Distributed Koopman learning with incomplete measurements

September 2024 ยท W Hao, L Wang, A Rai, S Mou


Abstract

This paper considers collaborative system identification when agents have incomplete local measurements. It combines Koopman lifting, deep neural networks, and consensus coordination so agents exchange lifted-state information and reconstruct global nonlinear dynamics. Experiments on Lunar Lander show performance close to centralized deep Koopman learning with full-state access.


Demo

Demo coming soon.


Citation

Hao, Wenjian, Lili Wang, Ayush Rai, and Shaoshuai Mou. 2024. "Distributed Koopman Learning with Incomplete Measurements." arXiv preprint arXiv:2409.11586.

@techreport{WHao2024distribu,
  author = {W Hao, L Wang, A Rai, S Mou},
  year = {2024},
  title = {Distributed Koopman Learning with Incomplete Measurements},
  number = {arXiv:2409.11586},
  url = {https://arxiv.org/pdf/2409.11586}
}