STATE-NAV: Stability-Aware Traversability Estimation for Bipedal Navigation on Rough Terrain
Ziwon Yoon, Lawrence Y. Zhu, Jingxi Lu, Lu Gan, Ye Zhao
AI summary
Problem
Bipedal robots face high fall risks on rough terrain, yet existing traversability methods rely on manual geometric rules or transfer features from wheeled/quadrupedal platforms that poorly predict bipedal instability. Traditional navigation planners also require tedious manual weight tuning to balance stability and speed.
Approach
A transformer network predicts body-to-stance-foot angle instability with uncertainty from terrain maps and velocity commands. This prediction defines traversability as the maximum stable command velocity, which guides a hierarchical TravRRT* global planner and MPC local planner.
Key results
- Identified body-to-stance-foot angle as the most effective self-supervised signal for predicting bipedal fall risk
- Developed TravFormer, a transformer-based estimator that predicts instability with aleatoric uncertainty from elevation maps and velocity commands
- Formulated traversability as a stability-aware command velocity, eliminating environment-specific weight tuning in path planning
- Validated the framework on the Digit humanoid in simulation and real-world tests, demonstrating improved stability, time efficiency, and robustness
Why it matters
Enables safer, faster, and more autonomous navigation for humanoids in complex environments without tedious manual tuning, advancing practical deployment of bipedal robots.
Abstract
Bipedal robots have advantages in maneuvering human-centered environments, but face greater failure risk com- pared to other stable mobile platforms, such as wheeled or quadrupedal robots. While learning-based traversability has been widely studied for these platforms, bipedal traversability has in- stead relied on manually designed rules with limited consideration of locomotion stability on rough terrain. In this work, we present the first learning-based traversability estimation and risk-sensitive navigation framework for bipedal robots operating in diverse, uneven environments. TravFormer, a transformer-based neural network, is trained to predict bipedal instability with uncertainty, enabling risk-aware and adaptive planning. Based on the network, we define traversability as stability-aware command velocity—the fastestcommandvelocitythatkeepsinstabilitybelowauser-defined limit. This velocity-based traversability is integrated into a hi- erarchical planner that combines traversability-informed Rapid RandomTreeStar(TravRRT*)fortime-efficientpathplanningand Model Predictive Control (MPC) for safe execution. We validate our method in MuJoCo simulator and the real world, demon- strating improved stability, time efficiency, and robustness across diverse terrains compared with existing methods.