TL;DR of the paper, generated automatically.

BirdDreams is a driving world model that generates perspective video and a bird's-eye-view video together.

• Input: a single camera image, with no HD map or vehicle-state inputs needed.
• Ego control: the predicted BEV is moved along an input trajectory to steer the ego car.
• Results: the paper says it follows ego trajectories more faithfully than prior models on PhysicalAI-AV and nuScenes.