🤖 AI Summary
This work addresses the high energy consumption of bipedal robots by proposing an energy-efficient walking control method that integrates passive dynamics with end-to-end reinforcement learning (RL). We design Duke Humanoid—an open-source, anatomically inspired 10-DOF humanoid robot—optimized for static balance and co-modeled across hardware and simulation. We introduce a novel PPO/SAC-based RL algorithm explicitly incentivizing passive dynamic exploitation, enabling zero-shot sim-to-real transfer. Experiments demonstrate a 31% reduction in measured cost of transport (CoT) and a 50% reduction in simulation. To our knowledge, this is the first zero-shot hardware deployment of variable-speed, dynamically stable walking on an open-source ROS platform. The core contributions are: (i) a passive-dynamics–guided RL policy design that prioritizes energy-efficient gait emergence, and (ii) a cross-domain generalizable locomotion control framework bridging simulation fidelity and real-world robustness.
📝 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/ .