Reproduce “Energy-Efficient Quadruped Locomotion with Compliant Feet” on OpenResearch

About this paper

Pramod Pal, Shishir Kolathaya, Ashitava Ghosal

Researchers at the Indian Institute of Science explored the effect of passive foot compliance on quadruped robot energy consumption when combined with reinforcement learning (RL) controllers. Their custom robot, featuring compliant feet and RL-trained gaits, demonstrated that an optimal intermediate foot stiffness reduced mechanical energy consumption by up to 16.94% on physical hardware compared to a stiff foot design.

Key insights

• An optimal intermediate foot stiffness (14,500 N/m) exists that minimizes mechanical energy consumption in quadruped locomotion, demonstrating a balance between impact absorption and excessive compression.
• Integrating passive mechanical compliance directly at the foot-ground interface effectively synergizes with learned control policies, allowing the hardware to manage impacts and store energy while the controller optimizes motion.
• Policies trained in more compliant environments (e.g., with softer springs) tend to learn more energy-efficient strategies that generalize better across different foot stiffnesses.

Summary by alphaXiv.