NVIDIA (Junyoung Seo, Jaewoo Seo, Seungryong Kim, Laura Leal-Taixé, Xiuming Zhang, Seung Wook Kim) · Driving World Model

BirdDreams

BirdDreams is an autoregressive driving world model that jointly generates perspective video and a schematic bird's-eye-view video using two joint-attention video diffusion transformers. The BEV stream makes vehicle layout part of the prediction, and ego control is applied by transforming the BEV frame along an input trajectory. On PhysicalAI-AV and nuScenes, the authors report that it follows ego trajectories more faithfully than prior driving world models, and human evaluation favours it over a perspective-only baseline for surrounding-vehicle behaviour.