arXiv 2411.15139

DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving

By Bencheng Liao, Shaoyu Chen, et al.

Published 2024-11-22

Citation lineage

Review the prior work and downstream research connected to this paper.

Recently, the diffusion model has emerged as a powerful generative technique for robotic policy learning, capable of modeling multi-mode action distributions. Leveraging its capability for end-to-end autonomous driving is a promising direction. However, the numerous denoising steps in the robotic diffusion policy and the more dynamic, open-world nature of traffic scenes pose substantial challenges for generating div…

View the original paper on arXiv