The research article presents Drive-WM, a world model for autonomous driving. This model is designed to predict future events and evaluate potential risks, thus helping autonomous vehicles to better plan their actions. This enhances safety and efficiency on the road. The model uses a joint spatial-temporal modeling facilitated by view factorization to generate high-fidelity multiview videos in driving scenes. Evaluation on real-world driving datasets showed that the method could generate high-quality, consistent, and controllable multiview videos, opening up possibilities for real-world simulations and safe planning.
Publication date: 29 Nov 2023
Project Page: https://drive-wm.github.io
Paper: https://arxiv.org/pdf/2311.17918
