arxivcs.RO2026-07-20
GeoWorldAD: Geometry World Action Model for Autonomous Driving
Songyan Zhang, Jinyuan Tian, Hanbing Li, Daqi Liu, Hao Chen, Wenhui Huang, et al.
Autonomous driving requires both safe and efficient planning decisions in dynamic 3D environments. Although recent Vision/Video-Action models learn policies directly from visual observations and scale well with advances in vision transformers and large-scale training data, they o…