Minimizing Acoustic Noise: Enhancing Quiet Locomotion for Quadruped Robots in Indoor Applications

Zhanxiang Cao, Buqing Nie, Yang Zhang, Yue Gao

Abstract

Recent advancements in quadruped robot research have significantly improved their ability to traverse complex and unstructured outdoor environments. However, the issue of noise generated during locomotion is generally overlooked, which is critically important in noise-sensitive indoor environments, such as service and healthcare settings, where maintaining low noise levels is essential. This study aims to optimize the acoustic noise generated by quadruped robots during locomotion through the development of advanced motion control algorithms. To achieve this, we propose a novel approach that minimizes noise emissions by integrating optimized gait design with tailored control strategies. This method achieves an average noise reduction of approximately 8 dBA during movement, thereby enhancing the suitability of quadruped robots for deployment in noise-sensitive indoor environments. Experimental results demonstrate the effectiveness of this approach across various indoor settings, highlighting the potential of quadruped robots for quiet operation in noise-sensitive environments.

Introduction

While quadruped robots have achieved remarkable agility in rugged outdoor terrains, their transition to indoor service roles—such as healthcare assistance, library maintenance, or residential support—has encountered a significant barrier: acoustic noise. In these environments, the rhythmic "clack-clack" of robotic feet striking hard surfaces is more than a nuisance; it is a disruption that can increase human stress levels and impair productivity.

Most existing research in quadrupedal locomotion prioritizes speed, stability, and power efficiency, often ignoring the acoustic footprint. The primary source of this noise is the kinetic energy released during foot-ground collisions. This paper introduces MUTE (Minimizing acoUsTic noisE), a reinforcement learning-based control strategy designed to maintain stability and speed while drastically softening the robot's acoustic impact.

Methodology

The Physics of Quiet Locomotion

The MUTE framework treats noise reduction as a constrained optimization problem within a Reinforcement Learning (RL) pipeline. The core physical insight is that sound amplitude is proportional to the square root of impact energy ($E = \frac{1}{2}mv^2$). Consequently, the most effective way to reduce noise is to minimize the foot's downward velocity at the exact moment of contact.

Phase Estimation and Control

To achieve a "soft" landing, the robot must predict the exact moment of contact. The researchers utilized an LSTM (Long Short-Term Memory) encoder-decoder to estimate the "gait phase." This allows the controller to anticipate the transition from the swing phase (foot in air) to the stance phase (foot on ground) with high precision.

A novel scalar input, the Quiet Factor ($\beta$), was introduced to allow dynamic adjustment:

  • $\beta = 0$: Prioritizes speed and agility (standard operation).
  • $\beta = 1$: Prioritizes noise minimization (stealth mode).

This parameter enables a single policy to trade off between operational speed and acoustic discretion depending on the environment.

Training Architecture

The system was trained using an Asymmetric Actor-Critic (AAC) framework with Proximal Policy Optimization (PPO). During simulation, the "Critic" utilizes privileged information—such as exact terrain height and contact forces—to guide the learning process. The "Actor" (the policy deployed on the physical robot) relies solely on onboard proprioceptive sensors, ensuring the system remains functional in real-world scenarios without external sensors.

Experimental Setup

Simulation and Real-World Transfer

Training was conducted in Isaac Gym, utilizing 4,096 parallel agents to simulate approximately 6,000 episodes (8 hours of training). The resulting policy was transferred "zero-shot" to a Unitree Go1 EDU robot, meaning no further tuning was required for the physical hardware.

Evaluation Surfaces

The robot's performance was evaluated on three common indoor surfaces: wooden flooring, carpet, and porcelain tiles. These surfaces represent the variety of acoustic challenges found in modern indoor infrastructure.

Results and Discussion

Quantitative Noise Reduction

The MUTE framework achieved a significant reduction in acoustic output. On wooden floors at a speed of 0.5 m/s, MUTE (with $\beta=1$) produced 65.78 dBA, compared to 79.74 dBA for the robot's built-in MPC controller. While an 8-14 dBA reduction may seem modest numerically, the decibel scale is logarithmic; an 8 dBA reduction represents a 2.5-fold decrease in sound pressure.

Performance Across Environments

In a long-distance reliability test, the robot traversed 91.7 meters across mixed indoor surfaces. It maintained an average noise level of 68.25 dBA, staying safely below the 70 dBA threshold generally considered comfortable for human hearing. Interestingly, even when MUTE was set to its loudest setting ($\beta=0$), it remained quieter than baseline controllers like DreamWaQ, suggesting that the optimized gait is inherently more acoustically efficient.

Limitations

Despite its success, the approach faces certain trade-offs. To operate quietly, the robot must move its legs more deliberately, which can increase velocity tracking errors at higher speeds. Additionally, the robot’s internal components—cooling fans and motors—produce a baseline noise of approximately 55 dBA, setting a floor for how quiet the system can ultimately be.

Conclusion

The MUTE framework demonstrates that software-based control strategies can significantly mitigate the acoustic impact of quadruped robots. By using foot-drop velocity as a proxy for sound and employing phase-aware reward functions, the researchers have enabled a "cat-like" landing capability. Future work may integrate exteroception (vision or LiDAR) to allow the robot to automatically adjust its Quiet Factor based on the detected floor material, further enhancing its suitability for sensitive indoor environments like hospitals and libraries.

Summary updated

Abstract

Recent advancements in quadruped robot research have significantly improved their ability to traverse complex and unstructured outdoor environments. However, the issue of noise generated during locomotion is generally overlooked, which is critically important in noise-sensitive indoor environments, such as service and healthcare settings, where maintaining low noise levels is essential. This study aims to optimize the acoustic noise generated by quadruped robots during locomotion through the development of advanced motion control algorithms. To achieve this, we propose a novel approach that minimizes noise emissions by integrating optimized gait design with tailored control strategies. This method achieves an average noise reduction of approximately 8 dBA during movement, thereby enhancing the suitability of quadruped robots for deployment in noise-sensitive indoor environments. Experimental results demonstrate the effectiveness of this approach across various indoor settings, highlighting the potential of quadruped robots for quiet operation in noise-sensitive environments.
BibTeX
@inproceedings{cao2025minimizing,
  author = {Zhanxiang Cao and Buqing Nie and Yang Zhang and Yue Gao},
  title = {{Minimizing Acoustic Noise: Enhancing Quiet Locomotion for Quadruped Robots in Indoor Applications}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2025},
  doi = {10.1109/iros60139.2025.11246563},
}
Figure