs. RP1 builds on ROBOTO ORIGIN (RPO), which RoboParty open-sourced in January 2026 and which has accumulated more than 2,500 GitHub stars. The company describes RP1 as a high-performance platform for researchers, educators, and embodied AI teams who want to modify hardware, train policies, and deploy on real robots. Peak joint torque is listed at up to 160 N·m from in-house ROMOMO actuator modules. Configurations include 24-DoF and 30-DoF versions. The face-down recovery sequence in the slowed video is one of the motion capabilities the company has shown.

RoboParty states that full-stack open source means more than mechanical files. The planned releases cover motion-control systems, simulation environments, training tools, and the PartyOS development foundation. Developers can inspect and modify layers from mechanical structure and actuator response through control policies. The company says RP1 has demonstrated continuous walking, recovery from disturbances, getting back up after a fall, and human motion tracking. The open toolchain is intended to support iteration across models, control, and hardware. Pricing has not been disclosed in the announcements. A mass-production program and further open-source details are planned for later in the fourth quarter. The developer community is described as more than 5,000 participants.

Open-source humanoid platforms remain uncommon at the full-stack level. Most commercial humanoids keep hardware designs and low-level control closed. RoboParty’s approach starts from the earlier RPO project, which released mechanical and electrical designs, and extends that model to a higher-performance platform with stronger actuators and a more complete software stack. The recovery demonstration shows the robot starting face-down and returning to a standing position without hands attached, so the body and lower-limb motion are visible. That sequence is a practical test of balance and whole-body control after a fall. The company presents it as one capability of the open platform rather than a finished product claim.

Researchers who want to train locomotion or manipulation policies need a robot they can modify and a simulation that matches the hardware. RoboParty’s releases of URDF, MJCF, and actuator details are intended to support that workflow. The 160 N·m peak torque figure is higher than many small research humanoids. Whether the platform reaches the reliability needed for continuous lab use will depend on the quality of the open-source documentation and the number of community builds that succeed. The IROS debut and the subsequent GitHub releases are the public evidence. The recovery video is a visual example of one motion the control stack can produce.

RoboParty was founded in 2025 and has focused on open-source humanoid development from the start. The team includes members from Tsinghua, Peking University, Stanford, Carnegie Mellon, and other institutions. The company positions RP1 as a shared foundation rather than a closed product. The open-source roadmap includes progressive release of additional software and hardware details. For labs that want to study fall recovery, sim-to-real transfer, or whole-body control without building a robot from scratch, the platform is one of the more complete open options currently available. The video shows the recovery motion; the GitHub repositories contain the designs and tools.

RoboParty has opened RP1 (ROBOTO 01) as a full-stack open-source humanoid platform. Mechanical designs, actuators, firmware, simulation models, and control tools are available. The robot has shown fall recovery, walking, and disturbance rejection. It builds on the earlier ROBOTO ORIGIN project. The slowed recovery video illustrates one of the motion capabilities. The platform is intended for researchers and developers who want to modify and extend a real humanoid rather than treat it as a black box. Further open-source releases and a production program are planned for later in the year.

View on GitHub RP1 →

Also Today

Linkerbot L30 Hand Plays High-Speed Rhythm Game

Linkerbot has shown its L30 tendon-driven dexterous hand playing a rhythm game at high speed. The hand is a flagship model with a tendon-driven architecture that moves actuators out of the fingers, reducing distal mass. Published specifications list 17 active degrees of freedom plus passive joints, for a total around 21 joints, with fingertip force sensors, repeatability of ±0.20 mm, and joint speeds exceeding 400°/s on core motions. The hand weighs approximately 1.4 kg. Control is over CAN FD. The rhythm-game demonstration shows rapid, coordinated finger movements timed to on-screen prompts. Linkerbot has previously shown the L30 performing fine-motor tasks such as screw tightening and object manipulation. The game clip is a speed and coordination demonstration rather than a new product claim.

Tendon-driven hands can achieve higher finger speeds than designs that place motors in the fingers because the moving mass is lower. The L30 is marketed for research, industrial assembly, and tasks that require human-like dexterity. Company materials list fingertip forces in the 3.5–4 N range for individual fingers and higher grasp forces with multiple fingers. The rhythm-game video is one of several high-speed clips Linkerbot has released. It does not include independent timing measurements or a comparison against human players under controlled conditions. The visual result is a hand that keeps up with rapid note sequences. Whether the same control precision holds for longer, more variable tasks depends on the software and sensing used with the hand.

Dexterous hands remain a bottleneck for many humanoid and manipulation systems. A hand that can move quickly and return to precise positions is useful for assembly, tool use, and research into high-frequency control. Linkerbot’s L30 is one of several tendon-driven options available to labs and integrators. The rhythm-game demonstration is a clear visual of speed. It does not replace quantitative benchmarks on grasp success, force control, or durability under continuous industrial use. The hand is sold as a modular end effector that can be mounted on robot arms or humanoid platforms. The video shows the L30 in a controlled entertainment-style task.

Linkerbot is a Chinese company focused on dexterous hands. The L30 is positioned as its highest-performance tendon-driven model. Public specifications and demonstration videos are the main evidence available. The rhythm-game clip has circulated as an example of the hand’s speed. For researchers evaluating end effectors, the relevant questions are repeatability under load, thermal performance during continuous high-speed motion, and integration effort with existing controllers. Those details are not fully answered by a short game video. The demonstration is useful as a qualitative illustration of the platform’s response time.

Linkerbot’s L30 hand is shown playing a rhythm game at high speed. The hand uses a tendon-driven design with approximately 17 active degrees of freedom, fingertip sensing, and joint speeds above 400°/s. The video is a coordination and speed demonstration. The L30 is a commercial dexterous hand for research and industrial use. No independent performance metrics from the game sequence have been published. The clip illustrates the hand’s rapid finger control on a timed task.

By humanoid.press · Editor: Humanoid Press Team · October 19, 2026

L30 Linkerbot

17+

Active DoF

400°/s

Speed

1.4 kg

Weight

Watch Demo L30 Rhythm →

NEURA Details Neuron AI Perception-Language-Action Model

NEURA Robotics has published further explanation of Neuron AI, the multimodal foundation model it introduced on October 14. The model uses a Perception-Language-Action architecture that combines 3D vision, tactile and force signals, audio, robot state, and language into a shared representation updated over time, with a memory layer for object permanence and causality. CEO David Reger and Business Unit Head Dr. Jens Buchner describe the approach in a new video. The company says vision alone is insufficient for many physical tasks; touch and sound confirm contact events that cameras miss. Neuron AI is trained on data from robots in NEURA Gyms, simulation, and human demonstrations in sensor-equipped suits. It runs across five NEURA robot embodiments and their grippers and hands. Customers can fine-tune skills via Neuraverse and deploy them to fleets. Inference is local. An 11-minute autonomous laundry sequence on the 4NE1 humanoid is scheduled for NVIDIA GTC Berlin.

The memory component is presented as necessary for physical action. A robot must remember that an object still exists after it leaves the camera frame and must track the consequences of its own prior actions. NEURA says the architecture was adapted specifically for these memory tasks. Training draws on three pillars: real robot data collected in the company’s gyms and environments, simulation that varies scenarios, and human motion captured with sensor suits. After base training across the five embodiments, the model is fine-tuned on a specific robot for individual skills. Time-to-Skill is the metric the company uses to measure how quickly a new task can be learned and performed reliably. No independent benchmarks comparing Neuron AI to other foundation models have been released. The laundry demonstration will be the first extended public test on hardware.

Multisensory fusion and memory address known limitations of vision-only models in unstructured settings. Tasks such as confirming a connector is seated, judging a weld seam by sound, or continuing a sequence after an object is occluded require information beyond the current camera frame. NEURA’s claim is that Neuron AI incorporates those signals by design. The evidence so far is the architectural description, the training sources, and the scheduled demonstration. The video with Reger and Buchner walks through the same points. Whether the model generalizes to customer environments and reduces time-to-skill as claimed will be tested when customers begin fine-tuning and when the laundry sequence runs in public. Until then, the model is a company system with a clear technical thesis.

NEURA has positioned Neuron AI as the intelligence layer across its robot family, including multiple generations of the 4NE1 humanoid and other bimanual platforms. The Neuraverse platform is the delivery mechanism for fine-tuned skills. The company has said it will scale NEURA Gyms globally to increase data and reduce training time for new skills. The October 14 announcement and the subsequent explanatory video are the public materials. The model is developed and trained in Metzingen, Germany. For potential customers, the open questions are the data requirements for fine-tuning, the reliability of transferred skills, and the cost of local inference on the robot. Those details are not fully specified in the current materials. The laundry task is the next scheduled public data point.

NEURA Robotics is providing additional detail on Neuron AI, its Perception-Language-Action foundation model. The model fuses 3D vision, touch, audio, robot state, and language with a memory layer, and runs across five embodiments. It is trained on robot, simulation, and human demonstration data. Customers fine-tune via Neuraverse. An 11-minute autonomous laundry sequence is planned for NVIDIA GTC Berlin. The new video explains the architecture and training approach. No independent performance numbers have been published. The model remains a company announcement with a scheduled hardware demonstration.

By humanoid.press · Editor: Humanoid Press Team · October 19, 2026

Neuron AI PLA

PLA

Architecture

5

Bodies

Local

Inference

DEEP Robotics Tests DR02 as Traffic Support in Hangzhou

The company’s humanoid is being trialled on streets in the West Lake District for navigation, obstacle avoidance, violation reminders, and multilingual pedestrian interaction.

By humanoid.press · Editor: Humanoid Press Team · October 18, 2026

DR02 Hangzhou

Video credit: DEEP Robotics ▶ Via YouTube

DEEP Robotics has been testing its DR02 humanoid robot on the streets of Hangzhou’s West Lake (Xihu) District. The company describes the work as real-world testing across urban scenarios, including autonomous navigation, dynamic obstacle avoidance, identification of traffic violations, voice and gesture alerts, and interaction with pedestrians in multiple languages. Early October trials took place at busy intersections such as Zheda Road and West Wenyi Road in Lingyin Subdistrict, and at sites including Xixi National Wetland Park. The robot is shown directing traffic, reminding riders to wear helmets or stay behind stop lines, announcing lights, and answering visitor questions. DEEP Robotics presents the tests as exploring support for smarter urban mobility rather than routine deployment.

The DR02 carries an IP66 rating for dust and water protection. The company used rainy conditions during the National Day holiday period to test outdoor operation. In company materials and video captions, the robot is described as supporting traffic officers with roadside patrol and monitoring, including in heavy rain. It can identify behaviours such as e-bike riders without helmets, overloaded bikes, pedestrians crossing lines, and vehicles stopping beyond lines, then issue spoken or gestured reminders. Enforcement decisions remain with human officers; the robot does not issue tickets or detain people. Multilingual voice interaction is tested for tourist guidance on routes and attractions. The number of robots deployed, exact operating hours, and degree of remote supervision have not been fully detailed in public releases.

Hangzhou has previously deployed traffic-management robots around the West Lake area. A separate squad of units began operations earlier in 2026 and has been used for direction, violation reminders, and public consultation. DEEP Robotics’ DR02 trials are a company field test of its own humanoid platform in similar environments. The video shows the robot navigating intersections, avoiding people and vehicles, and interacting with pedestrians. Company engineers have described the intended role as reducing repetitive advisory work so human staff can focus on emergencies and enforcement. Data from the trials is intended to refine the robot for additional public-service applications. No timeline for wider deployment has been announced.

Outdoor humanoid testing in rain, crowds, and moving traffic is a higher bar than controlled indoor demos. Navigation, obstacle avoidance, and reliable voice interaction under variable lighting and weather are the practical challenges. DEEP Robotics’ materials emphasise the IP66 rating and the ability to continue duty in wet conditions. The public evidence is the video and the company description of the early October tests. Independent measurements of uptime, accuracy of violation detection, or success rate of pedestrian interactions have not been published. The trials are framed as learning on real streets rather than a finished product for continuous unsupervised operation.

DEEP Robotics is a Hangzhou-based company known for legged robots used in inspection, rescue, and industrial settings. The DR02 humanoid trials in Xihu District are an expansion into public urban scenarios. The company says the work explores what humanoid robots can do from the lab to the streets. The video documents navigation, alerts, and interaction. The tests do not claim that the robot has replaced traffic officers or that it operates without oversight. They show a company evaluating its platform for potential support roles in traffic management and visitor guidance under real conditions.

DEEP Robotics is testing its DR02 humanoid in Hangzhou’s West Lake District for traffic support and pedestrian interaction. The robot is shown navigating, avoiding obstacles, issuing voice and gesture reminders, and answering questions in multiple languages. Trials occurred in early October at intersections and tourist sites, including in rain. The company presents the work as real-world testing for potential urban mobility support, not as a routine deployment. The video matches the described tasks. Broader operational data and any path to sustained use remain with the company and local authorities.

Watch Video DEEP Robotics DR02 →

Also Today

AheadForm Brings Skyborne Elf-Xuan 2.0 to Shanghai WoW Event

AheadForm is presenting its Elf-Xuan 2.0 humanoid at the World of Warcraft Infinite Party — Dalaran on Earth event in Shanghai. The company says it collaborated with World of Warcraft to give the robot a Skyborne-inspired appearance drawn from the new race in World of Warcraft: Forever. Elf-Xuan 2.0 features expressive facial movements, natural gaze, and coordinated body motion. The event runs October 16–18, 2026 at the National Exhibition and Convention Center (Hall 1.2H). AheadForm describes the robot as blurring the boundary between virtual characters and physical reality. The video shows the Skyborne-styled unit. The appearance is a promotional collaboration for the convention, not a claim of new general-purpose capabilities.

Elf-Xuan 2.0 is AheadForm’s lifelike bionic humanoid platform. The company has previously shown versions focused on facial expression and upper-body movement. Founder Yuhang Hu has described the work as bringing characters from games and virtual worlds into physical form for interaction and entertainment. The Shanghai event is a Blizzard and NetEase production marking 21 years of World of Warcraft in China, with a life-sized Dalaran recreation. AheadForm’s presence is a brand and character collaboration at that event. The robot is not described as performing autonomous tasks beyond the expressive demonstration. Specs, degrees of freedom, and price for Elf-Xuan 2.0 have been reported in secondary sources in broad ranges; the company materials for this event focus on the visual and interactive presentation.

Bionic humanoids designed for expressive interaction are a distinct category from industrial or warehouse platforms. AheadForm’s approach emphasises realistic faces, gaze, and motion so the robot can occupy entertainment and fan-event roles. The Skyborne collaboration places that capability inside a major game-convention setting. Attendees can meet the robot in person during the three-day event. The video is the company’s promotional clip for the appearance. No independent technical evaluation of the robot’s autonomy, reliability, or interaction quality at the event has been published. The demonstration is the physical embodiment of a virtual character for the duration of the show.

AheadForm is a Shanghai-based company focused on realistic humanoid heads and bodies for emotional expression and interaction. Its Elf-Xuan series has been shown at previous events. The World of Warcraft tie-in is a marketing collaboration that places the robot in a high-visibility fan setting. The company states that characters from games and films no longer have to exist only behind screens. The practical outcome is a styled humanoid present at the Shanghai convention. Whether the same platform moves into other commercial entertainment or interactive roles depends on further development and customer interest beyond this event.

AheadForm is displaying Elf-Xuan 2.0 with a Skyborne-inspired appearance at the World of Warcraft Infinite Party in Shanghai from October 16–18, 2026. The robot features expressive facial movements and coordinated motion. The collaboration is promotional for the event. The video shows the unit. It is a character embodiment demonstration rather than a new technical capability claim. The event is the public venue; the robot is the physical presence AheadForm is bringing to it.

By humanoid.press · Editor: Humanoid Press Team · October 18, 2026

Elf-Xuan WoW

2.0

Elf-Xuan

Oct 16–18

Shanghai

Skyborne

Theme

Learn More AheadForm →

Kinova Shows Dual Gen3 Arms for Coordinated Manipulation

Kinova has released a demonstration of two Gen3 robotic arms performing coordinated manipulation. The Gen3 is a lightweight, modular arm platform developed for research and professional applications. It supports grasping, vision-based manipulation, assembly, and robotics development work. Dual-arm setups allow researchers to explore coordinated movements and tasks that require two arms. The video shows the arms working together on a manipulation sequence. Kinova, based in Quebec, Canada, designs the Gen3 series for labs and industrial settings where portability, safety around people, and open software interfaces matter. The Gen3 is available in 6-DoF and 7-DoF configurations with options for integrated vision and various end effectors.

The Gen3 platform is widely used in research because of its ROS support, high-level and low-level control, and modular design. Dual-arm configurations appear in academic and industrial projects for tasks such as assembly, teleoperation, and collaborative manipulation. One documented use is an augmented-reality teleoperation system in which two Gen3 arms are controlled remotely inside a containment cell. The arms’ compact size and 7 degrees of freedom help with collision avoidance in constrained spaces. Kinova’s Kortex software and open APIs support custom development. The video is a company demonstration of dual-arm capability rather than a new product announcement. It illustrates coordinated motion that single-arm systems cannot perform without additional fixturing or sequencing.

Lightweight collaborative arms are common in research labs because they can be moved between experiments and operated near people with lower risk than large industrial robots. Dual-arm setups add the ability to hold one object while manipulating another, or to perform bimanual tasks such as folding, assembly, or tool use. The Gen3’s modular joints and optional vision modules make it suitable for vision-guided grasping and insertion tasks. Researchers use the platform with MoveIt, Gazebo, and custom controllers. The demonstration shows the arms executing a sequence; it does not claim autonomous high-level planning or long-horizon reliability. Performance depends on the software and sensing the user adds.

Kinova has positioned the Gen3 for education, research, and light industrial applications. Dual-arm work expands the range of experiments possible on the same hardware. Labs that already own single Gen3 arms can add a second unit for coordination studies. The video is a visual example of that configuration in action. No new specifications or software releases are announced. The interest for humanoid and manipulation research is the availability of a documented, commercially supported dual-arm setup that can be used for algorithm development and teleoperation studies. The arms themselves remain the same Gen3 platform; the demonstration is the coordinated use of two of them.

Kinova is showing two Gen3 robotic arms performing a coordinated manipulation task. The Gen3 is a modular research and professional arm used for grasping, vision-based work, and assembly studies. Dual-arm setups support exploration of bimanual tasks. The video is a company demonstration of the configuration. It does not introduce new hardware. Researchers use the platform for teleoperation, assembly, and algorithm development. The demonstration illustrates coordinated motion available on existing Gen3 hardware.

By humanoid.press · Editor: Humanoid Press Team · October 18, 2026

Gen3 Kinova

Dual

Arms

6/7

DoF

Research

Use

Learn More Kinova Gen3 →

Humanoid Daily News — 10/17/2026

Humanoid Daily News - October 17th, 2026

10/17/2026

Robot Completes Full Apple-Picking Loop in a Real Orchard

From detection to basket, perception and manipulation run without a human in the loop.

Read full story →

10/17/2026

Unitree Shows UniUni Swarm Control

Rapid choreography, live status, and safety braking for groups of robots.

Read full story →

10/17/2026

NEURA Launches Neuron AI for Physical Robots

A PLA model that adds touch, sound, and memory to 3D vision and language.

Read full story →

10/17/2026

Codey Offers an Unscripted Fist Bump

Mind Children’s social humanoid is built for connection, not factory work.

Read full story →

10/17/2026

Boston Dynamics Shows How Atlas Is Built

In-house machining and same-roof assembly aim for faster iteration to walking hardware.

Read full story →

Humanoid Press — Daily Brief · October 16, 2026

Agility Selects Nyobolt for Digit 5 Fast Charging

Digit 5 is planned to run 90 minutes and recharge in 9 minutes, supporting more than 20 hours of work in a day through short charge cycles.

By humanoid.press · Editor: Humanoid Press Team · October 16, 2026

Digit 5 Nyobolt

Video credit: Agility Robotics / Nyobolt ▶ Via YouTube

Agility Robotics has selected Nyobolt as the ultra-fast charging battery technology partner for its Digit 5 humanoid. The companies announced on October 15 that the system is designed to deliver 90 minutes of runtime and recharge in 9 minutes, for a 10:1 run-to-charge ratio. That is an improvement from the 2:1 ratio Agility has attributed to Digit 4. Agility says the resulting duty cycle can support more than 20 hours of productive work in a 24-hour day, enough to cover more than two full shifts. Chief Business Officer Daniel Diez said Digit’s uptime is the business case for customers. Early-access units with the Nyobolt system are targeted for warehouse operators in the first half of 2027, with broader availability expected over the following six to nine months.

Nyobolt, based in Cambridge, UK, develops high-power battery technology for demanding industrial conditions. Co-founder and CEO Sai Shivareddy said the packs were chosen because Digit 5 needs more power from a smaller, lighter pack in environments where reliability is not optional. Agility has described Digit 5 as its first humanoid designed for cooperatively safe work at scale near people. The battery upgrade is one of several changes on the new model. Pack capacity, charging power, and quantified cycle-life data were not included in the announcement. The performance figures are company targets for the system, not measured results from deployed Digit 5 units.

The partnership addresses a practical constraint in humanoid fleets: downtime for charging. A 10:1 ratio leaves a high share of the day available for work if the robot can dock, charge, and return without long interruptions. Agility has existing commercial deployments of earlier Digit models at sites including Schaeffler, GXO, and Toyota facilities. Digit 5 is intended to expand that work into closer proximity to people and longer daily operation. The Nyobolt system is framed as one of the pieces that makes the higher utilization target reachable. Whether the 9-minute charge holds under real warehouse vibration, temperature, and duty cycles will be tested in the planned pilots.

Agility is completing its business combination with Churchill Capital Corp XI. The SEC has declared the S-4 registration statement effective. The deal values Agility at a $2.5 billion pre-money equity value and is expected to deliver more than $620 million in gross proceeds. The combined company is expected to trade on Nasdaq under the ticker AGLT by the end of the fourth quarter of 2026, subject to remaining conditions including a shareholder vote. The battery announcement arrives as Agility moves toward that public listing. Digit 5 and its charging performance will be part of the commercial story the company presents to public-market investors.

Humanoid fleets in warehouses and manufacturing need high utilization to improve economics. A robot that spends a large fraction of the day charging reduces the effective capacity of the fleet. Agility’s claim is that the Nyobolt system changes that ratio enough to support two-shift coverage from a single robot. The 90-minute runtime and 9-minute charge are the stated targets. Early-access pilots in the first half of 2027 will be the first real-world check of those numbers under continuous industrial use. The partnership is a supplier selection for Digit 5’s power system, not a claim that the full performance has already been demonstrated in customer facilities.

Agility Robotics has named Nyobolt as the battery technology partner for Digit 5’s planned 90-minute runtime and 9-minute recharge. The company says the resulting 10:1 ratio can support more than 20 hours of work in a day. Early access is targeted for the first half of 2027. The figures are company targets. The announcement coincides with Agility’s planned Nasdaq listing as AGLT following its business combination with Churchill Capital Corp XI. The video shows the charging approach; the operational test will be the pilots.

Read Announcement Agility Nyobolt →

Also Today

Hyundai, Qualcomm Join Korea AI and Robot Association

The Korea AI and Robot Industry Association announced on October 15, 2026 that Hyundai Motor, Qualcomm Technologies, and STATS ChipPAC Korea have joined the group. Hyundai Motor joins as a vice-chairman company. Qualcomm joins as a board member company. STATS ChipPAC Korea joins as a regular member. The association said it will use the companies’ expertise and networks to support policy development, proof-of-concept projects, international market expansion, and cross-industry collaboration. Standing vice chairman Bae Seok-tae said the new members strengthen the association’s ability to help South Korea lead in the global AI and robot market. The announcement comes ahead of RoboWorld 2026, scheduled for November 4–7 in South Korea.

Hyundai Motor has been a general member of the association since 2009. The elevation to vice-chairman status gives it a larger role in the association’s direction. Hyundai Motor Group has expanded its robotics footprint through the acquisition of Boston Dynamics and investments in AI and autonomous systems. Qualcomm’s participation brings semiconductor and edge-compute expertise relevant to robot platforms. STATS ChipPAC Korea is a semiconductor packaging firm; its membership adds advanced packaging capability to the association’s network. The association frames the additions as a way to break down barriers among automotive, semiconductor, AI, and robotics sectors.

The Korea AI and Robot Industry Association works on policy recommendations, field demonstrations, overseas market support, and cross-industry models. Adding large automotive and semiconductor companies is intended to broaden the industrial base available for those programs. Hyundai’s existing membership and the group’s broader robotics activities, including Boston Dynamics, give the association a direct link to a major Korean industrial player already investing in humanoids and physical AI. Qualcomm and STATS ChipPAC bring component and packaging perspectives that are relevant to the hardware side of robot development and production. The practical effect will depend on how the association uses the new members’ networks in specific projects.

South Korea has been expanding activity in physical AI and humanoid robotics, including regulatory sandbox pilots for factory deployments and investments by major groups. The association’s membership expansion is one institutional step in that broader activity. RoboWorld 2026 next month will be an early public venue where the new members’ involvement may become visible through exhibits, sessions, or joint announcements. The association’s website is korearobot.or.kr. The October 15 announcement is a membership update, not a product or technology launch. It signals continued consolidation of industrial participants around Korea’s AI and robotics policy and project framework.

Hyundai Motor, Qualcomm, and STATS ChipPAC Korea have joined the Korea AI and Robot Industry Association at different levels. Hyundai is a vice-chairman company, Qualcomm is a board member, and STATS ChipPAC Korea is a regular member. The association plans to draw on their expertise for policy work, demonstrations, and market expansion. The move comes ahead of RoboWorld 2026 in November. It is an industry-association membership announcement that reflects ongoing alignment between large Korean and international companies and the country’s AI and robotics sector. No specific joint projects were detailed in the announcement.

By humanoid.press · Editor: Humanoid Press Team · October 16, 2026

KAR Hyundai

Vice

Hyundai

Board

Qualcomm

Nov 4–7

RoboWorld

Visit Site KAR →

Gemini 4 Argon Builds Cornhole Skill for Apollo from Video

Google DeepMind researchers used Gemini 4 Argon in an internal test to watch a video of a human playing cornhole and generate the simulation and control needed for Apptronik’s Apollo 2 humanoid to sink seven consecutive shots on real hardware. The model mapped the human arm swing and release onto the robot’s joints, used the robot’s cameras to locate the board, built a MuJoCo physics simulation of the body and board, practiced the throw in simulation, and then transferred the result to the physical robot. On the real robot it adjusted release points based on where bags landed until seven shots went in. Google calls the workflow Video-to-Sim-to-Real and says it compresses weeks of manual modeling into a few hours. The demonstration was announced with Gemini 4 Argon on October 13.

Gemini 4 Argon is Google’s new frontier model, made available to developers and enterprises. In the cornhole test it was instructed to learn a physical skill outside its prior training distribution by watching ordinary video. The company reports that Argon scored 87 percent on its Embodied Reasoning benchmark, which tests understanding of physical and spatial problems in three-dimensional environments, and that it outperformed other frontier models on that test. The cornhole result is an internal demonstration of one throwing task. Google has not published details on board distance, total number of throws before the seven consecutive successes, or how the result generalizes to other objects, viewpoints, or contact dynamics. The video shows the sequence from human footage through simulation to the successful throws.

Apptronik’s Apollo is a humanoid platform that has been used in pilots with companies including Mercedes-Benz, GXO, and Jabil. The cornhole test is not a commercial deployment; it is a research demonstration of a training pipeline. The claimed advantage is the reduction in manual work required to build a usable simulation from video of a human performing a task. Once the simulation exists, the model can practice and then transfer to hardware, with further refinement from real-world feedback. Google has described similar pipelines on other tasks, including raising a robot’s success rate on a different manipulation task from 56 percent to 75 percent. Those figures are company-reported. Independent replication of the cornhole result has not been published.

Video-to-sim-to-real is one proposed way to expand the amount of training data available for physical AI. Ordinary videos of people performing tasks become potential sources of simulation and control targets. The practical limits are the accuracy of the inferred physics, the safety of transferring untested motions to hardware, and the need for real-world feedback when the simulation does not perfectly match the robot or the environment. The cornhole demo shows that, for this particular throwing task under controlled conditions, the pipeline produced a working result in hours rather than weeks of manual setup. It does not establish that the same approach will work for arbitrary industrial tasks or without human supervision during the real-world phase.

Google DeepMind has shown Gemini 4 Argon watching a video of cornhole, building a MuJoCo simulation, and enabling Apptronik’s Apollo 2 to sink seven consecutive shots on real hardware. The company says the Video-to-Sim-to-Real workflow reduces weeks of manual modeling to a few hours. The result is an internal demonstration of one skill. Argon’s 87 percent score on an embodied reasoning benchmark is a separate measure of the model’s performance on spatial and physical questions. The video documents the cornhole sequence; broader generalization and safety bounds remain open questions for future tests.

By humanoid.press · Editor: Humanoid Press Team · October 16, 2026

Gemini Apollo

7

Shots

MuJoCo

Sim

Internal

Test

Learn More Gemini 4 Argon →

Humanoid Press — Daily Brief · October 15, 2026

Menlo Opens the SDK for Asimov 1

The company is open-sourcing the Python SDK and the onboard Asimov Edge software so developers can write applications that run on the open-source humanoid.

By humanoid.press · Editor: Humanoid Press Team · October 15, 2026

Asimov SDK

Video credit: Menlo Research / Asimov ▶ Via YouTube

Menlo Research has open-sourced the Menlo SDK and Asimov Edge software for its Asimov 1 humanoid. The Python SDK connects application code to the robot’s movement, joint state, camera, and audio. Asimov Edge is the software that runs on the robot itself: it checks incoming commands and arbitrates control between an application, a gamepad, and other controllers. Developers can build on existing interfaces, extend the SDK, and use coding agents that understand it. Possible uses include replaying a motion, building a VR control interface, connecting a voice application, or running a trained policy on the hardware. Source is on GitHub; documentation is at docs.menlo.ai/sdk.

Asimov 1 is a 1.2-meter, 35-kilogram biped with 25 actuated degrees of freedom. Legs have six degrees of freedom each; arms have five; there is a waist yaw and a two-degree-of-freedom neck. The head carries a monocular camera and stereo microphones; the torso has a speaker and an IMU. Onboard compute is a Raspberry Pi 5 for media and networking plus a Radxa CM5 for motion control. Menlo has already open-sourced the mechanical CAD, electrical design, simulation models, and bill of materials. A DIY kit is listed around $20,000. The robot is positioned as a research and educational platform for whole-body control and locomotion.

The SDK is a client of Asimov Edge, so commands pass through the robot’s safety layer. Cockpit software and a paired gamepad outrank application commands. Connection modes include UDP on the robot’s network, a hybrid mode that adds LiveKit for camera and audio, and a full LiveKit mode. Developers can stand the robot, set velocity, damp, run trajectories, balance, and set joints. State reads return mode, joints, IMU, battery, and alerts. The package is on PyPI as menlo-sdk and requires Python 3.12 or newer.

Menlo Research is the company behind both Asimov and the earlier open-source AI project Jan. The team is distributed across Singapore, Vietnam, and the United States. Hardware is treated as an open reference design that manufacturing partners can build; software and intelligence remain the focus. Asimov 1 launched earlier in 2026. The open-source approach now extends from CAD and BOM to the control software developers write against. That combination—buildable hardware plus a documented, open SDK—is the practical claim.

What the open source does not yet do is turn the robot into a finished application platform. Full-body teleoperation is still listed as in development. The locomotion policy ships in the firmware; direct motor control and telemetry are available through the Edge software or the SDK. A developer who wants to collect data, test a policy, or write a simple control loop can do so. A developer who wants a production-grade unsupervised skill is further along the path. The open SDK lowers the barrier. It does not replace the need for a working robot and a safe workspace.

Menlo has made the control software for Asimov 1 available in the same open way as the hardware designs. Developers can write Python that stands the robot, walks it, reads its sensors, and uses its camera and audio, with the onboard Edge software still checking commands. The GitHub repositories and documentation are public. The DIY kit remains the route to hardware. The announcement is the software side of the open platform. Whether teams ship useful applications will depend on how many robots reach labs and how quickly the examples mature.

View SDK Menlo SDK →

Also Today

NEURA Introduces Neuron AI for Physical Robots

NEURA Robotics introduced Neuron AI on October 14, 2026. The company describes it as a multimodal foundation model for Physical AI and calls it the brain of the physical world. The architecture is Perception-Language-Action, or PLA. It combines 3D vision, touch and force data, audio, robot state, and language, then aligns them over time with a memory component so the robot can act on information that cameras alone do not capture and can retain context when objects leave view. The model is developed and trained in-house in Metzingen, Germany. Founder and CEO David Reger said an intelligent robot needs a brain that brings its senses together and remembers what happened. Independent benchmarks have not been published.

Neuron AI is trained on real-world robot data collected in NEURA’s own gyms, simulation-amplified data, and human demonstrations. The company says one model runs across five of its robot embodiments and their grippers and hands. Customers can fine-tune the model for their own tasks and deploy skills across fleets through the Neuraverse platform. Time-to-Skill is presented as the central measure of progress. Inference is described as local and real-time on the robot. A public demonstration is planned for NVIDIA GTC Berlin, where NEURA’s 4NE1 humanoid is scheduled to complete an eleven-minute laundry sequence fully autonomously. That sequence will be the first extended public test of the model on hardware.

The practical distinction NEURA draws is between models that primarily see the world through cameras and a model that also uses touch, sound, and memory of prior actions. Examples include confirming a lock has opened by its click and resistance, or gripping an object that is no longer in view. Object permanence—remembering that an object still exists after it leaves the camera frame—is presented as a requirement for physical tasks that cameras cannot solve alone. Whether those capabilities hold under real workplace variability is the open question. The company has not released quantitative comparisons. The architecture description and the planned laundry demo are the public evidence.

NEURA is a German robotics company that has positioned itself in the humanoid and cognitive robotics space. Its product family includes the 4NE1 humanoid across generations, wheeled variants, and other bimanual platforms. The Neuron AI launch is framed as the intelligence layer that can sit on top of those bodies and be fine-tuned by customers rather than rebuilt for each new task. For a buyer, the relevant questions are how long fine-tuning takes, what data the customer must supply, and how the model behaves when the task drifts from the training distribution. Those numbers are not in the announcement. The model is the offering; the laundry sequence is the scheduled public check.

Memory and multi-sensory fusion are harder problems than vision alone, and they are also the problems that limit many current robot systems in unstructured settings. A robot that can only act on what is currently in frame will fail when a tool is set down out of view or when contact must be judged by force rather than by sight. NEURA’s claim is that Neuron AI addresses those cases by design. The evidence offered so far is the architectural description, the training sources, and the upcoming demonstration. Until the laundry task or other independent tests are public, the model remains a company announcement with a clear technical thesis and a scheduled first long-horizon public run.

NEURA Robotics has released Neuron AI as its multimodal foundation model for physical robots. The model is designed to fuse 3D vision, touch, audio, robot state, and language with memory, and to run across multiple NEURA embodiments. Customers are expected to fine-tune skills and deploy them via Neuraverse. The first extended public test is an eleven-minute autonomous laundry sequence on the 4NE1 humanoid at NVIDIA GTC Berlin. The announcement is the model and the thesis; the demonstration will be the next data point. No independent performance numbers have been published yet.

By humanoid.press · Editor: Humanoid Press Team · October 15, 2026

Neuron AI NEURA

PLA

Architecture

5

Bodies

Oct 14

Launch

Autowash Uses KUKA Robots for Touchless Car Washes

Autowash is using KUKA industrial robots at its Colorado locations to perform touchless car washes. The system relies on KR CYBERTECH robots, 3D vision, and software that scans each vehicle and generates a customized path for washing and drying. Four cameras create a 3D model of the car; the planning software then produces a unique wash plan for that vehicle’s shape and attachments. The company says the approach delivers a faster, damage-free wash while reducing water and energy use. First commercial robotic bays are operating at the Lakeside location in Wheat Ridge. Autowash plans to roll the system out across more of its roughly 26–27 sites. The application is unusual for industrial arms that more commonly appear in factories.

Autowash was founded by Dennis and Erin Dreeszen and began with a single touchless location around 2013. The chain has grown to more than two dozen sites in Colorado. The founders have described the robotic system as a way to keep the wash completely touchless while improving consistency and resource use. In company materials they report washes in about four minutes versus a prior eight-minute cycle, water use around 20 gallons per car, and reductions in electricity compared with conventional touchless equipment. Those figures are company claims based on their own installations. Independent third-party measurements have not been published.

The technical approach is straightforward once the 3D scan exists. The robots carry custom tool heads with spray nozzles and air knives. Because the path is generated from the actual vehicle model rather than from a fixed template, the system can accommodate different body styles and attachments without a separate program for each. KUKA’s KR CYBERTECH series provides the reach and payload for the bay. The software layer handles the scan-to-path generation. The result is a wash that stays clear of the vehicle’s surfaces while still covering the body. That is the practical advantage over a fixed gantry or brush system designed around an average car.

Industrial robots are rarely the first choice for a consumer car wash. They cost more than conventional equipment, require a controlled bay, and need integration work that most wash operators do not perform in-house. Autowash’s decision to develop the system with KUKA and then deploy it commercially suggests the company sees a path to better margins or higher throughput once the capital is amortized. The chain already operates at a scale where a repeatable automated bay can be rolled out. The Lakeside site is the proof of concept in revenue service. Subsequent sites will test whether the same configuration can be replicated without custom engineering for every location.

The limits are the usual ones for any robotic cell that touches the physical world. A 3D scan can miss a trailer hitch or a damaged body panel. High-pressure water and air still require the nozzles to stay at a safe distance. Weather, customer behavior, and maintenance of the arms and vision system will all affect uptime. Autowash has not published failure rates or a full cost comparison against its previous equipment. What it has published is the architecture—scan, plan, execute with industrial arms—and the early commercial deployment. For the broader robotics industry the interest is the application itself: industrial robots performing a high-volume, consumer-facing service outside a factory.

Autowash is deploying KUKA KR CYBERTECH robots with 3D vision to generate per-vehicle wash and dry paths at its Colorado touchless car washes. The first commercial bays are running at the Lakeside site in Wheat Ridge, with plans to expand across the chain. The company reports faster cycles and lower water and energy use relative to its prior systems. Those are company figures from the installations. The system is an industrial-robot solution for a service that has historically used fixed machinery. The video shows the arms and the scan-to-path process; the business test is whether the same setup can be replicated profitably at the remaining sites.

By humanoid.press · Editor: Humanoid Press Team · October 15, 2026

KUKA Autowash

3D

Scan

~4 min

Cycle

Colorado

Sites

Wiley Publishes Alaoui’s The Humanoid Economy

The book argues that humanoids will scale as infrastructure, not as a consumer app, and that the missing piece is the ecosystem around the robot.

By humanoid.press · Editor: Humanoid Press Team · October 14, 2026

Wiley Book

The Humanoid Economy · Wiley ▶ Via YouTube

Wiley published The Humanoid Economy on October 13, 2026. The full title is The Humanoid Economy: What Physical AI Means for Investment, Industry, and the Future of Work. The author is Modar Alaoui, founder and general partner of ALM Ventures and founder and chair of the Humanoids Summit. The hardcover is 240 pages, ISBN 978-1-394-45364-1. An audiobook narrated by Brandon Pollock is listed. A simplified Chinese edition is licensed to CITIC Press, with a 2027 release. The book is for investors, business leaders, robotics founders, and policymakers who want a map of the system around the robot, not a product review of any one machine.

Alaoui’s argument is that the robot is not the economy. The economy is the set of things that have to exist around it before a buyer can put one on a floor and leave it there. Capital, manufacturing, infrastructure, safety standards, public policy, labor markets, and human acceptance are the chapters. The book treats humanoids as an industrial technology that will scale more like a telephone network than a consumer app. That is the line he has used in the publication materials. It is a claim about pace and infrastructure, not a forecast of unit sales. A demo that walks and a plant that runs a shift are different problems, and the book is about the second one.

Alaoui is an investor and a convener, not a robot maker. ALM Ventures is a Silicon Valley firm focused on humanoids, physical AI, and spatial intelligence. The Humanoids Summit has editions in Silicon Valley, Tokyo, Seoul, and London. He is also founder and chair of the Bay Area Robotics Association and creator of International Humanoids Day. Those roles put him in the rooms where capital, founders, and operators meet. The book draws on that position. It is an insider’s account of how the field is trying to move from demonstrations to commercial scale, written by someone who has been building the rooms rather than the bodies.

Across seven chapters the book traces why the humanoid form fits environments already built for people, how AI maturity and labor shortages lined up, what it takes to engineer and deploy a system, and how venture capital, policy, safety standards, and summits combined to build an ecosystem. Market projections of multi-trillion-dollar scale by mid-century, including figures up to $5 trillion by 2050, appear in the publisher materials. Those are projections, not results. The useful part of the book is the map: what has to be true before a humanoid is infrastructure rather than a stage act. Endorsements from robotics leaders, including Hiroshi Ishiguro and Werner Kraus of Fraunhofer IPA, are listed on the publisher pages.

Publication events followed the release. A book signing and fireside chat in San Francisco on October 13, and another at MassRobotics in Boston on October 15. The site is thehumanoideconomy.com. For a field that still measures itself in clips, a 240-page hardcover from a major publisher is a different artifact. It does not prove that humanoids will reach commercial scale. It sets out the conditions Alaoui thinks have to be met, and the audience he thinks needs to hear them. Investors and policymakers are the readers. A plant manager looking for a price and a failure rate will not find one here.

The Humanoid Economy is out from Wiley. The argument is that the next phase of the field is the ecosystem, not the next demo. Alaoui’s position as an investor and summit chair is the vantage. The book is the map he is offering. Whether the map is right is a question for the next two years of deployments, not for the publication week. The clip is the author on the thesis. The book is the longer version.

DEEP Robotics Puts a Lynx on Patrol at a Grain Store

DEEP Robotics has put a Lynx wheeled-legged quadruped on station at the Fuyang Grain Storage Area in China. The company says the robot is carrying out all-weather inspections and autonomous patrols. The machine combines wheels with four legs. On a flat yard it rolls. On steps or rough ground it uses the legs. That hybrid is the point of the design. A pure wheeled robot stalls at a curb. A pure quadruped is slower on a long flat patrol. The grain yard has both. The film shows the robot moving through the site. It is a deployment claim, not an independent audit of how many hours it has already run.

Lynx is part of a family. The industrial M20 weighs about 33 kg, carries a 15 kg payload, climbs steps up to 80 cm, and is rated IP66. It runs in temperatures from about -20 °C to 55 °C. The sport model is lighter, around 30 kg, with a lower payload and IP54. The company has used the same hybrid idea in other outdoor work: grape transport in Turpan heat, tea-leaf runs in Zhejiang, desert crossings. Grain storage is a different job. The value is a patrol that can go out in weather a person would rather skip, and that can handle a step or a rough path without a second machine.

Grain storage is a safety and maintenance problem as much as a logistics one. Temperature, moisture, pests, and access matter. Traditional wheeled robots need a prepared surface. Conventional quadrupeds can cross rough ground but give up speed on the long aisles of a yard. The wheeled-leg design is meant to keep both. DEEP Robotics has described the Fuyang work as all-weather inspection and autonomous patrol. Those are company claims. The film is the public evidence. A buyer who wants a shift length, a failure rate, or a cost per patrol will not find them in the clip. What the clip shows is a robot that can move through the site in more than one mode.

DEEP Robotics is a Hangzhou company focused on embodied AI and legged systems. The Lynx line is its answer to industrial outdoor work. The M20 was positioned as a mid-sized wheeled-legged robot for complex terrain and hazardous environments. Hot-swap batteries, SLAM, and dual-spectrum cameras for anomaly detection are part of the industrial pitch. Grain storage fits the pattern: a large outdoor site, weather, and a need for repeated patrols rather than a one-off demo. The company has put similar machines into vineyards, orchards, and deserts. A grain yard is the next setting on that list.

The limit is the same as any outdoor robot. A film of a machine moving through a yard does not say how long it runs between charges, how it handles a blocked path, or what happens when the weather is worse than the day of the shoot. DEEP Robotics has not published a shift log or a cost comparison with a human patrol for this site. The useful fact is narrower and real: a wheeled-legged robot is being used for inspection and patrol at a grain storage area, not only in a lab or a desert highlight. That is a practical application. It is not a proof that the category is finished.

DEEP Robotics has deployed a Lynx at the Fuyang Grain Storage Area for all-weather inspection and autonomous patrol, according to the company and the film. The hybrid design is the reason it can do both a long flat run and a step. The clip is the robot on the site. The numbers that would turn this into a buy decision are not in the video. The deployment itself is the news.

By humanoid.press · Editor: Humanoid Press Team · October 14, 2026

Lynx Patrol

Wheel-leg

Design

Fuyang

Site

All-weather

Claim

Watch Patrol Lynx →

Michigan Visits Genix and T-Robotics in Korea

Officials from the Michigan Economic Development Corporation and the Michigan Department of Transportation visited Korean robotics companies Genix Robotics and T-Robotics on October 12–13, 2026. The Korea AI and Robot Industry Association was part of the meetings. The focus, according to the association, was business cooperation, technology exchange, investment opportunities, and expanding Korean robotics companies’ presence in North America, particularly in logistics automation and autonomous mobile robots. The visits are a continuation of Michigan’s longer effort to draw Korean manufacturing and technology investment into the state.

Michigan has been active in Korea. Governor Gretchen Whitmer led an investment mission to South Korea and Taiwan in September 2026, with the MEDC involved. That trip covered shipbuilding, semiconductors, robotics, drones, and future mobility. Korean companies already operating or expanding in Michigan include POEN, BRILS, HL Mechatronics, and Hansae Mobility, which announced a $93 million expansion and 102 jobs during the September trip. The October 12–13 visits to Genix and T-Robotics are a narrower robotics conversation inside that larger pattern. MEDC and MDOT on the same trip suggests both economic development and infrastructure or mobility interests are in the room.

T-Robotics has been building on its existing vacuum robot and AMR work. In late September it was selected as lead on a national project for AI-based mobile humanoid technology for display manufacturing, running through 2030 with government funding. Genix Robotics is a Korean robotics firm that has appeared in industry coverage around regulatory and market moves. The association’s account of the Michigan visits centers on logistics automation and AMRs, not on a signed deal. Meetings of this kind are how a company starts a conversation about a North American presence. They are not a plant announcement.

The practical interest for Michigan is logistics and manufacturing automation. The state has a dense auto supply chain, engineering talent, and a history of Korean investment. Korean robotics companies looking for a North American base have reasons to look there: customers, a manufacturing workforce, and an economic development agency that has already run missions and closed projects. The October visits put Genix and T-Robotics in front of MEDC and MDOT, with the Korea AI and Robot Industry Association facilitating. What comes next is a proposal, a pilot, or nothing. The meetings are the first public step.

The limit is that these are visits, not contracts. The association has described the focus as cooperation, technology exchange, investment, and North American expansion in logistics and AMRs. No investment figure, job count, or site has been announced from these two days. That is normal for an early meeting. The useful fact is that Michigan’s economic development and transportation agencies spent two days with two Korean robotics companies, alongside the industry association. It fits the pattern of the September mission and the existing Korean footprint in the state.

MEDC and MDOT visited Genix Robotics and T-Robotics on October 12–13, with the Korea AI and Robot Industry Association. The stated focus is cooperation, technology exchange, investment, and a larger Korean robotics presence in North America, especially logistics automation and AMRs. The meetings are the news. A deal, if one follows, would be a later story. Source: Korea AI and Robot Industry Association, korearobot.or.kr.

By humanoid.press · Editor: Humanoid Press Team · October 14, 2026

MEDC Korea

Oct 12–13

Visits

2

Companies

AMR

Focus

Watch Visit Michigan–Korea →

Humanoid Press — Daily Brief · October 13, 2026

HIVE Is Training a Unitree to Stack Cans at føtex

A Unitree G1 learns to restock canned goods at a Danish føtex supermarket, as HIVE works toward autonomous overnight shelf replenishment.

By humanoid.press · Editor: Humanoid Press Team · October 13, 2026

HIVE Training

Video credit: HIVE Robots ▶ Via YouTube

HIVE Robots is training a Unitree on a supermarket aisle in Denmark, and the job in this clip is cans. At a føtex near Copenhagen, a G1 is being guided through the motions of picking a tin, lining it with the row, and setting it so the next one still fits. This is a supervised training session, not a finished shift. A can is a specific problem: rigid, round, and meant to sit flush. A shelf of tins fails the eye if one of them sits proud. HIVE is a Danish robotics software and services company based in Kongens Lyngby. It provides the software, remote pilots, and robot service by the hour. føtex, part of Salling Group, is the store. Footage credit: HIVE Robots.

HIVE and føtex announced a multi-year partnership on March 11, 2026, to pilot humanoid robots for overnight shelf replenishment. The company said the first pilot was meant to start before summer. Training sessions with human operators have been under way since June, with operators guiding the movements. Independent overnight replenishment is targeted for early 2027. That remains a target, not a demonstrated capability. The current work is supervised, and the robot is still slower than a person. HIVE markets the service as available. The film shows a guided session. A can that has to sit flush is a useful case, because the failure is visible the moment the robot lets go.

Product variety is the actual job. A night list is not one SKU. It is tins, then a soft pack, then a bottle, then a box that will crush if the grip is wrong. A model that only knows one product will fail on the next. HIVE’s store app lets staff pick which products get filled, each with a photo and a shelf location, and send that list at close. Potential future tasks include expiry checks, empty facings, and items in the wrong place. Those are not in this clip. What is in the clip is the hand learning a can, under supervision, which is the step that has to happen before a morning crew can trust more than one facing.

Height is the other limit, and HIVE has said so in its own posts. The company has called its first retail robot, a Unitree G1 it names Danni, a “reliable colleague” on the night shift, and has said its only weakness is that it cannot reach the top shelves. In the same post, HIVE said a taller Unitree H2, Frankie, is expected to start in November. That is a company post, not a completed deployment. HIVE lists retail robots around 140 and 180 centimeters so the same software can cover a low facing and a high one. The machine in this short is the shorter G1. A can on a bottom shelf is in reach. A can on the top shelf is a different body.

Behind the can is the rest of the stack. Heimdall is HIVE’s teleoperation and data platform. A trained remote robot pilot can take the controls when the robot needs a hand, while operational data can support further training. Whole-body control is HIVE’s, so the robot can bend for a low shelf and reach a high one. The company says its models are trained on GPUs in Denmark and that store data is kept on European infrastructure. Charlotte Blou Sand is co-founder and chief executive. Jakob Sand is co-founder. For now the pilot is overnight, when the store is closed to customers. Expanding into busy shopping hours would introduce additional safety and regulatory questions. The can clip does not change that. It is one product in the training.

HIVE is doing the unglamorous version of a humanoid, one product at a time, with a person still able to take the controls. The partnership is in place, but the work shown remains supervised. Early 2027 is still the horizon they have put on independence. A taller body the company has said is expected in November would change the reach, not the fact that the work is still guided. The commercial shape is unchanged: the store does not buy a fleet. It pays an hourly rate, and HIVE handles the machine. A staffed night, delivered as an hourly service with store data kept on European infrastructure, is the business model. This short is the can, and it is still training.

Watch Aisle HIVE at føtex →

Also Today

MindOn Claims 142 Folds an Hour on Mind-1

MindOn has put a number on the speed claim it made at the end of September. In an October 12 video, nearly an hour long, a dual-arm station folds shirts under Mind-1. The company says the run reached 142 garments, about one every 25 seconds, and labels the footage 1.0x speed and fully autonomous. A person throws the next shirt onto the table after each fold. The folding is what the company calls autonomous. The feed is a human. That is a company claim for a human-fed station, not a laundry that runs itself. The arithmetic is consistent with the number MindOn published. It does not independently confirm the video, the autonomy label, or any ranking.

Mind-1 is a model, not a body. MindOn introduced it on September 30 as physical AI aimed at human-speed task cycle times, and showed it on a dual-arm robot, a wheeled dual-arm, and a humanoid. Parcel handling, shirt folding, and a humanoid collecting laundry were the scenes, labeled the same way: real time, fully autonomous. The company said packaging and parcel sorting came in faster than a person in its own tests, and folding at 25 seconds against a human 15. Those figures are MindOn’s. They are cycle times on chosen tasks, not an independent time study, and they do not include retries or a person standing beside the table.

The hour-long film is the useful follow-up, because a two-minute demo cannot show whether the pace holds. A counter on screen climbs toward the company’s 142. The description talks about sustained operation, throughput, and reliability as the requirements for real work. MindOn has called the result the industry’s fastest robotic clothes folding. That ranking is a company claim. It would need a comparable test: same garments, same start, same fold quality, same help from a person. In the materials reviewed, that side-by-side comparison was not published. What the company has published is a long, labeled clip of one station, with a person feeding each shirt.

Speed is harder than it looks in a highlight reel. Slow teleoperation data teaches a slow policy, because the model learns the tempo of the demonstrations as well as the motions. At an end-effector speed of 3 m/s, 10 milliseconds of latency corresponds to about 3 cm of movement—enough to affect a precise grasp. Consecutive guesses that disagree make the arm correct itself, which at speed becomes a shake. MindOn says Mind-1 was trained on people working at natural speed, that inference latency dropped, and that a whole-body controller holds the trajectory. Those are design claims on the company blog. The film is the public check the company is offering: does the fold keep happening for an hour, at a counted pace, on the shirts someone keeps throwing onto the table.

A folding station is a narrow slice of laundry. Someone still has to bring the shirt, judge the fold, and move the pile. Other groups are working the handoffs around the fold, wash to dry to shelf, which is a different problem from cycle time at the table. MindOn’s number is the fold itself. A humanoid collecting laundry in a room, shown in the September film, is a different job from two arms at a green table. The model is meant to travel between those bodies. The October video tests it on one of them, for long enough to count, and leaves the rest of the workflow to the person standing next to the station.

Mind-1 is MindOn’s bid to make a robot finish work at a useful pace, not just finish it. The October 12 film is the company’s claim of 142 folds in an hour on a dual-arm station, with a person feeding each shirt. The autonomy label covers the fold, as MindOn frames it. The throughput is the company’s number. A shop that wants the number should ask what happens when the shirts change, when the fold quality is checked, and when nobody is standing at the table. Completing a task was the old bar. Doing it for an hour, at a counted pace, is the new one. This film is that count, and it is still a claim.

By humanoid.press · Editor: Humanoid Press Team · October 13, 2026

Mind-1 Folding

142

Company claim

~25 s

Per fold

Oct 12

Film

Watch Hour Mind-1 →

Microducks Sync Over Bluetooth and Start a Choir

Put two Microducks in a room and they start to sing. The small biped from Pollen Robotics, the Bordeaux team that joined Hugging Face in April 2025, syncs with its neighbors over Bluetooth and joins a piece that is being made as it goes. The company’s clip says there is no score and no conductor, and that two, three, or four ducks are enough for a choir to break out. That is the product working as designed: an affordable robot with open-source software that does something visible the moment a second one walks in. Pollen’s published open-source offering covers its software and training stack. The product materials reviewed do not list complete hardware design files. The film is the choir.

Microduck is 25 centimeters tall and weighs under 800 grams. It has 15 motors, an articulated beak for grasping objects, a front camera, LiDAR, and two inertial measurement units. Those figures are from Pollen’s press kit. Pollen Robotics says it can walk, sit, crouch, roller-skate with the appropriate attachments, and recover from many common falls. Each unit generates a voice the first time it wakes and keeps that voice. It does not talk. It quacks and sings. Pre-order price is $399 before taxes and shipping, with first deliveries targeted before Christmas 2026. Seven trained moves ship in the box. The rest is for the owner to add.

The choir is the interesting bit of the software. Ducks find each other over Bluetooth, lock to a shared beat, and take parts. In the open code, the lowest duck takes the bass, a roster is shared, and the piece is chosen for that performance, then another after a breath. The marketing line is no score and no conductor. The mechanism is a beat beacon and a seating rule, which is how a small group stays in time without a person waving a hand. Mouth and head move with the part, so the choir is visible as well as audible. Two ducks are enough. A third joins what it finds. That is the demo, and it is also a small lesson in multi-robot timing.

This is Hugging Face’s second consumer robot, after Reachy Mini, the desktop companion that stays put. Microduck is the one that moves. Pollen’s point is physical AI you can put on a table and retrain: reinforcement learning in simulation, then on the duck, with the open software stack. Races, a kickabout, and a choir are the multi-robot demos, a way to try group behavior without a lab full of hardware. A $399 biped will not stock a shelf or fold a shirt. It will let a student see what a policy looks like when it falls and gets back up. That is a different product from a humanoid, and it should be read as one.

The limit is the size, and it is the point. A duck that sings with its neighbor is a demonstration of sync, identity, and a shared clock, packed into a body light enough to post. It is not a humanoid, and it should not be read as a step on the way to one. The open stack is the serious part: a public software runtime, a public training path, and a robot priced so a classroom can own two. The choir is how that stack shows itself without a slide. A buyer who wants a night aisle or a folding station is in the wrong room. A buyer who wants an affordable robot with open-source software is in the right one.

Microducks sing together because they can hear each other over Bluetooth and agree on a beat. Pollen and Hugging Face are selling that as a $399 robot with open-source software, first deliveries targeted before Christmas, with its own voice and a beak. The clip is the choir. The product is a small body you can retrain. There is no score on the film. There is a beacon in the code. Both are true. Open source, as published, covers the software and training stack. It is a duck that starts a song when another duck walks in.

By humanoid.press · Editor: Humanoid Press Team · October 13, 2026

Microduck Open source

$399

Pre-order

25 cm

Tall

15

Motors

Watch Choir Microduck →

Humanoid Press — Daily Brief · October 12, 2026

HIVE Is Teaching a Unitree to Stock Pasta at føtex

A Danish robotics company is testing a Unitree G1 for overnight shelf replenishment at a føtex near Copenhagen. Operators still guide its movements.

By humanoid.press · Editor: Humanoid Press Team · October 12, 2026

HIVE Retail

HIVE Robots Unitree G1 restocking at føtex

HIVE Robots · føtex shelf trial ▶ Via YouTube

HIVE Robots is putting a humanoid on a supermarket aisle in Denmark, and the job is the dull one. At a føtex store near Copenhagen, a Unitree G1 is being trained to restock shelves after the doors close, including pasta in the clip on this page. The small motions are the hard ones: find the pack, get a grip, place it so the next one still fits. HIVE is a Danish robotics software and services company based in Kongens Lyngby. It provides the software, remote pilots, and robot service by the hour. føtex, part of Salling Group, is providing a real supermarket environment for testing the technology outside a laboratory.

HIVE and føtex announced their multi-year partnership on March 11, 2026. Training began in June, with human operators guiding the robot as it learns to handle different products and shelf configurations. The stated ambition is independent overnight shelf replenishment by early 2027. That remains a target, not a demonstrated capability. The current work is supervised, and the robot is still slower than a person. The film is that training session, not a shift a store can already book. HIVE markets the service as available. The footage shows a supervised trial.

The labor math is why a grocer would bother with a robot that is still learning to hold a box. føtex has said staff spend 40 to 50 percent of their time filling shelves, which is the repetitive half of a retail shift and the half customers never see. HIVE’s offer is to take the night work so the morning crew walks into a stocked aisle and spends the open hours on customers. The commercial shape is robot-as-a-service: the store does not buy a fleet, a maintenance bench, or a robotics team. It pays an hourly rate, and HIVE handles the machine. A store app lets staff select products for replenishment, each with a photo and shelf location, and send the task list for the night shift. Potential future tasks include checking expiry dates, identifying empty shelf spaces, and spotting misplaced products.

Behind the shelf work is Heimdall, HIVE’s teleoperation and data platform. A trained remote robot pilot can take the controls when the robot needs assistance, while operational data can support further training. HIVE also develops whole-body control software to help its robots bend toward lower shelves and reach higher ones. The company says its models are trained on GPUs in Denmark and that store data is kept on European infrastructure. HIVE lists retail robots around 140 and 180 centimeters tall. The machine shown in the clip is a Unitree G1.

Charlotte Blou Sand is co-founder and chief executive. Jakob Sand is co-founder. In June, Jakob Welner, formerly on the Atlas controls team at Boston Dynamics, joined as chief robotics architect, which is the hire that tells you the shelf work is being treated as a controls problem and not a demo. HIVE has also partnered with Danfoss on five-finger grasping for factory work, and it showed a restocking demo on a stand at NRF in Paris. For now, the føtex pilot focuses on overnight work when the store is closed to customers. Expanding into busy shopping hours would introduce additional safety and regulatory questions. HIVE’s current approach relies on human teleoperators who can step in when the robot needs a hand.

HIVE is doing the unglamorous version of a humanoid: a Unitree body, a Danish grocer, a night aisle, and a person still able to take the controls while the model learns the shelf. The partnership with føtex is real, and early 2027 is the date they have put on independence. The clip is that training. A staffed night shift, delivered as an hourly service with store data kept on European infrastructure, is the business model HIVE is developing around the technology.

Watch Aisle HIVE at føtex →

Also Today

CES 2027 Is a Week of Stages, Not One Floor

CES 2027 is not a single hall that opens on a Wednesday. The week starts on Monday, January 4, with media day at Mandalay Bay and CES Unveiled, the preview where companies put a product in front of press two days before the floor opens to the industry. Tuesday, January 5, is the second media day. Conference programming starts that day, and C Space, the media and entertainment campus at ARIA, opens with it. The exhibit floor itself runs Wednesday, January 6, through Saturday, January 9, across the Las Vegas Convention Center, Venetian Expo, and a set of official venues that CES counts as three campuses: LVCC, the Venetian campus, and C Space. A badge that only thinks about the convention center is a badge that will miss half the week.

Hours on the floor are not the same every day, which is the detail that wrecks a late booking. Wednesday opens at 10 a.m. and closes at 6 p.m. Thursday and Friday run 9 a.m. to 6 p.m. Saturday closes at 4 p.m. A robotics team that lands on Saturday morning has already missed the keynote morning, the Unveiled briefings, and most of the meetings that get booked off the Wednesday aisle while people are still fresh. The useful map is the week, not the badge: Unveiled on the 4th, conference on the 5th, floor and stages from the 6th. The film on this page is a highlight from the last CES floor, robots beside other hardware. It is a picture of the mix. It is not the 2027 calendar.

The stages already named are mobility and computing, not a humanoid keynote, and that is worth knowing before anyone promises a robot moment. On January 6, the State of the Industry address is set for 8:30 a.m. in the Palazzo Ballroom at the Venetian. Stellantis chief executive Antonio Filosa follows, on connected vehicles and software across the group’s brands, with the talk tied to a West Hall exhibit so the audience can walk from the ballroom to the car. Waymo co-chief executive Tekedra Mawakana is listed the same day at 1 p.m. in that ballroom. Lenovo is taking its Tech World keynote to Sphere. None of these is a robot launch. They are the rooms the rest of the industry will be walking out of, which is where a humanoid meeting either gets booked or does not.

Foundry is the program closest to this beat. After a two-day debut in 2026, the AI and quantum destination returns to Fontainebleau for three days, January 6–8, with talks, demos, and the kind of hallway that software and hardware people actually use when they are not performing for a booth camera. Eureka Park remains the startup zone, the place a small humanoid team can still be found without a West Hall budget. CES Innovation Awards honorees are due in November, with embargoed products shown from January 4, which is another reason the media day matters more than the Saturday close. A short film of the last floor is a fair picture of how those zones feel in person. It does not tell you which room to be in at 8:30 on the 6th.

For a humanoid company, the week splits cleanly, and the split is the story. Unveiled is where a new hand or a new body gets a first write-up, before the floor noise starts. The exhibit floor is where a buyer walks the aisle and asks for a meeting that is not a stage demo with a marked path. Foundry is where the model people are, the ones who will ask what the policy was trained on. C Space is a different crowd, media and entertainment, useful for a bionic face and less useful for a payload chart. A team that only staffs a booth from Wednesday noon will meet the people who already had their briefing on Monday and have moved on. The last January showed how crowded the robot corner can get. The calendar above is how to be in the right room before that happens.

CES 2027 runs January 6–9 on the floor, with the real week opening January 4 at Mandalay Bay. Keynotes already set are Stellantis and Waymo on the 6th, plus Lenovo at Sphere. Foundry fills Fontainebleau for three days. Eureka Park is still the startup hall. The film is last year’s mix of robotics, hardware, and the Physical AI pitch. The program is the reason to book the earlier flight, and to know which campus you are actually walking into.

By humanoid.press · Editor: Humanoid Press Team · October 12, 2026

CES 2027 Program

Jan 4

Media day

Jan 6–9

Floor

3 days

Foundry

Watch Highlight CES floor →

AheadForm Elf-Xuan 2.0 at Inclusion 2026

AheadForm’s Elf-Xuan 2.0 Is Built to Be Looked At

AheadForm brought Elf-Xuan 2.0 to Inclusion 2026 in Shanghai, and the point of the machine is the face. The company’s own clip, posted in September, holds on the eyes and the small movements around the mouth long enough that a viewer has to remind themselves it is a robot. AheadForm calls it a bionic humanoid, lifelike design and natural expression set on embodied AI. It is a showpiece, not a factory body, and the film is honest about that. Nobody in the frame is handing it a bin, a tool, or a shift sheet. The product is the moment a person across the room forgets, briefly, what they are looking at.

Inclusion, on the Bund, is an AI conference rather than a robotics fair, which is why a singing android fits the bill better than a palletizing demo would. Reports from the week said Elf-Xuan 2.0 performed a song written by its own system, with gesture and expression timed to the vocal, on a stage rather than in a booth corner. Coverage described synthetic skin and a face driven by many small actuators, the usual route to a microexpression that does not look like a mask sliding across a skull. Those are descriptions of a stage act, written by people who watched it. They are not a spec sheet, and AheadForm has not published degrees of freedom, payload, battery life, or runtime with the clip.

The company behind the face raised about $15.7 million in July 2026, in a round reported by the trade press rather than in a filing a buyer can read. That is enough to keep a small bionic team in parts, silicone, and stage time. It is not a humanoid factory round, and it should not be read as one. AheadForm’s public trail is the look: elves, singers, machines that hold a gaze and are filmed for the gaze. The Inclusion appearance is the clearest version of that bet, a body designed so the first reaction is recognition rather than a count of joints or a question about the hourly rate.

There is a real split in the humanoid market here, and the clip sits on one side of it without apology. Industrial machines are judged on a shift, a hand, a failure rate, and a price a plant manager can defend. Bionic machines are judged on whether a person across a table forgets, for a second, what they are looking at, and whether they stay to watch the next gesture. Both can be serious engineering. They are not the same product, and a buyer who confuses them will be disappointed by both. Elf-Xuan 2.0 is not applying for a night aisle in a Danish grocer or a line in a factory. It is applying for the room where people stay and watch.

The limit is the same as any stage robot, and it is the limit that keeps the rating at zero. A song, a look back, and a smooth turn of the head do not say what the machine does at 7 a.m., whether the face still tracks after an hour under conference lighting, or what it costs to keep the skin looking like that. AheadForm has not claimed a deployment, a wage, a lobby installation, or a second unit doing the same act the next night. The value of the September film is narrower and real: a Chinese bionic maker put a convincing face on a conference stage, and the audience treated it as a performer. That is the product working as designed.

Elf-Xuan 2.0 is AheadForm’s answer to the uncanny valley, shown at Inclusion 2026 in Shanghai and filmed so the eyes do the work. The clip is the company letting the face do the talking, which is a choice, not an omission. It is a bionic showpiece, funded like a startup, not an industrial humanoid, and it should be read as one. The aisle in Denmark and the line in a factory are different businesses. This one is the face.

By humanoid.press · Editor: Humanoid Press Team · October 12, 2026

Elf-Xuan Bionic

2.0

Model

Shanghai

Inclusion

Face

The product

Watch Face Elf-Xuan →

Hyundai, Qualcomm & Stats ChipPAC Join Korea AI-Robotics Association as New Members

2026-10-16 02:00 The Korea AI & Robotics Industry Association (Chairman Oh Joon-ho, Samsung Electronics) recently held its board meeting and appointed Hyundai Motor Company (vice-chairman), Qualcomm Technologies Inc. (director), and Stats ChipPAC Korea (regular member) as new executive and regular members, it was announced on October 15.@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=Playfair+Display:wght@600&display=swap'); :root { --primary: #0a0a0a; --accent: #d4af77; } body { margin: 0; font-family: 'Inter', system_ui, sans-serif; line-height: 1.65; color: #222; background: #f8f5f0; } .article-container { max-width: 980px; margin: 40px auto; background: white; box-shadow: 0 10px 40px rgba(0, 0, 0, 0.08); overflow: hidden; } header { background: linear-gradient(135deg, #1a1a1a, #2d2d2d); color: white; padding: 60px 40px 50px; text-align: center; position: relative; } header::after { content: ''; position: absolute; bottom: 0; left: 0; width: 100%; height: 4px; background: linear-gradient(to right, transparent, var(--accent), transparent); } .headline { font-family: 'Playfair Display', serif; font-size: 2.8rem; line-height: 1.1; font-weight: 600; margin: 0 0 16px; letter-spacing: -0.02em; } .subheadline { font-size: 1.25rem; opacity: 0.9; margin: 0; font-weight: 400; } .meta { margin-top: 30px; display: flex; align-items: center; justify-content: center; gap: 20px; font-size: 0.95rem; opacity: 0.85; } .meta-dot { width: 4px; height: 4px; background: var(--accent); border-radius: 50%; } .article-body { padding: 50px 40px 70px; } .lead { font-size: 1.35rem; font-weight: 500; color: #333; margin-bottom: 32px; border-left: 4px solid var(--accent); padding-left: 20px; } h2 { font-family: 'Playfair Display', serif; font-size: 1.85rem; margin: 50px 0 18px; color: var(--primary); position: relative; } h2::before { content: ''; position: absolute; left: -8px; top: 8px; width: 22px; height: 3px; background: var(--accent); } p { font-size: 1.08rem; margin-bottom: 24px; } .image-container { margin: 40px 0; text-align: center; border-radius: 12px; overflow: hidden; box-shadow: 0 8px 25px rgba(0, 0, 0, 0.12); } .image-container img { width: 100%; display: block; max-height: 420px; object-fit: cover; } .caption { font-size: 0.95rem; font-style: italic; color: #666; margin-top: 12px; text-align: center; } .quote { background: #f8f5f0; border-left: 5px solid var(--accent); padding: 22px 26px; margin: 32px 0; font-size: 1.1rem; font-style: italic; color: #333; } .highlight-box { background: #f8f5f0; padding: 28px 32px; border-radius: 12px; margin: 40px 0; } footer { background: #111; color: #ccc; padding: 40px 40px 60px; font-size: 0.9rem; text-align: center; } .byline { font-weight: 500; color: var(--accent); } HUMANOID PRESS • OCTOBER 16, 2026 Hyundai Motor, Qualcomm, and Stats ChipPAC Korea Join Korea AI & Robotics Industry Association as New Executive and Regular Members Major global companies strengthen AI-robotics collaboration by joining the association’s leadership and membership By Humanoid Press Team Featured: Korea AI & Robotics Industry Association new executive and regular members To accelerate the expansion of the AI and robotics ecosystem — a core pillar of future advanced industries — leading domestic and international companies have united. Featured Photo: Korea AI & Robotics Industry Association welcomes new executive and regular members (Photo credit: Korea AI & Robotics Industry Association) Breaking Down Industry Boundaries The Korea AI & Robotics Industry Association (Chairman Oh Joon-ho, Samsung Electronics) recently held its board meeting and appointed Hyundai Motor Company (vice-chairman), Qualcomm Technologies Inc. (director), and Stats ChipPAC Korea (regular member) as new executive and regular members, it was announced on October 15. With this new membership, Hyundai Motor will participate directly in the association’s operations and key initiatives, Qualcomm will serve as a director driving technology collaboration and policy proposals, and Stats ChipPAC Korea will contribute as a regular member with expertise in semiconductor backend processing. “Hyundai Motor, Qualcomm, and Stats ChipPAC Korea, along with other globally innovative companies representing each industry, have come together at the association,” said Baek Seok-tae, standing deputy chairman. Industry Synergies Expected to Reach New Heights The addition of Hyundai Motor, a global leader in complete vehicles and robotics; Qualcomm, which holds on-device AI and semiconductor technology; and Stats ChipPAC Korea, a global leader in semiconductor backend processing, is expected to create maximum synergy by dismantling barriers between automotive, semiconductor, AI, and robotics, and significantly expanding Korea’s AI-robotics ecosystem. The association wil...

Agility Robotics Picks Nyobolt for Digit 5’s 90-Minute Charge

2026-10-16 02:00 Early-Access Pilots Targeted for First Half of 2027 Early-access units with the Nyobolt system are targeted for warehouse operators in the first half of 2027, with broader availability expected over the following six to nine months. What This Means for Humanoid Fleets The partnership directly addresses a major constraint in humanoid fleets: downtime for charging. A 10:1 ratio leaves a high share of the day available for work if the robot can dock, charge, and return without long interruptions. Agility has existing commercial deployments of earlier Digit models at sites including Schaeffler, GXO, and Toyota facilities. Digit 5 is designed to expand that work into closer proximity to people and longer daily operation.@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=Playfair+Display:wght@600&display=swap'); :root { --primary: #0a0a0a; --accent: #d4af77; } body { margin: 0; font-family: 'Inter', system_ui, sans-serif; line-height: 1.65; color: #222; background: #f8f5f0; } .article-container { max-width: 980px; margin: 40px auto; background: white; box-shadow: 0 10px 40px rgba(0, 0, 0, 0.08); overflow: hidden; } header { background: linear-gradient(135deg, #1a1a1a, #2d2d2d); color: white; padding: 60px 40px 50px; text-align: center; position: relative; } header::after { content: ''; position: absolute; bottom: 0; left: 0; width: 100%; height: 4px; background: linear-gradient(to right, transparent, var(--accent), transparent); } .headline { font-family: 'Playfair Display', serif; font-size: 2.8rem; line-height: 1.1; font-weight: 600; margin: 0 0 16px; letter-spacing: -0.02em; } .subheadline { font-size: 1.25rem; opacity: 0.9; margin: 0; font-weight: 400; } .meta { margin-top: 30px; display: flex; align-items: center; justify-content: center; gap: 20px; font-size: 0.95rem; opacity: 0.85; } .meta-dot { width: 4px; height: 4px; background: var(--accent); border-radius: 50%; } .article-body { padding: 50px 40px 70px; } .lead { font-size: 1.35rem; font-weight: 500; color: #333; margin-bottom: 32px; border-left: 4px solid var(--accent); padding-left: 20px; } h2 { font-family: 'Playfair Display', serif; font-size: 1.85rem; margin: 50px 0 18px; color: var(--primary); position: relative; } h2::before { content: ''; position: absolute; left: -8px; top: 8px; width: 22px; height: 3px; background: var(--accent); } p { font-size: 1.08rem; margin-bottom: 24px; } .image-container { margin: 40px 0; text-align: center; border-radius: 12px; overflow: hidden; box-shadow: 0 8px 25px rgba(0, 0, 0, 0.12); } .image-container img { width: 100%; display: block; max-height: 420px; object-fit: cover; } .caption { font-size: 0.95rem; font-style: italic; color: #666; margin-top: 12px; text-align: center; } .quote { background: #f8f5f0; border-left: 5px solid var(--accent); padding: 22px 26px; margin: 32px 0; font-size: 1.1rem; font-style: italic; color: #333; } .highlight-box { background: #f8f5f0; padding: 28px 32px; border-radius: 12px; margin: 40px 0; } footer { background: #111; color: #ccc; padding: 40px 40px 60px; font-size: 0.9rem; text-align: center; } .byline { font-weight: 500; color: var(--accent); } HUMANOID PRESS • OCTOBER 16, 2026 Agility Robotics Names Nyobolt as Battery Partner for Digit 5 Humanoid 90 minutes of runtime and 9-minute recharge deliver a 10:1 run-to-charge ratio — enough for more than 20 hours of daily work By Humanoid Press Team Featured: Nyobolt ultra-fast charging system for Agility Robotics Digit 5 Agility Robotics has selected Nyobolt as the ultra-fast charging battery technology partner for its Digit 5 humanoid. Featured Photo: Agility Robotics Digit 5 with Nyobolt ultra-fast charging system (Photo credit: Agility Robotics / Nyobolt) 90 Minutes of Runtime, 9 Minutes of Charge The companies announced on October 15 that the system is designed to deliver 90 minutes of runtime and recharge in 9 minutes, for a 10:1 run-to-charge ratio. That is an improvement from the 2:1 ratio Agility has attributed to Digit 4. The resulting duty cycle can support more than 20 hours of productive work in a 24-hour day, enough to cover more than two full shifts. “Digit’s uptime is the business case for customers,” said Agility Chief Business Officer Daniel Diez. Early-Access Pilots Targeted for First Half of 2027 Early-access units with the Nyobolt system are targeted for warehouse operators in the first half of 2027, with broader availability expected over the following six to nine months. What This Means for Humanoid Fleets The partnership directly addresses a major constraint in humanoid fleets: downtime for charging. A 10:1 ratio leaves a high share of the day available for work if the robot can dock, charge, and return without long interruptions. Agility has existing commercial deployments of earlier Digit models at sit...

HIVE Robots is continuing its supermarket robotics trials in Denmark, using a Unitree G1 humanoid

2026-10-13 03:00 Restocking involves more than grasping an object. The robot must approach, establish a reliable grip, position items precisely without disturbing neighbors, and release them cleanly. Rigid cans, soft packaging, tall bottles — each presents unique challenges in perception, manipulation, and placement. Product variety is the central difficulty. A single successful system must adapt across bags, boxes, cans, bottles, and dozens of SKUs. HIVE’s current work remains fully supervised: human operators guide the robot through difficult cases using the company’s teleoperation platform Heimdall. @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=Playfair+Display:wght@600&display=swap'); :root { --primary: #0a0a0a; --accent: #d4af77; } body { margin: 0; font-family: 'Inter', system_ui, sans-serif; line-height: 1.65; color: #222; background: #f8f5f0; } .article-container { max-width: 980px; margin: 40px auto; background: white; box-shadow: 0 10px 40px rgba(0, 0, 0, 0.08); overflow: hidden; } header { background: linear-gradient(135deg, #1a1a1a, #2d2d2d); color: white; padding: 60px 40px 50px; text-align: center; position: relative; } header::after { content: ''; position: absolute; bottom: 0; left: 0; width: 100%; height: 4px; background: linear-gradient(to right, transparent, var(--accent), transparent); } .headline { font-family: 'Playfair Display', serif; font-size: 2.8rem; line-height: 1.1; font-weight: 600; margin: 0 0 16px; letter-spacing: -0.02em; } .subheadline { font-size: 1.25rem; opacity: 0.9; margin: 0; font-weight: 400; } .meta { margin-top: 30px; display: flex; align-items: center; justify-content: center; gap: 20px; font-size: 0.95rem; opacity: 0.85; } .meta-dot { width: 4px; height: 4px; background: var(--accent); border-radius: 50%; } .article-body { padding: 50px 40px 70px; } .lead { font-size: 1.35rem; font-weight: 500; color: #333; margin-bottom: 32px; border-left: 4px solid var(--accent); padding-left: 20px; } h2 { font-family: 'Playfair Display', serif; font-size: 1.85rem; margin: 50px 0 18px; color: var(--primary); position: relative; } h2::before { content: ''; position: absolute; left: -8px; top: 8px; width: 22px; height: 3px; background: var(--accent); } p { font-size: 1.08rem; margin-bottom: 24px; } .image-container { margin: 40px 0; text-align: center; border-radius: 12px; overflow: hidden; box-shadow: 0 8px 25px rgba(0, 0, 0, 0.12); } .image-container img { width: 100%; display: block; max-height: 420px; object-fit: cover; } .caption { font-size: 0.95rem; font-style: italic; color: #666; margin-top: 12px; text-align: center; } .quote { background: #f8f5f0; border-left: 5px solid var(--accent); padding: 22px 26px; margin: 32px 0; font-size: 1.1rem; font-style: italic; color: #333; } .highlight-box { background: #f8f5f0; padding: 28px 32px; border-radius: 12px; margin: 40px 0; } footer { background: #111; color: #ccc; padding: 40px 40px 60px; font-size: 0.9rem; text-align: center; } .byline { font-weight: 500; color: var(--accent); } HUMANOID PRESS • OCTOBER 13, 2026 HIVE Robots Trains Unitree G1 Humanoid to Restock føtex Shelves From pasta to canned goods: practical supervised trials bring everyday retail one step closer to autonomous overnight replenishment in early 2027 By Humanoid Press Team Featured: Unitree G1 in a real Danish supermarket setting A Danish humanoid robotics startup is now training a Unitree G1 robot to perform real-world supermarket shelf restocking — the kind of repeatable, customer-facing task that marks the next phase in the commercialization of general-purpose humanoids. From pasta to canned goods, HIVE Robots is conducting supervised training sessions with a Unitree G1 humanoid (internally dubbed “Danni”) inside an active føtex supermarket. The goal is not flashy demos, but reliable handling of everyday grocery products that everyday shoppers take for granted. Featured Photo: Unitree G1 reaching for a supermarket shelf (Credit: Jeppe B. Nielsen) Supermarket Restocking Is No Longer Just About Picking and Placing Restocking involves more than grasping an object. The robot must approach, establish a reliable grip, position items precisely without disturbing neighbors, and release them cleanly. Rigid cans, soft packaging, tall b...

HIVE Robots Tests Unitree G1 for føtex Shelf Restocking

2026-10-12 06:00 HIVE Robots is testing a humanoid robot for one of retail’s most repetitive tasks: restocking supermarket shelves. At a føtex store near Copenhagen, Denmark, a Unitree G1 is being trained to replenish products after the store closes, including pasta shown in the accompanying footage. The trial focuses on a practical challenge for humanoid robotics. A robot must do more than walk through a store. It needs to reach different shelf heights, handle packaged products, position items accurately, and repeat those movements reliably in a real retail environment. HIVE, a Danish robotics software and services company based in Kongens Lyngby, is developing the software and operating model behind the application. Rather than manufacturing its own humanoid body, the company works with existing robotic platforms and provides the software, remote-pilot support, and operational services needed to deploy them.@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=Playfair+Display:wght@600&display=swap'); :root { --primary: #0a0a0a; --accent: #d4af77; } body { margin: 0; font-family: 'Inter', system_ui, sans-serif; line-height: 1.65; color: #222; background: #f8f5f0; } .article-container { max-width: 980px; margin: 40px auto; background: white; box-shadow: 0 10px 40px rgba(0, 0, 0, 0.08); overflow: hidden; } header { background: linear-gradient(135deg, #1a1a1a, #2d2d2d); color: white; padding: 60px 40px 50px; text-align: center; position: relative; overflow: hidden; } header::after { content: ''; position: absolute; bottom: 0; left: 0; width: 100%; height: 4px; background: linear-gradient(to right, transparent, var(--accent), transparent); } .headline { font-family: 'Playfair Display', serif; font-size: 2.8rem; line-height: 1.1; font-weight: 600; margin: 0 0 16px; letter-spacing: -0.02em; } .subheadline { font-size: 1.25rem; opacity: 0.9; margin: 0; font-weight: 400; } .meta { margin-top: 30px; display: flex; align-items: center; justify-content: center; gap: 20px; font-size: 0.95rem; opacity: 0.85; } .meta-dot { width: 4px; height: 4px; background: var(--accent); border-radius: 50%; } .article-body { padding: 50px 40px 70px; } .lead { font-size: 1.35rem; font-weight: 500; color: #333; margin-bottom: 32px; border-left: 4px solid var(--accent); padding-left: 20px; } h2 { font-family: 'Playfair Display', serif; font-size: 1.85rem; margin: 50px 0 18px; color: var(--primary); position: relative; } h2::before { content: ''; position: absolute; left: -8px; top: 8px; width: 22px; height: 3px; background: var(--accent); } p { font-size: 1.08rem; margin-bottom: 24px; } .image-container { margin: 40px 0; text-align: center; border-radius: 12px; overflow: hidden; box-shadow: 0 8px 25px rgba(0, 0, 0, 0.12); } .image-container img { width: 100%; display: block; max-height: 420px; object-fit: cover; } .caption { font-size: 0.95rem; font-style: italic; color: #666; margin-top: 12px; text-align: center; } .quote { background: #f8f5f0; border-left: 5px solid var(--accent); padding: 22px 26px; margin: 32px 0; font-size: 1.1rem; font-style: italic; color: #333; } .highlight-box { background: #f8f5f0; padding: 28px 32px; border-radius: 12px; margin: 40px 0; } .youtube-embed { margin: 40px 0; text-align: center; border-radius: 12px; overflow: hidden; box-shadow: 0 8px 25px rgba(0, 0, 0, 0.12); } .youtube-embed iframe { width: 100%; max-width: 640px; height: 360px; display: block; } .links { margin: 30px 0; text-align: center; font-size: 1.05rem; } .links a { color: var(--accent); text-decoration: none; } .links a:hover { text-decoration: underline; } footer { background: #111; color: #ccc; padding: 40px 40px 60px; font-size: 0.9rem; text-align: center; } .byline { font-weight: 500; color: var(--accent); } /* ANIMATION OVERLAYS (NEW - subtle floating robot + data flow) */ .robot-float { position: absolute; top: 20px; right: 40px; width: 110px; ...

Humanoid Press – Humanoid Robots News Hub | 2026 Monthly Archive

2026 Monthly Archive

October 2026

Boston Dynamics Unveils New Atlas Hand, Names Rohit Prasad CEO

October 6, 2026

Boston Dynamics revealed a redesigned 13-DoF Atlas hand built for mass manufacturing and modern AI, then named former Amazon AI leader Rohit Prasad as its new CEO to drive physical-AI commercialization.

October 2026

Agility Digit Wins AGCO and GXO Expansions, Adds FORT Safety Partnership

October 2026

Agility signed a commercial deal to deploy Digit at AGCO’s Kansas factory, expanded with GXO into North Carolina and Tennessee sites, and partnered with FORT Robotics on Digit 5 safety architecture.

October 2026

IDC: H1 Humanoid Shipments Near 25,000, Up 432%

October 2026

IDC data put first-half 2026 humanoid shipments at almost 25,000 units, with China dominating volume. Meanwhile a report says Meta has built a prototype humanoid with Chinese startup Fourier.

September 2026

Tesla Optimus Production Ramps to Hundreds/Week

September 2026

Supply-chain reports confirm Tesla ordering parts for ~15,000 Optimus units in 2026, targeting ~1,000 units/week by late year. Production focuses on the dedicated “Optimus Academy” inside Fremont for real-world training data collection; supply-chain hurdles remain significant.

September 2026

Agility Unveils Digit 5 — Safe Cooperative Humanoid

September 15, 2026

Agility unveiled Digit 5, built on over 65,000 hours of commercial operation and new 13-DoF hands. The platform enables true close-proximity human collaboration without barriers; IPO via Churchill Capital Corp XI merger and Fort Robotics MoU for safety infrastructure also announced.

September 2026

Boston Dynamics Opens Atlas Training Center at Hyundai Metaplant

September 23, 2026

Boston Dynamics opened a dedicated Atlas training and testing center at Hyundai’s Georgia Metaplant. The facility prepares production-intent Atlas units for automotive manufacturing roles, emphasizing tool handling and reinforcement-learning policies. All 2026 Atlas output remains committed to Hyundai + Google DeepMind.

August 2026

Figure 03 Hits 1,000 Units; BMW Logistics Sequencing Live

August 25, 2026

BotQ production sustained at one robot per hour. Figure 03 now sorting and sequencing parts at Spartanburg; continuous multi-day package demos draw industry attention.

August 2026

China H1 Shipments Top 19,000 — AgiBot Leads Volume

August 2026

Global humanoid shipments surge 272% YoY. Chinese makers hold ~97% of volume; Unitree IPO and factory deployments accelerate the commercial phase.

August 2026

Tesla Optimus Production Ramp Begins at Fremont

Late July – August 2026

Former Model S/X space converted; low-volume builds start for internal training and data collection. External sales still targeted later.

July 2026

Figure Crosses 1,000-Robot Milestone at BotQ

July 23, 2026

Figure AI becomes the first U.S. humanoid company to surpass 1,000 units produced. Throughput jumps and supply-chain localization continue.

July 2026

Agility Digit: 100k+ Totes & Multi-Customer RaaS

July 2026

Digit fleets expand at GXO and new sites including Toyota Canada. Operating hours and contracted pipeline strengthen commercial leadership claims.

July 2026

Boston Dynamics Electric Atlas Production Underway

July 2026

All 2026 Atlas production committed to Hyundai and Google DeepMind. Heavy-lift and whole-body control demos advance the industrial case.

June 2026

BMW Brings Figure 03 Back for Logistics Sequencing

June 2026

After successful Figure 02 body-shop work, Figure 03 takes on more complex parts sorting and sequencing in a live production environment.

June 2026

Catalyst Brands Signs Figure for Warehouse Deployment

June 2026

JCPenney parent deploys Figure 03 fleets at Reno logistics center — one of the first major retail-supply-chain humanoid contracts.

June 2026

NVIDIA Cosmos 3 World Foundation Model Launches

Late May – June 2026

Open physical-AI foundation model unifies world generation, visual reasoning and action simulation — core infrastructure for humanoid training pipelines.

May 2026

Figure 02 Completes 11-Month BMW Spartanburg Run

May 2026

1,250+ operating hours, 90,000+ parts handled, contribution to 30,000+ X3 vehicles — the clearest Western commercial proof point to date.

May 2026

Apptronik Apollo Expands with Mercedes & Jabil

May 2026

Manufacturing and logistics pilots deepen; Jabil partnership advances “robots that build robots” manufacturing model.

May 2026

Chinese Volume Leaders Accelerate Factory Pilots

May–June 2026

AgiBot, Unitree, UBTech and Xiaomi push high-volume production and real factory task training (nut installation, panel sorting, logistics).

April 2026

Agility Robotics Multi-Site Digit Expansion

April 2026

Additional logistics and automotive customers come online; RaaS model and operating-hour data continue to differentiate Agility.

April 2026

Tesla Optimus Internal Factory Use Grows

April–May 2026

Limited Optimus units operate on battery-cell and parts-handling tasks inside Fremont and Texas while production lines are prepared.

April 2026

Physical AI Simulation & Foundation Model Stack Matures

Spring 2026

Omniverse, Cosmos and human-motion foundation models tighten the loop between synthetic data, policy learning and real humanoid control.

March 2026

NVIDIA GTC Focus on Physical AI & Humanoid Stacks

March 2026

World models, simulation pipelines and reference humanoid platforms take center stage as the industry shifts from demos to scalable training data.

March 2026

Boston Dynamics Atlas Electric Enters Production

March 2026

First production units allocated to Hyundai industrial sites and Google DeepMind research; strength and whole-body control emphasized.

March 2026

Humanoid Deployment Trackers Become Industry Standard

Q1 2026

Live trackers, HPS-style leaderboards and side-by-side capability databases emerge as key reference points for operators and investors.

February 2026

Agility Signs Toyota Canada RaaS Agreement

February 2026

Digit expands from pure logistics into automotive manufacturing environments under a formal Robots-as-a-Service contract.

February 2026

Figure & Western Peers Push Commercial Proof Points

February 2026

BMW results, early warehouse deals and production-rate targets set the competitive baseline that Chinese volume players and Tesla will be measured against.

February 2026

Unitree & Domestic Chinese Capacity Build-Out

Early 2026

High shipment numbers and new production capacity announcements reinforce China’s early lead in pure unit volume.

January 2026

Electric Atlas Formally Launches at CES

January 2026

Boston Dynamics presents the production-intent electric Atlas; 2026 output fully committed to Hyundai and research partners.

January 2026

Tesla Optimus Lines Prepared at Fremont

January 2026

Conversion of former vehicle production space begins; internal deployment and data-collection focus remains the near-term priority.

January 2026

Early Commercial Humanoid Deployments Scale

January 2026

Agility, Figure, Apptronik and Chinese platforms move from pilots into multi-site, multi-customer operations as the industry enters its first true deployment year.

Humanoid Press Footer

Humanoid Press

Humanoid Daily News

Foundation Faces Scrutiny Over Humanoid Military Claims and Ukraine Deployments

NC13 Unmanned Ground Systems Hold Position for 45 Days in Ukraine

Open-Source CLMD Driver Closes the Loop on Cheap Joints

Apptronik Apollo 2 Runs Automotive Kitting on Mercedes-Benz Lines

Xiaomi Humanoid Completes Extended Shift on EV Factory Line

More headlines

Walker C1 Delivers First Keynote at ZEISS Innovation Summit 2026

LimX Luna Platform for Creative and Open-Ended Use

FACEGOOD Opens Pre-Orders and Source for Aurora A7X Lifelike Face

Boston Dynamics Adds Wheels to Atlas to Stress Whole-Body Control

Humanoid Daily News

Tesla Begins Production-Quality Optimus Units at Fremont

Deploying Whole-Body Policies on S10 + Piper Arm

Actuator Dynamics Curricula for Narrow-Viability Tasks

Open-Source CLMD Driver Closes the Loop on Cheap Joints

Humanoid Footstep Planning with Graphs of Convex Sets

More headlines

X-Humanoid Launches Wise Kaiwu Platform and Hi-WBC Controller

BrainCo Tests Vector-Based Retargeting on Revo 3 Using MANUS Gloves

3D Vision-Guided Welding Demonstrated in South Korea

T2 Translates Human Dynamic Motion into Stable Robotic Actions

Humanoid Daily News

Agile Robots Raises $500 Million Series F at $8.5 Billion Valuation

DEEPRobotics LYNX Supports 24/7 Inspection at Tongxiang Wastewater Plant

WOLKER Robotics Shows X30 Plant-Room Inspection Round

UC San Diego Releases OpenRoboKit Open-Source Robotics Suite

FW-02 Omnidirectional Wheeled Platform for Complex Terrain

More headlines

Astrall Dynamics Hypertron-SW01 for Industrial Inspection

ANSCER AR1250 Compact AMR Lifts Up to 1,250 kg

Generalist AI GEN-1.5 Shows One-Shot Learning of Physical Skills

Invariant EKF Enables Real-Time State Estimation on Non-Inertial Ground

Humanoid Daily News

Figure AI Ramps Figure 03 Production Past 1,000 Units at One Per Hour

Chinese Makers Dominate Humanoid Shipments as AGIBOT and Unitree Scale

Paxini Shows Humanoid Manipulation with Light-Bulb Task

Nyobolt Supplies Ultra-Fast Charging for Symbotic Warehouse Robots

PUDU MP2000 Maneuvers Smoothly Through Narrow Aisles

More headlines

TouchDIVER Pro Adds Force, Temperature and Texture Feedback to Humanoid Teleop

Generalist AI GEN-1.5 Demonstrates Engine Housing Assembly

HFUN-TECH T01 Targets Fully Unmanned Warehouse-to-Door Logistics

Astrall Dynamics Hypertron-SW01 Quadruped for Industrial Inspection

Humanoid Daily News

Boston Dynamics Adds Wheels to Atlas to Test Whole-Body Control

WonderDuck Starts Walking with Fully Open Hardware and Code

Handheld LiDAR SLAM Maps Crowded RobotWorld 2026 Hall in Real Time

Anchor 1 Advances Whole-Body Intelligence for Reliable Intent-to-Action

PUDU D5 Enables Autonomous Multi-Level Facility Patrols in Three Setup Steps

More headlines

InDro Robotics Introduces Prowler Wheeled-Legged Security Robot

MABEL Open-Source Mobile Manipulator Built for Approximately $8,100

AGIBOT A3 Ultra Completes Hotel Tasks in One Continuous Autonomous Take

osstrich Tests LocomotionV1 Policy on Real Hardware

Humanoid Daily News

X-Humanoid Launches Wise Kaiwu Platform with Hi-WBC Full-Body Control

Humanoid Footstep Planning Using Graphs of Convex Sets

Generalist AI Releases GEN-1.5 Foundation Model

Autonomous Pen Spinning Under External Disturbances on WUJI Hand 2

TeleXperience Enables Continuous Multi-Task Dual-Arm Operation

More headlines

Fashion Star TPU90 Flexible Gripper Adapts to Irregular Objects

Anchor 1 Advances Whole-Body Intelligence for Reliable Intent-to-Action

NVIDIA Isaac and AI Agents Accelerate Robot Skill Development

Additional Humanoid Developments

Humanoid Daily News

1X Shows Close-Up of the Motors and Actuators Inside NEO’s Tendon-Driven Hand

TESOLLO DG-5F-M Robotic Hand Mirrors Human Finger Movements in Real Time

BrainCo Revo 3 Hand Twists a Rubber Band Around Its Fingers

Xynova Flex 2 Puts a 23-DoF Hybrid-Drive Dexterous Hand in Focus

ETH Zurich Robotic Hand Crawls Without an Attached Robot Body

More headlines

Unidentified Home Robot Teaser Shows Chores, Block Play and Bedtime Assistance

ROBROS Presents IGRIS-MAX with Claimed 30 kg Payload

SpeedBot Presents Weldroid-W1 Humanoid Welding Robot

Chestnut Robotics (formerly TetherIA) Presents Aero Hand Open

Humanoid Daily News

Astribot T1 Performs Lab Task Under Teleoperation