Testing Rare Downstream Safety Violations via Upstream Adaptive Sampling of Perception Error Models

TL;DR

This paper introduces a method combining perception error models with adaptive importance sampling to efficiently estimate rare safety violations in perception-based control systems within simulation environments.

Contribution

It presents a novel approach that improves the efficiency of testing rare failures in black-box perception systems using adaptive sampling and perception error models.

Findings

Accurately estimates rare failure probabilities with fewer simulations.

Demonstrates effectiveness on an autonomous braking system with RGB obstacle detection.

Shows safety metric choice impacts learning of failure sampling distributions.

Abstract

Testing black-box perceptual-control systems in simulation faces two difficulties. Firstly, perceptual inputs in simulation lack the fidelity of real-world sensor inputs. Secondly, for a reasonably accurate perception system, encountering a rare failure trajectory may require running infeasibly many simulations. This paper combines perception error models -- surrogates for a sensor-based detection system -- with state-dependent adaptive importance sampling. This allows us to efficiently assess the rare failure probabilities for real-world perceptual control systems within simulation. Our experiments with an autonomous braking system equipped with an RGB obstacle-detector show that our method can calculate accurate failure probabilities with an inexpensive number of simulations. Further, we show how choice of safety metric can influence the process of learning proposal distributions capable of reliably sampling high-probability failures.