Driving in Corner Case: A Real-World Adversarial Closed-Loop Evaluation Platform for End-to-End Autonomous Driving

2025-12-17

AI Summary

End-to-end autonomous driving systems struggle to efficiently detect safety-critical corner cases in real-world deployment. Method: This paper introduces the first closed-loop adversarial evaluation platform designed for real-vehicle deployment. It integrates a flow-matching–based realistic image generator with reinforcement learning–driven adversarial traffic policies to enable dynamic, interactive, and closed-loop generation of long-tail scenarios on real-road backgrounds—overcoming the limitations of conventional simulation-centric approaches. The method supports high-fidelity image synthesis and online closed-loop integration with end-to-end models for evaluation. Contribution/Results: Evaluated on state-of-the-art models including UniAD and VAD, the platform significantly enhances corner-case elicitation capability and safety issue detection rate. It establishes a scalable, reproducible paradigm for robustness verification of end-to-end autonomous driving systems.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Security and Privacy: Large-scale security measurements

Abstract

Safety-critical corner cases, difficult to collect in the real world, are crucial for evaluating end-to-end autonomous driving. Adversarial interaction is an effective method to generate such safety-critical corner cases. While existing adversarial evaluation methods are built for models operating in simplified simulation environments, adversarial evaluation for real-world end-to-end autonomous driving has been little explored. To address this challenge, we propose a closed-loop evaluation platform for end-to-end autonomous driving, which can generate adversarial interactions in real-world scenes. In our platform, the real-world image generator cooperates with an adversarial traffic policy to evaluate various end-to-end models trained on real-world data. The generator, based on flow matching, efficiently and stably generates real-world images according to the traffic environment information. The efficient adversarial surrounding vehicle policy is designed to model challenging interactions and create corner cases that current autonomous driving systems struggle to handle. Experimental results demonstrate that the platform can generate realistic driving images efficiently. Through evaluating the end-to-end models such as UniAD and VAD, we demonstrate that based on the adversarial policy, our platform evaluates the performance degradation of the tested model in corner cases. This result indicates that this platform can effectively detect the model's potential issues, which will facilitate the safety and robustness of end-to-end autonomous driving.

Problem

• Generates adversarial corner cases for real-world autonomous driving evaluation
• Evaluates end-to-end models' performance degradation in safety-critical scenarios
• Detects potential issues to improve autonomous driving safety and robustness

Innovation

• Closed-loop platform generates adversarial interactions in real-world scenes
• Flow matching generator creates realistic driving images efficiently
• Adversarial traffic policy models challenging interactions to create corner cases