DreamPlan: Efficient Reinforcement Fine-Tuning
of Vision-Language Planners via Video World Models

Emily Yue-Ting Jia^*1, Weiduo Yuan^*1, Tianheng Shi¹, Vitor Guizilini², Jiageng Mao^†1, Yue Wang^†1

¹USC Physical Superintelligence (PSI) Lab    ²Toyota Research Institute

^*Equal contribution   ^†Equal advising

---

Code is coming soon!

Citation

@misc{jia2026dreamplanefficientreinforcementfinetuning,
      title={DreamPlan: Efficient Reinforcement Fine-Tuning of Vision-Language Planners via Video World Models}, 
      author={Emily Yue-Ting Jia and Weiduo Yuan and Tianheng Shi and Vitor Guizilini and Jiageng Mao and Yue Wang},
      year={2026},
      eprint={2603.16860},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2603.16860}, 
}

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.