Evaluation as Evolution: Transforming Adversarial Diffusion into Closed-Loop Curricula for Autonomous Vehicles

TL;DR

This paper presents a novel closed-loop evaluation framework for autonomous vehicles that adaptively generates adversarial scenarios to improve safety testing and policy robustness.

Contribution

It introduces Evaluation as Evolution ($E^2$), transforming static adversarial testing into an adaptive, evolutionary curriculum using transport-regularized control over a learned SDE prior.

Findings

$E^2$ improves collision failure discovery by 9.01% on nuScenes.

$E^2$ achieves up to 21.43% improvement on nuPlan.

Recycling boundary cases for policy fine-tuning enhances robustness.

Abstract

Autonomous vehicles in interactive traffic environments are often limited by the scarcity of safety-critical tail events in static datasets, which biases learned policies toward average-case behaviors and reduces robustness. Existing evaluation methods attempt to address this through adversarial stress testing, but are predominantly open-loop and post-hoc, making it difficult to incorporate discovered failures back into the training process. We introduce Evaluation as Evolution ($E^2$), a closed-loop framework that transforms adversarial generation from a static validation step into an adaptive evolutionary curriculum. Specifically, $E^2$ formulates adversarial scenario synthesis as transport-regularized sparse control over a learned reverse-time SDE prior. To make this high-dimensional generation tractable, we utilize topology-driven support selection to identify critical interacting agents, and introduce Topological Anchoring to stabilize the process. This approach enables the targeted discovery of failure cases while strictly constraining deviations from realistic data distributions. Empirically, $E^2$ improves collision failure discovery by 9.01% on the nuScenes dataset and up to 21.43% on the nuPlan dataset over the strongest baselines, while maintaining low invalidity and high realism. It further yields substantial robustness gains when the resulting boundary cases are recycled for closed-loop policy fine-tuning.