Website | Paper | Video

We are currently in the process of organizing our code, and will release the code on GitHub upon acceptance of the paper.

Stay tuned for updates!

---

CurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models

Zihao Sheng^1,*, Zilin Huang^1,*, Yansong Qu², Yue Leng³, Sruthi Bhavanam³, Sikai Chen^1,✉

¹University of Wisconsin-Madison, ²Purdue University, ³Google

^*Equally Contributing First Authors, ^✉Corresponding Author

😮 Highlights

🔥 To the best of our knowledge, CurricuVLM is the first work to utilize VLMs for dynamic curriculum generation in closed-loop autonomous driving training.

🏁 CurricuVLM outperforms state-of-the-art baselines, across both regular and safety-critical scenarios, achieving superior performance in terms of navigation success, driving efficiency, and safety metrics..

Before training with CurricuVLM

Case 1	Case 2	Case 3	Case 4	Case 5

After training with CurricuVLM

Case 1	Case 2	Case 3	Case 4	Case 5

🎯 Citation

If you find CurricuVLM useful for your research, please consider giving us a star 🌟 and citing our paper:

@article{sheng2025curricuvlm,
  title={CurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models},
  author={Sheng, Zihao and Huang, Zilin and Qu, Yansong and Leng, Yue and Bhavanam, Sruthi and Chen, Sikai},
  journal={arXiv preprint arXiv:2502.15119},
  year={2025}
}