Learning Quiet Walking for
a Small Home Robot

Ryo Watanabe1,2, Takahiro Miki1, Fan Shi1,3, Yuki Kadokawa1,4,
Filip Bjelonic1, Kento Kawaharazuka1,5, Andrei Cramariuc1, Marco Hutter1
1ETH Zurich,2Sony Group Corporation,3National University of Singapore,
4Nara Institute of Science and Technology,5The University of Tokyo
ICRA 2025
Video arXiv

Abstract

As home robotics gains traction, robots are increasingly integrated into households, offering companionship and assistance. Quadruped robots, particularly those resembling dogs, have emerged as popular alternatives for traditional pets. However, user feedback highlights concerns about the noise these robots generate during walking at home, particularly the loud footstep sound. To address this issue, we propose a sim-to-real based reinforcement learning (RL) approach to minimize the foot contact velocity highly related to the footstep sound. Our framework incorporates three key elements: learning varying PD gains to actively dampen and stiffen each joint, utilizing foot contact sensors, and employing curriculum learning to gradually enforce penalties on foot contact velocity. Experiments demonstrate that our learned policy achieves superior quietness compared to a RL baseline and the carefully handcrafted Sony commercial controllers. Furthermore, the trade-off between robustness and quietness is shown. This research contributes to developing quieter and more user-friendly robotic companions in home environments.

MY ALT TEXT

Method

We focus on minimizing the footstep sound by using sim-to-real based RL to reduce foot contact velocity in the physics simulator, which is highly related to the footstep sound in the real world. Firstly, we incorporate foot switch contact sensors to detect when the feet touch the ground softly. Secondly, the policy estimates the joint PD gain scale as an action to stiffen joints during the stance phase for body support and dampen them during the contact phase. Thirdly, we employ a curriculum learning approach that initially trains the robot on basic locomotion and progressively applies penalties to noisy walking such as foot contact velocity to ensure quieter walking.

Quietness

The quietness evaluation is shown. The microphone at the rear head measures the magnitude of the sound. As you can hear, the proposed RL policy is quieter than the other baselines.

Robustness

To assess robustness, we varied the slope angle and used this value as a metric to compare robustness. The detailed experimental results are shown as follows. While the quiet RL proposed policy can climb up a small slope, it cant climb up a steeper slope such as 7 degree. On the other hands, the loud RL baseline can climb up the steeper slope. There is a trade-off between quietness and robustness. Domain randomization parameters can tune it.