qwbu/univla-iros-manipulation-challenge-baseline

🌏 UniVLA

This is the baseline model for the AgiBot World Challenge @ IROS 2025 - Manipulation track.

Performance

This model is fine-tuned across all tasks in our manipulation challenge, building upon our publicly released pretrained checkpoint.
Additional fine-tuning using the Agibot-World dataset is encouraged to further enhance performance.

Results of baseline models. More detailed task descriptions and metric definitions can be found here.

Model Name	Total Score	Clear the countertop waste	Open drawer and store items	Heat the food in the microwave	Pack moving objects from conveyor	Pickup items from the freezer	Restock supermarket items	Pack in the supermarket	Make a sandwich	Clear table in the restaurant	Stamp the seal
UniVLA	2.336	0.194	0	0.198	0	0.08	0.55	1	0.064	0.25	0

How to use

Please visit our offical repo for detailed guidelines.

Citation

@article{bu2025univla,
  title={UniVLA: Learning to Act Anywhere with Task-centric Latent Actions}, 
  author={Qingwen Bu and Yanting Yang and Jisong Cai and Shenyuan Gao and Guanghui Ren and Maoqing Yao and Ping Luo and Hongyang Li},
  journal={arXiv preprint arXiv:2505.06111},
  year={2025}
}

@article{bu2025agibot,
  title={Agibot world colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems},
  author={Bu, Qingwen and Cai, Jisong and Chen, Li and Cui, Xiuqi and Ding, Yan and Feng, Siyuan and Gao, Shenyuan and He, Xindong and Huang, Xu and Jiang, Shu and others},
  journal={arXiv preprint arXiv:2503.06669},
  year={2025}
}