RiskFlow: Fast and Faithful Safety-Critical Traffic Scenario Generation

2026-06-04

AI Summary

This work addresses the high computational cost and error accumulation in existing diffusion models when generating safety-critical traffic scenarios, which often lead to distorted trajectories in long-horizon closed-loop simulation. The authors propose RiskFlow, a framework that formulates multi-agent trajectory generation as an optimal transport problem in action space. By learning a mean velocity field and leveraging Jacobian-Vector Products (JVPs), RiskFlow enables a single-step forward mapping that directly transforms Gaussian action sequences into acceleration and yaw-rate commands, while incorporating output-space guidance and vehicle dynamics constraints. Eliminating iterative denoising, the method achieves significantly faster inference and enhanced trajectory realism on the nuScenes dataset, without compromising its capability to generate safety-critical scenarios, thereby effectively balancing realism and adversarial fidelity.

Abstract

Safety-critical traffic scenario generation is essential for evaluating autonomous driving systems under rare but high-risk interactions. Existing diffusion-based methods offer strong controllability in closed-loop generation, but their iterative denoising process is computationally expensive and may accumulate sampling and guidance errors over long rollouts, causing unrealistic motion artifacts such as jitter, abnormal acceleration, and off-road behavior. To address these issues, we propose RiskFlow, a closed-loop safety-critical multi-agent traffic generation framework that formulates future trajectory generation as transport in the action space. Instead of relying on iterative denoising, RiskFlow learns an average velocity field over a finite interval to transform Gaussian action sequences into future acceleration and yaw-rate commands with a single forward pass, using a JVP-based objective for efficient and stable training. At test time, RiskFlow applies output-space guidance to the generated actions, steering selected critical agents toward risky interactions while regularizing off-road behavior, and reconstructs physically feasible trajectories through vehicle dynamics. Experiments on nuScenes with tbsim closed-loop evaluation show that RiskFlow achieves a strong adversariality-realism trade-off across multi-agent and long-horizon settings. Compared with representative baselines, RiskFlow consistently improves realism while maintaining competitive safety-critical generation capability, and substantially reduces inference time for evaluation.

Problem

• safety-critical scenario generation
• diffusion-based methods
• motion artifacts
• computational efficiency
• trajectory realism

Innovation

• trajectory generation
• action-space transport
• closed-loop simulation
• output-space guidance
• vehicle dynamics