News

• Sep. 7th, 2025: We reorganize code for better readability. Code & Models are released.
• Aug. 08, 2025: We release the DistillDrive paper on arXiv.
• Jun. 26, 2025: DistillDrive is accepted to ICCV 2025!
• Oct. 26, 2025: DistillDrive is published at ICCV 2025 on CVF Open Access Paper.

Introduction

We introduce DistillDrive, an end-to-end knowledge distillation-based autonomous driving model that leverages diversified instance imitation to enhance multi-mode motion feature learning

• We propose a distillation architecture for multi-mode instance supervision in end-to-end planning, tackling single-target imitation learning limitations.
• We introduce reinforcement learning-based state optimization to enhance state-to-decision space understanding and mitigate ego motion state leakage.
• To address missing motion-guided attributes, we use a generative model for distribution-wise interaction between expert trajectories and instance features.
• We conduct open- and closed-loop planning experiments on the nuScenes and NAVSIM datasets, achieving a 50% reduction in collision rate and a 3-point increase in both EP and PDMS over the baseline.

Overview

The Overview of our proposed DistillDrive. Initially, we train a teacher model with scene-structured annotation data, integrating reinforcement and imitation learning to enhance multi-mode planning. Subsequently, we constructed an end-to-end student model and used a generative model to implement motion-oriented distribution interactions in latent space. Ultimately, multi-stage knowledge distillation and multi-mode supervision synergistically enhance the planning diversity and safety margins of autonomous driving models.

• Open-loop planning results on nuScenes.

Model	Input	1s	2s	3s	Avg.	1s	2s	3s	Avg.	FPS ↑
		L2 (m) ↓				Collision (%) ↓
FF	LiDAR	0.55	1.20	2.54	1.43	0.06	0.17	1.07	0.43	-
EO	LiDAR	0.67	1.36	2.78	1.60	0.04	0.09	0.88	0.33	-
ST-P3	Camera	1.33	2.11	2.90	2.11	0.23	0.63	1.27	0.71	1.6
UniAD	Camera	0.45	0.70	1.04	0.73	0.62	0.58	0.63	0.61	1.8
VAD	Camera	0.41	0.70	1.05	0.72	0.03	0.19	0.43	0.21	4.5
SparseDrive	Camera	0.31	0.60	1.00	0.63	0.01	0.08	0.30	0.13	6.5
DistillDrive (Teacher)	Annotation	0.27	0.51	0.82	0.53	0.01	0.04	0.10	0.05	31.6
DistillDrive (Student)	Camera	0.28	0.54	0.83	0.57	0.00	0.03	0.17	0.06	6.0

• Close-loop planning results on NAVSIM.

Model	Input	Backbone	NC ↑	DAC ↑	TTC	Comf. ↑	EP ↑	PDMS ↑
Const Velocity	-	-	69.0	57.8	58.0	100	19.4	20.6
Ego Status MLP	-	-	93.0	77.3	83.6	100	62.8	65.6
UniAD	C	ResNet-34	97.8	91.9	92.2	100	78.8	83.4
VADV2	C	ResNet-34	92.2	89.1	91.6	100	76.0	80.9
Transfuser	C & L	ResNet-34	97.8	92.3	92.9	100	78.6	83.5
Hydra-MDP	C & L	ResNet-34	97.9	91.7	92.9	100	77.6	83.0
DistillDrive (Teacher)	GT	-	97.5	96.0	92.8	100	81.0	86.5
DistillDrive (Student)	C & L	ResNet-34	98.1	94.6	93.6	100	81.0	86.2

Modle Name	Stage	Weight Download
distilldrive_stage0_distribution.pth	stage0	huggingface
distilldrive_stage0_label.pth	stage0	huggingface
distilldrive_stage1_adamax.pth	stage1	huggingface
distilldrive_stage1_soap.pth	stage1	huggingface
distilldrive_stage2_distribution.pth	stage2	huggingface
distilldrive_stage2_label.pth	stage2	huggingface

Getting Started

• Preparation of DistillDrive environment
• Getting started from nuScenes environment preparation
• Training and Evaluation

Video Demo on Real-world Application

https://huggingface.co/RuiYuStudying/DistillDrive/blob/main/demo.mp4

Acknowledgement

• SparseDrive
• Sparse4D
• UniAD
• VAD
• StreamPETR
• StreamMapNet
• mmdet3d

Citation

If you find DistillDrive is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@inproceedings{yu2025distilldrive,
  title={Distilldrive: End-to-end multi-mode autonomous driving distillation by isomorphic hetero-source planning model},
  author={Yu, Rui and Zhang, Xianghang and Zhao, Runkai and Yan, Huaicheng and Wang, Meng},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={26188--26197},
  year={2025}
}