TrajDiff: End-to-end Autonomous Driving without Perception Annotation

Xingtai Gui¹, Jianbo Zhao², Wencheng Han¹, Jikai Wang¹, Jiahao Gong³, Feiyang Tan³, Cheng-zhong Xu¹, Jianbing Shen¹

¹SKL-IOTSC, CIS, University of Macau
²University of Science and Technology of China
³Mach Drive

News

• Dec. 22th, 2025: We release initial version of TrajDiff evaluation and the corresponding checkpoint.
• Nov. 30th, 2025: We release the TrajDiff on arxiv.

Todos

• Release the training and visualization scripts.

Abstract

End-to-end autonomous driving systems directly generate driving policies from raw sensor inputs. While these systems can extract effective environmental features for planning, relying on auxiliary perception tasks, developing perception annotation-free planning paradigms has become increasingly critical due to the high cost of manual perception annotation. In this work, we propose TrajDiff, a Trajectoryoriented BEV Conditioned Diffusion framework that establishes a fully perception annotation-free generative method for end-to-end autonomous driving. TrajDiff requires only raw sensor inputs and future trajectory, constructing Gaussian BEV heatmap targets that inherently capture driving modalities. We design a simple yet effective trajectoryoriented BEV encoder to extract the TrajBEV feature without perceptual supervision. Furthermore, we introduce Trajectory-oriented BEV Diffusion Transformer (TB-DiT), which leverages ego-state information and the predicted TrajBEV features to directly generate diverse yet plausible trajectories, eliminating the need for handcrafted motion priors. Beyond architectural innovations, TrajDiff enables exploration of data scaling benefits in the annotationfree setting. Evaluated on the NAVSIM benchmark, TrajDiff achieves 87.5 PDMS, establishing state-of-the-art performance among all annotation-free methods. With data scaling, it further improves to 88.5 PDMS, which is comparable to advanced perception-based approaches.

Results & Checkpoints

Method	NC	DAC	EP	TTC	Comfort	PDMS	Checkpoint
TrajDiff	98.2	96.1	81.9	94.2	99.9	87.7
TrajDiff(scaling)	98.2	97.0	82.7	94.6	99.9	88.6	Download

Getting Started

• Getting started from NAVSIM environment preparation
• Downloading the chekpoints
• Evaluation

sh scripts/evaluation/run_trajdiff_score_evaluation.sh

Acknowledgement

We thank the research community for their valuable support. TrajDiff is built upon the following outstanding open-source projects: NAVSIM, DiffusionDrive

Citation

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

@article{gui2025trajdiff,
  title={TrajDiff: End-to-end Autonomous Driving without Perception Annotation},
  author={Gui, Xingtai and Zhao, Jianbo and Han, Wencheng and Wang, Jikai and Gong, Jiahao and Tan, Feiyang and Xu, Cheng-zhong and Shen, Jianbing},
  journal={arXiv preprint arXiv:2512.00723},
  year={2025}
}