Learning an Adaptive Fall Recovery Controller for Quadrupeds on Complex Terrains

TL;DR

This paper presents an Adaptive Fall Recovery (AFR) controller for quadrupedal robots on challenging terrains such as rocky, breams, steep slopes, and irregular stones, which generalizes well to complex terrains and outperforms baseline methods in terms of success rate and recovery speed.

Abstract

Legged robots have shown promise in locomotion complex environments, but recovery from falls on challenging terrains remains a significant hurdle. This paper presents an Adaptive Fall Recovery (AFR) controller for quadrupedal robots on challenging terrains such as rocky, breams, steep slopes, and irregular stones. We leverage deep reinforcement learning to train the AFR, which can adapt to a wide range of terrain geometries and physical properties. Our method demonstrates improvements over existing approaches, showing promising results in recovery scenarios on challenging terrains. We trained our method in Isaac Gym using the Go1 and directly transferred it to several mainstream quadrupedal platforms, such as Spot and ANYmal. Additionally, we validated the controller's effectiveness in Gazebo. Our results indicate that the AFR controller generalizes well to complex terrains and outperforms baseline methods in terms of success rate and recovery speed.

Figures (6)

• Figure 1: Overview of the AFR framework. Trained on Go1 could be deployed on multiple platforms.
• Figure 2: Total reward comparison among AFR, AFR_ RAW and PPO.
• Figure 3: Target Posture Reward of AFR, AFR_ RAW and PPO.
• Figure 4: Simulation results of the Go1, Aliengo, Anymal B and Spot robots recovering from a prone position on uneven terrain. Each row (A-H) represents a different terrain: (A) Stairs, (B) Single Gaps, (C) Air Beams, (D) Beams, (E) Sparse Stones, (F) Slope, (G) Discrete Obstacles, and (H) Dense Stones. Columns show key phases in the recovery process across time steps.
• Figure 5: Joint torque profiles during recovery on stairs.