DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving

drivetransformer_bench2drive.mp4

Official implementation of paper DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving. Xiaosong Jia, Junqi You, Zhiyuan Zhang, Junchi Yan. ICLR 2025

DriveTransformer offers a unified, parallel, and synergistic approach to end-to-end autonomous driving, facilitating easier training and scalability. The framework is composed of three unified operations: task self-attention, sensor cross-attention, temporal cross-attention and has three key features:

• Task Parallelism: All agent, map, and planning queries direct interact with each other at each block.
• Sparse Representation: Task queries direct interact with raw sensor features.
• Streaming Processing: Task queries are stored and passed as history information.

Getting Started

• Download and installation
• Data preprocessing
• Training and Evaluation

Model and Result

Model	Driving Score	Success Rate(%)	Efficiency	Comfortness	Latency	Config	Download
DriveTransformer-Large	63.46	35.01	100.64	20.78	211.7ms	config	Google Drive/Baidu Cloud

Citation

@inproceedings{jia2025drivetransformer,
  title={DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving},
  author={Xiaosong Jia and Junqi You and Zhiyuan Zhang and Junchi Yan},
  booktitle={International Conference on Learning Representations (ICLR)},
  year={2025}
}