Enregistré dans:
Détails bibliographiques
Auteurs principaux: Lian, Weitong, Tang, Zecong, Li, Haoran, Gao, Tianjian, Wang, Yifei, Wang, Zixu, Meng, Lingyi, Ru, Tengju, Cui, Zhejun, Zhu, Yichen, Cao, Hangshuo, Kang, Qi, Chen, Tianxing, Qin, Yusen, Wang, Kaixuan, Zhang, Yu
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2601.21288
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917231084961792
author Lian, Weitong
Tang, Zecong
Li, Haoran
Gao, Tianjian
Wang, Yifei
Wang, Zixu
Meng, Lingyi
Ru, Tengju
Cui, Zhejun
Zhu, Yichen
Cao, Hangshuo
Kang, Qi
Chen, Tianxing
Qin, Yusen
Wang, Kaixuan
Zhang, Yu
author_facet Lian, Weitong
Tang, Zecong
Li, Haoran
Gao, Tianjian
Wang, Yifei
Wang, Zixu
Meng, Lingyi
Ru, Tengju
Cui, Zhejun
Zhu, Yichen
Cao, Hangshuo
Kang, Qi
Chen, Tianxing
Qin, Yusen
Wang, Kaixuan
Zhang, Yu
contents Autonomous driving is an important and safety-critical task, and recent advances in LLMs/VLMs have opened new possibilities for reasoning and planning in this domain. However, large models demand substantial GPU memory and exhibit high inference latency, while conventional supervised fine-tuning (SFT) often struggles to bridge the capability gaps of small models. To address these limitations, we propose Drive-KD, a framework that decomposes autonomous driving into a "perception-reasoning-planning" triad and transfers these capabilities via knowledge distillation. We identify layer-specific attention as the distillation signal to construct capability-specific single-teacher models that outperform baselines. Moreover, we unify these single-teacher settings into a multi-teacher distillation framework and introduce asymmetric gradient projection to mitigate cross-capability gradient conflicts. Extensive evaluations validate the generalization of our method across diverse model families and scales. Experiments show that our distilled InternVL3-1B model, with ~42 times less GPU memory and ~11.4 times higher throughput, achieves better overall performance than the pretrained 78B model from the same family on DriveBench, and surpasses GPT-5.1 on the planning dimension, providing insights toward efficient autonomous driving VLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2601_21288
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Drive-KD: Multi-Teacher Distillation for VLMs in Autonomous Driving
Lian, Weitong
Tang, Zecong
Li, Haoran
Gao, Tianjian
Wang, Yifei
Wang, Zixu
Meng, Lingyi
Ru, Tengju
Cui, Zhejun
Zhu, Yichen
Cao, Hangshuo
Kang, Qi
Chen, Tianxing
Qin, Yusen
Wang, Kaixuan
Zhang, Yu
Artificial Intelligence
Computer Vision and Pattern Recognition
Autonomous driving is an important and safety-critical task, and recent advances in LLMs/VLMs have opened new possibilities for reasoning and planning in this domain. However, large models demand substantial GPU memory and exhibit high inference latency, while conventional supervised fine-tuning (SFT) often struggles to bridge the capability gaps of small models. To address these limitations, we propose Drive-KD, a framework that decomposes autonomous driving into a "perception-reasoning-planning" triad and transfers these capabilities via knowledge distillation. We identify layer-specific attention as the distillation signal to construct capability-specific single-teacher models that outperform baselines. Moreover, we unify these single-teacher settings into a multi-teacher distillation framework and introduce asymmetric gradient projection to mitigate cross-capability gradient conflicts. Extensive evaluations validate the generalization of our method across diverse model families and scales. Experiments show that our distilled InternVL3-1B model, with ~42 times less GPU memory and ~11.4 times higher throughput, achieves better overall performance than the pretrained 78B model from the same family on DriveBench, and surpasses GPT-5.1 on the planning dimension, providing insights toward efficient autonomous driving VLMs.
title Drive-KD: Multi-Teacher Distillation for VLMs in Autonomous Driving
topic Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.21288