One Filter to Deploy Them All: Robust Safety for Quadrupedal Navigation in Unknown Environments
Albert Lin, Shuang Peng, Somil Bansal
AI summary
Problem
Existing safety methods for legged robots either require a priori knowledge of the environment or specific controller details, and fail to adapt to novel obstacles and unmodeled dynamics during real-world deployment.
Approach
The authors develop an observation-conditioned reachability (OCR) safety-filter framework that uses a neural network to predict real-time safety value functions from LiDAR scans and disturbance estimates, dynamically overriding nominal controllers when necessary.
Key results
- Policy-agnostic safety filtering across diverse quadruped controllers without retraining
- Real-time adaptation to novel obstacles via LiDAR-based observation conditioning
- Robustness to unmodeled dynamics through online disturbance estimation
- Successful simulation and hardware validation on a Unitree Go1 quadruped
Why it matters
Enables reliable, safe deployment of learning-based legged robots in unstructured real-world environments without costly policy-specific safety tuning.
Abstract
As learning-based methods for legged robots rapidly grow in popularity, it is important that we can provide safety as- surances efficiently across different controllers and environments. Existing works either rely on a priori knowledge of the environment and safety constraints to ensure system safety or provide assur- ances for a specific locomotion policy. To address these limitations, we propose an observation-conditioned reachability-based (OCR) safety-filter framework. Our key idea is to use an OCR value net- work (OCR-VN) that predicts the optimal control-theoretic safety value function for new failure regions and dynamic uncertainty during deployment time. Specifically, the OCR-VN facilitates rapid safety adaptation through two key components: a LiDAR-based input that allows the dynamic construction of safe regions in light of new obstacles and a disturbance estimation module that accounts for dynamics uncertainty in the wild. The predicted safety value function is used to construct an adaptive safety filter that overrides the nominal quadruped controller when necessary to maintain safety. Through simulation studies and hardware experiments on a UnitreeGo1quadrupedinagileplanarnavigationtasks,wedemon- strate that the proposed framework can automatically safeguard a wide range of hierarchical quadruped controllers, adapts to novel environments, and is robust to unmodeled dynamics without a priori access to the controllers or environments—hence, “One Filter to Deploy Them All.”