Traversability-Aware Legged Navigation by Learning from Real-World Visual Data
Hongbo Zhang, Zhongyu Li, Xuanqi Zeng, Laura Smith, Kyle Stachowicz, Dhruv Shah, Linzhu Yue, Zhitao Song, Weipeng Xia, Sergey Levine, Koushil Sreenath, Yunhui Liu
AI summary
Problem
Existing traversability-aware navigation methods rely on human-labeled features or simulation, failing to account for the robot's specific locomotion capabilities and struggling with the sim-to-real gap in unstructured environments.
Approach
The framework uses a four-stage hierarchical reinforcement learning pipeline that trains a high-level RGB-D navigation planner directly in the real world, guided by a traversability estimator derived from the robot's own low-level control performance.
Key results
- Robot-centric traversability estimator based on the low-level controller's value function
- Multi-modal RGB-D navigation planner trained directly in the real world
- Sample-efficient real-world RL pipeline combining offline demonstrations and online interactions
- Successful generalization to unseen off-road terrains with near-optimal path planning in 15-20 minutes
Why it matters
Enables quadrupedal robots to autonomously and efficiently navigate complex, unstructured outdoor environments without relying on accurate simulators or human-defined terrain labels.
Abstract
The enhanced mobility brought by legged locomo- tion empowers quadrupedal robots to navigate through com- plex and unstructured environments. However, optimizing agile locomotion while accounting for the varying energy costs of traversing different terrains remains an open challenge. Most previous work focuses on planning trajectories with traversability cost estimation based on human-labeled environmental features. This human-centric approach is insufficient because it does not account for the varying capabilities of the robot locomotion controllers over challenging terrains. To address this, we in- troduce a novel real-world learning pipeline that unifies offline demonstrations, online reinforcement learning, and multi-modal perception to achieve robust legged navigation. The framework employs multiple training stages to develop a planner that guides the robot in avoiding obstacles and hard-to-traverse terrains while reaching its goals. We first develop a novel traversability estimator in a robot-centric manner. The training of the navigation planner is directly performed in the real world using a sample efficient reinforcement learning method. With the proposed method, a quadrupedal robot learns to perform traversability-aware navigation through real-world interactions in diverse offroad and unstructured environments. Moreover, the robot demonstrates the ability to generalize the learned navigation skills to unseen scenarios.