Three‑dimensional modeling for autonomous driving must continuously estimate road geometry, obstacle occupancy, traffic participant behavior and observational uncertainty while operating under limited computational resources.
Existing reconstruction methods can synthesize realistic novel views, yet visual fidelity alone does not capture the geometric reliability required for safe driving. Generative world models can expand scenarios and predict future observations, but they may introduce content that lacks measurement support.
We introduce DriveWorld3D, a framework that combines temporal video learning, multi‑sensor measurement constraints and offline generative simulation. The online component maintains a unified world state comprising road surfaces, volumetric occupancy, dynamic entities and road relations, while separately tracking physical occupancy, observability and evidence provenance.
Action‑conditioned future prediction and risk‑aware planning link this representation to driving decisions. The offline component reconstructs recorded scenes using neural radiance fields or 3D Gaussian splatting, then expands long‑tail conditions with generative models, restricting generated content to scenario augmentation rather than geometric ground truth.
We specify data partitioning, sensor ablations, fault injection, closed‑loop testing and real‑to‑simulation consistency assessment. The central hypothesis is that explicit evidence management can reduce dangerous false free‑space predictions and occlusion risk within a matched computational budget, while preserving acceptable driving efficiency. All performance claims remain subject to experimental validation.
Review