🧲 Federated MARL x PINNs

There have been several works exploring the intersection between PINNs and MARL.

For example, Sebastián et al. [1] worked on physics-informed MARL for policy architecture for robotics. They used port-Hamiltonian descriptions of multi-robot systems as the policy parameterisation, encoding energy conservation and network topology directly into the actor architecture. Li et al. [2] proposed an F-MADRL algorithm where the federated learning mechanism trains the multi-agent system while preserving privacy and data security, and a physics-informed reward function encodes domain knowledge directly into the reward signal rather than into the network architecture. They applied it to multi-microgrid energy management.

References

[1] E. Sebastián, T. Duong, N. Atanasov, E. Montijano, and C. Sagüés, "Physics-informed multi-agent reinforcement learning for distributed multi-robot problems," IEEE Trans. Robot., 2025, doi: 10.1109/TRO.2025.3582836, arXiv:2401.00212.

[2] Y. Li, S. He, Y. Li, Y. Shi, and Z. Zeng, "Federated multi-agent deep reinforcement learning approach via physics-informed reward for multi-microgrid energy management," IEEE Trans. Neural Netw. Learn. Syst., vol. 35, no. 5, pp. 5902–5914, 2024, arXiv:2301.00641.