Compliant feet cut quadruped robot energy use by 17% in RL study

Researchers at the Indian Institute of Science, Robert Bosch Centre for Cyber Physical Systems, and Ahmedabad University have demonstrated that equipping quadruped robots with compliant (spring-loaded) feet can reduce energy consumption by roughly 17% during walking. Led by Pramod Pal, the team integrated foot compliance into a reinforcement learning (RL) locomotion controller, training eight policies for eight different spring stiffness values. They then cross-evaluated these policies in both simulation and on a physical quadruped robot, measuring mechanical energy consumed per meter traveled. The results consistently showed that an intermediate spring stiffness significantly outperformed both very stiff (rigid) and very flexible feet. The stiffest feet wasted energy by failing to absorb impact, while overly soft feet caused instability and energy loss. The optimum stiffness struck a balance, allowing the foot to deform on impact, store elastic energy, and release it during push-off—much like the spring-like tendons in animal legs.

This work addresses a key limitation of conventional quadruped robots, which typically use rigid feet for simplicity and stable control. Rigid feet cannot absorb ground reaction forces or recycle energy, leading to higher energy expenditure. By adding compliance and optimizing it via RL, the team achieved a measurable efficiency gain without sacrificing stability. The study, published on arXiv (2605.14411), provides a straightforward hardware modification—choosing the right spring for the feet—that can be applied to existing robot designs. Immediate applications include legged delivery robots, search-and-rescue quadrupeds, and industrial inspection drones, where longer battery life is critical. By showing that a low-cost mechanical change plus RL tuning yields substantial energy savings, the research paves the way for more agile and economical legged robots.