A unified world-action model that jointly generates depth maps, future videos, and actions.

LLM-based multimodal understanding with generative world models for end-to-end closed-loop autonomous driving.

A coupling selection algorithm that better align with the learned model.

The first comprehensive benchmark for generative driving world models.

A diffusion-based framework for video face swapping with superior identity preservation and temporal consistency.

A general and scalable self-supervised pretraining framework for motion prediction, designed to be both model-agnostic and dataset-agnostic.

A scenario-adaptive multi-round refinement strategy that boosts trajectory prediction with minimal extra compute.