# BridgeDrive Model Card ## Model Details Diffusion-based planners excel in autonomous driving by capturing multi-modal behaviors, but guiding them for safe, closed-loop planning remains challenging. Existing methods rely on anchor trajectories but suffer from a truncated diffusion process that breaks theoretical consistency. We introduce BridgeDrive, an anchor-guided diffusion bridge policy that directly transforms coarse anchors into refined plans while preserving consistency between forward and reverse processes. BridgeDrive supports efficient ODE solvers for real-time deployment. We achieve state-of-the-art performance on the Bench2Drive closed-loop evaluation benchmark, improving the success rate by 7.72% and 2.45% over prior arts with PDM-Lite and LEAD datasets, respectively. - **Developed by:** [Bosch] (https://www.bosch.de/), - **Model type:** An end-to-end autonomous driving model based on the diffusion bridge policy for closed-loop settings. ### Model Sources - **Repository:** https://github.com/shuliu-ethz/BridgeDrive - **Paper:** https://openreview.net/pdf?id=dJKhjK4zpp ## Uses The primary use of BridgeDrive is for the end-to-end autonomous driving in closed-loop settings. ## Citation Information @inproceedings{ liu2026bridgedrive, title={BridgeDrive: Diffusion Bridge Policy for Closed-Loop Trajectory Planning in Autonomous Driving}, author={Shu Liu and Wenlin Chen and Weihao Li and Zheng Wang and Lijin Yang and Jianing Huang and Yipin Zhang and Zhongzhan Huang and Ze Cheng and Hao Yang}, booktitle={The Fourteenth International Conference on Learning Representations}, year={2026}, url={https://arxiv.org/abs/2509.23589} }