Proposal-Conditioned Latent Diffusion for Closed-Loop Traffic Scenario Generation
For autonomous vehicle planning and simulation, this work addresses the need for efficient, controllable, and realistic multi-agent traffic scenario generation in time-constrained replanning loops.
The paper tackles closed-loop traffic simulation by proposing a diffusion-based scenario generation framework conditioned on instance-centric scene context and multimodal proposal priors. It achieves a favorable balance among realism, safety, and controllability on the Waymo Open Motion Dataset, with improved sampling efficiency and reduced per-step runtime.
Closed-loop traffic simulation remains challenging because it must generate interactive multi-agent behaviors that are scene-consistent and controllable throughout rollout. Prior diffusion-based approaches achieve strong realism, but their computational cost can hinder deployment in time-constrained replanning loops for autonomous vehicle planning and simulation. We present a diffusion-based scenario generation framework conditioned on instance-centric scene context and multimodal proposal priors, with optional test-time guidance for shaping safety-critical behaviors. A compact action-latent representation and proposal-based initialization improve sampling efficiency and reduce per-step runtime without retraining. Experiments on the Waymo Open Motion Dataset demonstrate a favorable balance among realism, safety, and controllability across diverse interactive scenarios, while showing that test-time guidance enables systematic trade-offs among competing objectives.