Learning robust perceptive locomotion for quadrupedal robots in the wild

TL;DR

This work tackles the challenge of robust locomotion for quadrupeds in the wild by integrating exteroceptive perception with proprioception through an attention-based recurrent encoder. The controller is trained via a privileged learning framework (teacher-student) in simulation and deployed zero-shot on real robots using a robot-centric elevation map, enabling fast, obstacle-aware locomotion even when perception is noisy or incomplete. It achieves superior speed and obstacle handling compared to proprioceptive baselines and demonstrates zero-falls across diverse terrains, including a lengthy Alps hike and DARPA Subterranean Challenge scenarios. The approach offers a general framework for robust, sensor-fused legged locomotion with practical implications for autonomous exploration in harsh environments.

Abstract

Legged robots that can operate autonomously in remote and hazardous environments will greatly increase opportunities for exploration into under-explored areas. Exteroceptive perception is crucial for fast and energy-efficient locomotion: perceiving the terrain before making contact with it enables planning and adaptation of the gait ahead of time to maintain speed and stability. However, utilizing exteroceptive perception robustly for locomotion has remained a grand challenge in robotics. Snow, vegetation, and water visually appear as obstacles on which the robot cannot step~-- or are missing altogether due to high reflectance. Additionally, depth perception can degrade due to difficult lighting, dust, fog, reflective or transparent surfaces, sensor occlusion, and more. For this reason, the most robust and general solutions to legged locomotion to date rely solely on proprioception. This severely limits locomotion speed, because the robot has to physically feel out the terrain before adapting its gait accordingly. Here we present a robust and general solution to integrating exteroceptive and proprioceptive perception for legged locomotion. We leverage an attention-based recurrent encoder that integrates proprioceptive and exteroceptive input. The encoder is trained end-to-end and learns to seamlessly combine the different perception modalities without resorting to heuristics. The result is a legged locomotion controller with high robustness and speed. The controller was tested in a variety of challenging natural and urban environments over multiple seasons and completed an hour-long hike in the Alps in the time recommended for human hikers.

Figures (9)

• Figure 1: Robust locomotion in the wild. The presented locomotion controller was extensively tested in a variety of complex environments over multiple seasons. The controller overcame a whole spectrum of real-world challenges, often encountering them in combination. These include slippery surfaces, steep inclinations, complex terrain, and vegetation in natural environments. In search-and-rescue scenarios, the controller dealt with steep stairs, unknown payloads, and perception-degrading fog. Reflective surfaces, loose ground, low light, and water puddles were encountered in underground cave systems. Soft and slippery snow piled up in the winter. The controller traversed these environments with zero failures.
• Figure 2: A hike on the Etzel mountain in Switzerland, completed by ANYmal with our locomotion controller. The 2.2km route -- with 120m of elevation gain and inclinations up to 38% -- encompasses a variety of challenging terrain. ANYmal reached the summit faster than the human time indicated in the official signage, and finished the entire route in virtually the same time as given by a hiking guide komoot.
• Figure 3: Our locomotion controller perceives the environment through height samples (red dots) from an elevation map (A). The controller is robust to many perception challenges commonly encountered in the field: missing map information due to sensing failure (B, C, G) and misleading map information due to non-rigid terrain (D, E) and pose estimation drift (F).
• Figure 4: Internal belief state inspection during perceptive failure using a learned belief decoder. Red dots indicate height samples given as input to the policy. Blue dots show the controller's internal estimate of the terrain profile. (A) After stepping on a soft obstacle that cannot support a foothold, the policy correctly revises its estimate of the terrain profile downwards. (B) A transparent obstacle is correctly incorporated into the terrain profile after contact is made. (C) With operational sensors, the robot swiftly and gracefully climbs the stairs, with no spurious contacts. (D) When the robot is blinded by covering the sensors, the policy can no longer anticipate the terrain but remains robust and successfully traverses the stairs. (E) When stepping onto a slippery platform, the policy identifies low friction and compensates for the induced pose estimation drift. The graph shows a decoded friction coefficient.
• Figure 5: Overview of the training methods and deployment. We first train a teacher policy with access to privileged simulation data using reinforcement learning (RL). This teacher policy is then distilled into a student policy, which is trained to imitate the teacher's actions and to reconstruct the ground-truth environment state from noisy observations. We deploy the student policy zero-shot on real hardware using height samples from a robot-centric elevation map.