ADV-0: Closed-Loop Min-Max Adversarial Training for Long-Tail Robustness in Autonomous Driving

TL;DR

ADV-0 introduces a closed-loop min-max adversarial training framework for autonomous driving, aligning attack and defense objectives to improve robustness against rare, safety-critical long-tail scenarios.

Contribution

It proposes a novel zero-sum Markov game approach with iterative preference learning, ensuring convergence to Nash Equilibrium and better capturing evolving failure modes.

Findings

Effectively exposes safety-critical failures.

Enhances policy robustness against unseen long-tail risks.

Converges to Nash Equilibrium with theoretical guarantees.

Abstract

Deploying autonomous driving systems requires robustness against long-tail scenarios that are rare but safety-critical. While adversarial training offers a promising solution, existing methods typically decouple scenario generation from policy optimization and rely on heuristic surrogates. This leads to objective misalignment and fails to capture the shifting failure modes of evolving policies. This paper presents ADV-0, a closed-loop min-max optimization framework that treats the interaction between driving policy (defender) and adversarial agent (attacker) as a zero-sum Markov game. By aligning the attacker's utility directly with the defender's objective, we reveal the optimal adversary distribution. To make this tractable, we cast dynamic adversary evolution as iterative preference learning, efficiently approximating this optimum and offering an algorithm-agnostic solution to the game. Theoretically, ADV-0 converges to a Nash Equilibrium and maximizes a certified lower bound on real-world performance. Experiments indicate that it effectively exposes diverse safety-critical failures and greatly enhances the generalizability of both learned policies and motion planners against unseen long-tail risks.