This paper proposes CrashDiffuser, a closed-loop VLM-guided diffusion framework that decouples semantic collision reasoning from continuous trajectory synthesis through a hierarchical collision-intent interface derived from the requested target contact region.
Abstract
Generating safety-critical scenarios is essential for evaluating autonomous driving systems. However, existing generators primarily focus on inducing collisions and offer limited control over where contact occurs on the target vehicle. In this paper, we study fine-grained safety-critical scenario generation, where success requires both a target collision and a specified head, rear, or side contact region. We propose CrashDiffuser, a closed-loop VLM-guided diffusion framework that decouples semantic collision reasoning from continuous trajectory synthesis through a hierarchical collision-intent interface derived from the requested target contact region. At initialization, the VLM extracts reusable scene-level context; at each replanning step, it predicts a structured action tuple describing speed change, turning behavior, and collision stage. This intent conditions a diffusion model to generate executable adversarial trajectories, while collision-guided sampling, candidate selection, and short-horizon replanning adapt generation to the target vehicle's evolving behavior. On WOMD-derived closed-loop scenarios, CrashDiffuser achieves a target-collision rate of 50.33% in a single attempt and 67.98% after three attempts, together with a contact-region control success rate of 40.05% and competitive trajectory naturalness. Component ablations further support the proposed design.
The generation of safety-critical traffic scenarios is essential for training and evaluating autonomous vehicles. Prior approaches typically perturb the trajectories of existing agents in a traffic scenario using simplified adversarial objectives to induce safety-critical interactions, which can limit the plausibility...
Safety-critical driving scenario generation has largely focused on manipulating the behavior of surrounding agents while starting from an initial scene from driving data. This assumption can limit the space of discoverable failures, since driving data can provide little opportunity for meaningful interaction. For examp...
Yin Wu, Jia-Rong Wei, Carl Esselborn et al.· 0 citations
MomADv2, a reliable state-space memory framework for long-horizon end-to-end autonomous driving, introduces a Selective State-Space Planning Memory Query Module, which filters historical planning queries based on temporal continuity and command consistency, and models the evolution of planning intentions through a sele...
Zi-Ying Song, Sheng-Kai Zhang, Lin Liu et al.· 0 citations
DriveReferee is introduced, which uses a learned geometry readout to predict the scene representation from camera observations and executes the geometric safety rule directly rather than learning it, and which matches or outperforms all learned-verifier and heuristic baselines.
Feng-Cheng Yu, Dhruv Parikh, Jun-Jie Ye et al.· 0 citations
Adversarial time-to-collision (aTTC), a risk metric that quantifies, for a given agent, how quickly any surrounding agent could reach it assuming adversarial intent, is introduced.
B. B. Sorensen, Mitchell Black, Erfaun Noorani et al.· 0 citations
Vision-based collision warning systems are increasingly recognised as a promising countermeasure against traffic collisions. However, their development is constrained by the limited explainability of deep learning-based collision prediction models. Vision-language models (VLMs), with inherent self-explainability, o...
Ruici Zhang, Yuxiang Feng, Jose Escribano et al.· Communications in Transporta...· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduJul 30, 2026
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks.