Learning quiet walking for a small home robot

This work addresses the problem of excessive foot-step noise when quadruped robots walk in home environments — a major obstacle for their acceptance as household companions. The paper proposes a sim-to-real reinforcement learning framework that minimizes foot contact velocity (a major contributor to footstep sound), by allowing the policy to dynamically adjust PD gains per joint, using foot-contact sensors to modulate damping/stiffening, and applying curriculum learning to gradually penalize high-contact velocities.

Experiments with a small home robot show that the learned policy produces significantly quieter walking compared both to a standard RL baseline and to carefully hand-tuned commercial controllers. The study demonstrates the feasibility of noise-aware locomotion and highlights the trade-off between noise reduction and robustness, paving the way for more user-friendly robots operating in domestic environments.

Recommended citation: R. Watanabe, T. Miki, F. Shi, Y. Kadokawa, F. Bjelonic, K. Kawaharazuka, A. Cramariuc, and M. Hutter, “Learning Quiet Walking for a Small Home Robot,” in Proc. IEEE Int. Conf. on Robotics and Automation (ICRA), 2025.
Download Paper | Download Bibtex