Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Duke Humanoid: Design and Control For Energy Efficient Bipedal Locomotion Using Passive Dynamics

Boxi Xia, Bokuan Li, Jacob Lee, Michael Scutari, Boyuan Chen

TL;DR

This work presents Duke Humanoid v1.0, an open-source, mid-sized 10-DoF biped designed to study dynamic locomotion and energy efficiency. It combines a zero-shot velocity-tracking RL policy with an end-to-end learning framework that explicitly modulates passive dynamics via a per-joint torque-activation parameter, enabling significant energy savings. Through extensive sim-to-real bridging (domain randomization and joint dynamics tuning), the authors demonstrate a 31% reduction in cost of transport on real hardware and up to 50% improvement in simulation at low walking speeds. The platform and methods offer a transparent, extensible path toward energy-efficient humanoid locomotion, while acknowledging current limitations such as lack of arms and onboard power, with plans for future enhancements.

Abstract

We present the Duke Humanoid, an open-source 10-degrees-of-freedom humanoid, as an extensible platform for locomotion research. The design mimics human physiology, with symmetrical body alignment in the frontal plane to maintain static balance with straight knees. We develop a reinforcement learning policy that can be deployed zero-shot on the hardware for velocity-tracking walking tasks. Additionally, to enhance energy efficiency in locomotion, we propose an end-to-end reinforcement learning algorithm that encourages the robot to leverage passive dynamics. Our experimental results show that our passive policy reduces the cost of transport by up to $50\%$ in simulation and $31\%$ in real-world tests. Our website is http://generalroboticslab.com/DukeHumanoidv1/ .

The Duke Humanoid: Design and Control For Energy Efficient Bipedal Locomotion Using Passive Dynamics

TL;DR

This work presents Duke Humanoid v1.0, an open-source, mid-sized 10-DoF biped designed to study dynamic locomotion and energy efficiency. It combines a zero-shot velocity-tracking RL policy with an end-to-end learning framework that explicitly modulates passive dynamics via a per-joint torque-activation parameter, enabling significant energy savings. Through extensive sim-to-real bridging (domain randomization and joint dynamics tuning), the authors demonstrate a 31% reduction in cost of transport on real hardware and up to 50% improvement in simulation at low walking speeds. The platform and methods offer a transparent, extensible path toward energy-efficient humanoid locomotion, while acknowledging current limitations such as lack of arms and onboard power, with plans for future enhancements.

Abstract

We present the Duke Humanoid, an open-source 10-degrees-of-freedom humanoid, as an extensible platform for locomotion research. The design mimics human physiology, with symmetrical body alignment in the frontal plane to maintain static balance with straight knees. We develop a reinforcement learning policy that can be deployed zero-shot on the hardware for velocity-tracking walking tasks. Additionally, to enhance energy efficiency in locomotion, we propose an end-to-end reinforcement learning algorithm that encourages the robot to leverage passive dynamics. Our experimental results show that our passive policy reduces the cost of transport by up to in simulation and in real-world tests. Our website is http://generalroboticslab.com/DukeHumanoidv1/ .
Paper Structure (15 sections, 3 equations, 9 figures, 4 tables)

This paper contains 15 sections, 3 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Humanoid Platforms
  4. Control Strategies for Humanoid Locomotion
  5. Strategies for Energy-Efficient Locomotion
  6. Method
  7. Hardware Design
  8. Reinforcement Learning for Bipedal Locomotion
  9. Policy for energy-efficient locomotion
  10. Bridging the Sim-to-Real Gap
  11. Experiments and Results
  12. Locomotion in the Real
  13. Passive Policy for Energy-Efficient Locomotion
  14. Passive Locomotion in the Real
  15. Conclusion, Limitation, and Future Work

Figures (9)

  • Figure 1: Duke Humanoid v1.0: a) The frontal plane symmetry of the hip enables static standing with straight knees. b) and c) Additional poses demonstrating the robot's range of motion.
  • Figure 2: Mechanical Design Overview: a) Major dimensions and the extensible body design. b) All joints in the left leg. c) Two parallel linkages in the knee and ankle.
  • Figure 3: Chronophotograph showing the Duke Humanoid walking using the baseline RL policy.
  • Figure 4: Comparison of Simulated and Real-World Baseline Walking: Target joint positions (from RL policy), actual joint positions, and joint velocities for the HFE and KFE joints.
  • Figure 5: Comparison of Simulated and Real-World Passive Walking: Target joint positions (from RL policy), actual joint positions, and joint velocities for the HFE and KFE joints.
  • ...and 4 more figures