Summary:

 

Adam is a state-of-the-art humanoid robot from PNDbotics, representing significant progress in Chinese humanoid robotics. Standing at 1.67 meters (5.5 feet) and weighing approximately 60–63 kg depending on the variant, Adam is designed for human-like agility, precision, and adaptability in complex tasks.

 

Powered by Quasi-Direct-Drive (QDD) actuators with peak torques up to 340 N·m in key leg joints (knee/hip), Adam achieves smooth, compliant, and energy-efficient motion. Its movements are driven by advanced AI, including deep reinforcement learning (DRL), imitation learning, and whole-body control algorithms.

 

The modular architecture, open-source SDK, and developer tools make it highly customizable for research, industrial prototyping, AI development, manipulation studies, and human-robot interaction.

 

Recent highlights include impressive demos like Charleston dance performances, synchronized routines, and the introduction of enhanced variants such as Adam-U Ultra for faster autonomous learning.

 

Editorial:

 

PNDbotics' Adam series marks a major advancement in humanoid robotics, combining biomimetic design (e.g., human-like pelvis and gait) with powerful AI to deliver truly anthropomorphic movement and interaction.

 

A key strength is the learning-based control system, leveraging reinforcement and imitation learning for dynamic balance, fall recovery, and expressive tasks like dancing or object handling. This enables real-world adaptability without rigid programming.

 

The modular platform—featuring interchangeable PSA (PND Smart Actuator) components—supports easy upgrades, maintenance, and tailoring. Open APIs, simulation environments, and RL tools accelerate development across academia and industry.

 

Practical demos continue to impress: recent 2026 videos show Adam nailing energetic Charleston routines with precise footwork and coordination, plus freestyle elements. At CES 2026 (January), PNDbotics showcased Adam (Standard/SP variants), Adam-U research platform, and modular actuators.

 

The newer Adam-U Ultra (early 2026 release) pushes boundaries further with an integrated Vision-Language-Action (VLA) model pre-trained on 10,000+ real-world samples. It performs complex manipulation out-of-the-box—no calibration needed—and learns/deploy new skills in hours, ideal for industrial and research acceleration.

In summary, Adam and its ecosystem offer a glimpse into scalable, AI-driven humanoid futures. With mass-production groundwork via PND-IMC and ongoing software refinements, it stands out as a versatile, developer-focused platform in the rapidly evolving humanoid landscape.

 

Specifications:

 

Physical Profile Height: 1.67 m (all main variants) Weight:

Adam Lite: ~60 kg

Adam Standard: ~61 kg

Adam SP: ~62 kg

Adam Pro: ~63 kg Body Material: Lightweight aluminum + composite shell (modular) Design Purpose: Research, AI locomotion/manipulation development, dexterous tasks, industrial R&D, HRI

Degrees of Freedom (DoF):

VariantTotal DoFNotes

Adam Lite25Core locomotion + basic spherical hands/arms

Adam Standard29Enhanced arms (wrist DoF added)

Adam SP41Dexterous hands (6 DoF fingers each), high manipulation

Adam Pro43Adds 2 DoF neck (yaw/pitch) for full expressiveness; ZED MINI sensors

(Note: Some media/demos cite up to 44 DoF for top configs, likely minor rounding or extras.)

Adam-U (separate upper-body/research variant): ~31 DoF, adjustable height 1.35–1.77 m. Adam-U Ultra (2026): Upper-body focused with advanced VLA AI; rapid skill learning (hours); precision QDD joints.

Actuation System: Actuator Type: Quasi-Direct-Drive (QDD), force-controlled, high compliance Peak Torque: Up to 340 N·m (knee/hip), ~46 N·m (ankle), ~60 N·m (arms) Control Characteristics: High compliance, low latency, smooth/human-like motion, energy-efficient torque control, modular PSA actuators

Compute & AI: Main Processor: Intel NUC 12th-Gen i7 (12 cores) AI/Perception Compute: NVIDIA Jetson Orin NX 16GB (100 TOPS) on Standard/SP/Pro AI Capabilities: Deep reinforcement learning (DRL), imitation learning, real-time sensor fusion, motion planning, VLA model (Adam-U Ultra) Open-source SDK: Yes, with real-time APIs, simulation, RL training tools, motion library

Perception & Sensors: Multi-sensor fusion stack 3D perception: RealSense D455 (Standard/SP) or ZED MINI (Pro) Force/torque sensing in limbs, high-precision proprioception Real-time feedback for balance/manipulation; supports LiDAR/camera expansion

Manipulation: Hands: Multi-DoF dexterous (SP/Pro: 6 joints per finger: thumb x2, index, middle, ring, pinky) Capabilities: Sorting, assembly, fine motor control, object interaction (stronger in SP/Pro/Ultra)

Locomotion & Mobility: Human-like bipedal walking (max ~1.5 m/s), dynamic balance (WBC + MPC) Fall recovery, obstacle avoidance, stair traversal Advanced demos: Charleston dance, freestyle, short running/jumping elements

Battery: Type: High-capacity modular pack (1172 Wh, 46.2V) Runtime: Up to ~4 hours active (varies; ~2 hours high-demand) Swappable: Yes, with intelligent BMS

Software & Developer Tools: Open-source SDK, real-time control APIs Simulation environment, reinforcement learning tools Motion library + behavior editor PND-Network protocol for sensor/integration

Positioning: Adam is a research-grade, modular humanoid optimized for: AI locomotion research, manipulation studies, robotics labs Industrial prototyping, human–robot interaction experiments Developer platforms with mass-production path via PND-IMC

 

Image: @XRoboHub

 

Company Source:

 

https://youtu.be/JyPiC0aArPg?si=LaLob489J4n0KtKu

 

Key News & Coverage:

 

China's humanoid robot nails Charleston dance: https://interestingengineering.com/ai-robotics/pndbiotics-humanoid-robot-aces-charleston-dance (Jan 2026)

 

Adam-U Ultra learns skills in hours: https://interestingengineering.com/ai-robotics/chinas-humanoid-robot-masters-complex-skills (Jan 2026)