DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915506774081536 |
|---|---|
| author | Shang, Shuyao Chen, Yuntao Wang, Yuqi Li, Yingyan Zhang, Zhaoxiang |
| author_facet | Shang, Shuyao Chen, Yuntao Wang, Yuqi Li, Yingyan Zhang, Zhaoxiang |
| contents | End-to-end autonomous driving has substantially progressed by directly predicting future trajectories from raw perception inputs, which bypasses traditional modular pipelines. However, mainstream methods trained via imitation learning suffer from critical safety limitations, as they fail to distinguish between trajectories that appear human-like but are potentially unsafe. Some recent approaches attempt to address this by regressing multiple rule-driven scores but decoupling supervision from policy optimization, resulting in suboptimal performance. To tackle these challenges, we propose DriveDPO, a Safety Direct Preference Optimization Policy Learning framework. First, we distill a unified policy distribution from human imitation similarity and rule-based safety scores for direct policy optimization. Further, we introduce an iterative Direct Preference Optimization stage formulated as trajectory-level preference alignment. Extensive experiments on the NAVSIM benchmark demonstrate that DriveDPO achieves a new state-of-the-art PDMS of 90.0. Furthermore, qualitative results across diverse challenging scenarios highlight DriveDPO's ability to produce safer and more reliable driving behaviors. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_17940 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving Shang, Shuyao Chen, Yuntao Wang, Yuqi Li, Yingyan Zhang, Zhaoxiang Robotics Computer Vision and Pattern Recognition End-to-end autonomous driving has substantially progressed by directly predicting future trajectories from raw perception inputs, which bypasses traditional modular pipelines. However, mainstream methods trained via imitation learning suffer from critical safety limitations, as they fail to distinguish between trajectories that appear human-like but are potentially unsafe. Some recent approaches attempt to address this by regressing multiple rule-driven scores but decoupling supervision from policy optimization, resulting in suboptimal performance. To tackle these challenges, we propose DriveDPO, a Safety Direct Preference Optimization Policy Learning framework. First, we distill a unified policy distribution from human imitation similarity and rule-based safety scores for direct policy optimization. Further, we introduce an iterative Direct Preference Optimization stage formulated as trajectory-level preference alignment. Extensive experiments on the NAVSIM benchmark demonstrate that DriveDPO achieves a new state-of-the-art PDMS of 90.0. Furthermore, qualitative results across diverse challenging scenarios highlight DriveDPO's ability to produce safer and more reliable driving behaviors. |
| title | DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving |
| topic | Robotics Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.17940 |