This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
AnchDrive is an end-to-end autonomous driving framework that effectively bootstraps diffusion policies to address the high computational costs of existing generative models in the autonomous driving field. Instead of denoising from pure noise, the framework initializes the planner using hybrid trajectory anchors derived from a static vocabulary of general driving prior knowledge and a set of dynamic, context-aware trajectories. The dynamic trajectories are decoded in real time by a Transformer that handles both dense and sparse perceptual features. The diffusion model learns to refine these anchors by predicting the distribution of trajectory offsets. This anchor-based bootstrap design efficiently generates diverse and high-quality trajectories. Experimental results on the NAVSIM benchmark demonstrate that AnchDrive achieves new state-of-the-art performance and demonstrates strong generalization capabilities.
Takeaways, Limitations
•
Takeaways:
◦
Bootstrapping the diffusion policy to improve computational efficiency.
◦
Improve the quality and diversity of trajectory generation by utilizing hybrid anchors that combine general driving prior knowledge with dynamic trajectories.
◦
Achieved new best performance on the NAVSIM benchmark, demonstrating strong generalization capabilities.
•
Limitations:
◦
There is no direct mention of Limitations in the paper.
