Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Junli, Hua, Zhihua, Liu, Xueyi, Xing, Zebin, Tian, Haochen, Ma, Kun, Ye, Hangjun, Chen, Guang, Chen, Long, Zhang, Qichao
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918511600730112
author Wang, Junli
Hua, Zhihua
Liu, Xueyi
Xing, Zebin
Tian, Haochen
Ma, Kun
Ye, Hangjun
Chen, Guang
Chen, Long
Zhang, Qichao
author_facet Wang, Junli
Hua, Zhihua
Liu, Xueyi
Xing, Zebin
Tian, Haochen
Ma, Kun
Ye, Hangjun
Chen, Guang
Chen, Long
Zhang, Qichao
contents Existing imitation learning methods for end-to-end autonomous driving predominantly learn from successful demonstrations by minimizing geometric deviations from expert trajectories. This paradigm implicitly assumes that spatial proximity implies behavioral safety, leading to a critical objective mismatch: trajectories with nearly identical imitation losses may exhibit drastically different safety outcomes, where one remains recoverable while the other results in collision. To address this limitation, we propose BeyondDrive, a failure-aware imitation learning framework that jointly learns from successful and failed driving behaviors. First, we introduce a flow matching-based negative trajectory generator that synthesizes safety-critical yet expert-proximate trajectories, enabling explicit modeling of safety asymmetry. Second, we develop a diversity-aware sampling strategy that mitigates mode collapse and improves coverage of diverse failure modes during negative trajectory generation. Third, we propose a Repulsive Distance Loss that simultaneously attracts predictions toward expert demonstrations while repelling them from hard negative trajectories, thereby establishing discriminative safety boundaries in trajectory space. Applied to the uni-modal baseline Latent TransFuser, BeyondDrive achieves 89.7 PDMS on the NAVSIMv1 closed-loop benchmark, outperforming prior state-of-the-art methods. Moreover, BeyondDrive generalizes effectively across different autonomous driving architectures, including multi-modal planners, and further demonstrates strong zero-shot transferability on the HUGSIM benchmark.
format Preprint
id arxiv_https___arxiv_org_abs_2605_19771
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives
Wang, Junli
Hua, Zhihua
Liu, Xueyi
Xing, Zebin
Tian, Haochen
Ma, Kun
Ye, Hangjun
Chen, Guang
Chen, Long
Zhang, Qichao
Robotics
Computer Vision and Pattern Recognition
Existing imitation learning methods for end-to-end autonomous driving predominantly learn from successful demonstrations by minimizing geometric deviations from expert trajectories. This paradigm implicitly assumes that spatial proximity implies behavioral safety, leading to a critical objective mismatch: trajectories with nearly identical imitation losses may exhibit drastically different safety outcomes, where one remains recoverable while the other results in collision. To address this limitation, we propose BeyondDrive, a failure-aware imitation learning framework that jointly learns from successful and failed driving behaviors. First, we introduce a flow matching-based negative trajectory generator that synthesizes safety-critical yet expert-proximate trajectories, enabling explicit modeling of safety asymmetry. Second, we develop a diversity-aware sampling strategy that mitigates mode collapse and improves coverage of diverse failure modes during negative trajectory generation. Third, we propose a Repulsive Distance Loss that simultaneously attracts predictions toward expert demonstrations while repelling them from hard negative trajectories, thereby establishing discriminative safety boundaries in trajectory space. Applied to the uni-modal baseline Latent TransFuser, BeyondDrive achieves 89.7 PDMS on the NAVSIMv1 closed-loop benchmark, outperforming prior state-of-the-art methods. Moreover, BeyondDrive generalizes effectively across different autonomous driving architectures, including multi-modal planners, and further demonstrates strong zero-shot transferability on the HUGSIM benchmark.
title Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives
topic Robotics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.19771