BirdDreams: Joint Perspective and BEV Video Generation for a Plausible, Reactive Driving World Model

Researchers introduce BirdDreams, an autoregressive driving world model that jointly generates perspective video and schematic bird's-eye-view (BEV) video from a single camera image. The predicted BEV serves as a geometric interface for ego control, allowing the model to follow input ego trajectories more faithfully than prior approaches. Evaluations on PhysicalAI-AV and nuScenes show that the model generates plausible ego motion and realistic surrounding-vehicle behavior. Open article →