ALTER: Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand

Dayi Dong, Maulik Bhatt, Aayushi Shrivastava, Lasse Peters, Negar Mehr
University of California, Berkeley

ALTER stands for Adaptation from Limited demonstrations for Team coordination with Existing-skill Retention. It adapts pretrained diffusion policies to coordinate with other robots while retaining their original independent skills. A trainable residual coordination head corrects a frozen base policy using limited collaborative demonstrations and single-agent replay distilled from the base policy itself.

Paper · Project website

Website

The project website is published from docs/. This public repository currently contains the paper website and supporting media. The research code is being prepared separately for release.

Citation

If you find ALTER useful in your research, please cite:

@article{dong2026alter,
  title={Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand},
  author={Dong, Dayi and Bhatt, Maulik and Shrivastava, Aayushi and Peters, Lasse and Mehr, Negar},
  journal={arXiv preprint arXiv:2609.32129},
  year={2026},
  url={https://arxiv.org/abs/2609.32129}
}