Sampling-Based Safety Filter with Probabilistic Restrictiveness Guarantee

Researchers Junyoung Park, Hyeontae Sung, and Heejin Ahn from the field of systems and control have introduced a novel sampling-based safety filter designed to provide formal safety guarantees for autonomous systems. The filter operates by evaluating the safety of arbitrary nominal control inputs at each timestep, using control sequence samples generated via Stein Variational Model Predictive Control (SV-MPC). This approach approximates a safety-conditioned posterior distribution over control sequences, enabling the filter to effectively capture multimodal safe regions in complex, non-convex environments where traditional methods may struggle.

The filter ensures safety by overriding the nominal input when all sampled control sequence candidates are deemed unsafe, providing a robust safety mechanism. By leveraging the scenario approach, the method offers a probabilistic guarantee on its restrictiveness, meaning it can ensure safety without being overly conservative. The researchers validated their approach through collision avoidance tasks in both single- and multi-vehicle settings, demonstrating its efficacy in navigating cluttered environments where nominal controllers may fail. This work addresses a critical challenge in autonomous systems, offering a practical solution for ensuring safety with formal guarantees.