SUMMARY:

 

Walker 2 is a full‑size bipedal humanoid robot developed by Shanghai DroidUp Co., Ltd. as a next‑generation locomotion research platform. It is engineered around bionic tendon‑driven actuation, Model Predictive Control (MPC), and Reinforcement Learning (RL) to achieve efficient, human‑like gait performance. Walker 2 is best known for completing the World Humanoid Robot Games half‑marathon in 4 hours 25 minutes without a battery swap, placing 3rd, demonstrating one of the strongest endurance benchmarks ever recorded for a humanoid robot.

 

With 6 RGB cameras, an 8‑microphone array, and a hybrid compute stack (6 TOPS onboard + 200 TOPS cloud), Walker 2 is designed for advanced gait research, multi‑terrain locomotion, and long‑duration mobility testing rather than commercial deployment.

 

EDITORIAL:

 

Walker 2 represents DroidUp’s commitment to locomotion‑first humanoid robotics, a field focused on endurance, stability, and terrain adaptability rather than manipulation or general‑purpose tasks. Its tendon‑driven actuation system is central to its identity, enabling biomimetic joint behavior, smoother motion, and reduced energy consumption during long‑duration walking.

 

Where many humanoids emphasize manipulation, expressive interaction, or industrial utility, Walker 2 is built around pure bipedal performance. Its hybrid MPC + RL control stack allows it to adapt to terrain changes, maintain balance under dynamic conditions, and optimize gait efficiency in real time.

 

The robot’s half‑marathon achievement — over four hours of continuous running — is a milestone for the field, proving that long‑range, real‑world locomotion is achievable for bipedal robots today. Walker 2 is not a consumer robot, not a warehouse robot, and not a social robot. It is a research‑grade locomotion platform, built to push the boundaries of gait control, multi‑terrain stability, and embodied AI.

 

SPECIFICATIONS:

 

Dimensions & Build

- Height: Not published  

- Weight: Not published  

- Design Intent: High‑endurance bipedal locomotion and multi‑terrain gait research  

- Construction:  

  - Full‑size bipedal humanoid  

  - Bionic tendon‑driven joints  

  - Lightweight performance‑oriented frame  

 

Note: No structural load ratings, torque values, or material composition published.

 

Mobility

- Locomotion: Full bipedal walking  

- Actuation: Bionic tendon‑driven actuators with adjustable stiffness  

- Control System: Model Predictive Control (MPC) + Reinforcement Learning (RL)  

- Terrain Capability: Multi‑terrain walking, obstacle climbing  

- Endurance: Completed humanoid half‑marathon in 4h 25m, 3rd place  

 

Note: No incline, step‑height, walking‑speed, or fall‑recovery metrics published.

 

Degrees of Freedom

- Total DOF: Not published  

- Arms: Present (details not published)  

- Hands: Not documented  

- Legs: Full bipedal DOF (exact count not published)

 

Actuation

- Actuator Type: Bionic tendon‑driven actuators  

- Capabilities:  

  - Smooth, biomimetic motion  

  - Adjustable joint stiffness  

  - Energy‑efficient gait cycles  

 

Note: No torque or joint‑speed values published.

 

Manipulation

- Capabilities:  

  - Basic arm presence  

  - No documented manipulation features  

 

Note: No payload, grip‑force, or hand‑DOF data published.

 

Perception & Sensors

Confirmed capabilities:

- 6 RGB cameras  

- 8‑microphone array  

- Multi‑camera mapping  

- Obstacle detection  

 

Note: No camera resolution, depth‑sensor type, FOV, IMU, or tactile sensor specifications published.

 

Compute & AI

- AI System: MPC + RL locomotion stack  

- On‑Device Compute: 6 TOPS  

- Cloud Compute: 200 TOPS  

- Capabilities:  

  - Adaptive gait control  

  - Multi‑terrain locomotion  

  - Energy‑optimized motion planning  

 

Note: No CPU/GPU details, SDK availability, or programming‑language support published.

 

Battery & Power

- Power Source: Electric  

- Max Operating Time: 360 minutes (6 hours)  

- Charging: Not published  

 

Note: No battery capacity, charge time, or power‑draw data published.

 

Connectivity

- Wireless: Not published  

- Integration: Not documented  

- Real‑Time Control Interfaces: Not documented  

 

Note: No protocol‑level details published.

 

Applications

- Locomotion research  

- Multi‑terrain gait development  

- RL/MPC control experiments  

- Endurance benchmarking  

- Academic and industrial R&D  

 

---

 

Pricing

- Not published  

- Research‑only platform  

 

DroidUp has NOT published:

- Height  

- Weight  

- Payload  

- Grip force  

- Joint torque  

- Joint speed  

- Camera resolution  

- Depth‑sensor specifications  

- IMU specifications  

- Battery capacity  

- Runtime details  

- Charge time  

- Environmental ratings (IP, temperature, humidity)  

- Step height  

- Max incline  

- Turning radius  

 

Image: @zoomvantage

 

Company Source

 

https://youtu.be/fiVmp6cjHSw?si=w17ptN9H5TOzKiWb

 

Key News & Coverage

 

https://youtu.be/Ol9nj2BxPH0?si=xMX1R1tzAotINuuz