Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quadruped robot traversing 3D complex environments with limited perception

Yi Cheng, Hang Liu, Guoping Pan, Linqi Ye, Houde Liu, Bin Liang

TL;DR

This work tackles robust quadruped locomotion in unknown 3D environments under limited perception by developing a proprioception-only, end-to-end motion controller. It introduces a history-based collision estimator and the Collision Domain with a Hybrid Imagination Model to implicitly characterize 3D obstacles, enabling accurate collision detection, localization, and agile responses via a teacher–student training framework optimized with PPO. The method is validated in large-scale simulation and real-world experiments on a Unitree Go2, showing improved traversability across Highlands, Barriers, Tunnels, and Cracks, and demonstrating the value of explicit collision-domain reasoning and offline-to-online transfer. The results highlight practical significance for navigation in low-visibility or sensor-degraded settings, with potential extensions to force-direction estimation and adaptive control across variable stiffness and contact conditions.

Abstract

Traversing 3-D complex environments has always been a significant challenge for legged locomotion. Existing methods typically rely on external sensors such as vision and lidar to preemptively react to obstacles by acquiring environmental information. However, in scenarios like nighttime or dense forests, external sensors often fail to function properly, necessitating robots to rely on proprioceptive sensors to perceive diverse obstacles in the environment and respond promptly. This task is undeniably challenging. Our research finds that methods based on collision detection can enhance a robot's perception of environmental obstacles. In this work, we propose an end-to-end learning-based quadruped robot motion controller that relies solely on proprioceptive sensing. This controller can accurately detect, localize, and agilely respond to collisions in unknown and complex 3D environments, thereby improving the robot's traversability in complex environments. We demonstrate in both simulation and real-world experiments that our method enables quadruped robots to successfully traverse challenging obstacles in various complex environments.

Quadruped robot traversing 3D complex environments with limited perception

TL;DR

This work tackles robust quadruped locomotion in unknown 3D environments under limited perception by developing a proprioception-only, end-to-end motion controller. It introduces a history-based collision estimator and the Collision Domain with a Hybrid Imagination Model to implicitly characterize 3D obstacles, enabling accurate collision detection, localization, and agile responses via a teacher–student training framework optimized with PPO. The method is validated in large-scale simulation and real-world experiments on a Unitree Go2, showing improved traversability across Highlands, Barriers, Tunnels, and Cracks, and demonstrating the value of explicit collision-domain reasoning and offline-to-online transfer. The results highlight practical significance for navigation in low-visibility or sensor-degraded settings, with potential extensions to force-direction estimation and adaptive control across variable stiffness and contact conditions.

Abstract

Traversing 3-D complex environments has always been a significant challenge for legged locomotion. Existing methods typically rely on external sensors such as vision and lidar to preemptively react to obstacles by acquiring environmental information. However, in scenarios like nighttime or dense forests, external sensors often fail to function properly, necessitating robots to rely on proprioceptive sensors to perceive diverse obstacles in the environment and respond promptly. This task is undeniably challenging. Our research finds that methods based on collision detection can enhance a robot's perception of environmental obstacles. In this work, we propose an end-to-end learning-based quadruped robot motion controller that relies solely on proprioceptive sensing. This controller can accurately detect, localize, and agilely respond to collisions in unknown and complex 3D environments, thereby improving the robot's traversability in complex environments. We demonstrate in both simulation and real-world experiments that our method enables quadruped robots to successfully traverse challenging obstacles in various complex environments.
Paper Structure (17 sections, 10 equations, 7 figures, 5 tables)

This paper contains 17 sections, 10 equations, 7 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Works
  3. Contributions
  4. Method
  5. Task Formulation
  6. Base Set
  7. Collision Estimator
  8. Implicit Collision Domain imagination
  9. EXPERIMENTS
  10. Training set up
  11. Hardware and Depoly
  12. Compared Method
  13. Simulation Experiments
  14. Real-World Experiments
  15. CONCLUSION
  16. ...and 2 more sections

Figures (7)

  • Figure 1: Teacher-student based two-stage training framework, where on the left is the schematic of the collision estimation.
  • Figure 2: Schematic of the collision domain
  • Figure 3: Locomotion through a tunnel: Adaptation strategies for collision avoidance.
  • Figure 4: Collision response strategies when facing a highland.
  • Figure 5: The t-SNE visualization for Collision Domain in the latent space.
  • ...and 2 more figures