A DDPG-based Solution for Optimal Consensus of Continuous-time Linear Multi-agent Systems

Y. Li*, Z. Liu, G. Lan, M. Sader, Z. Chen

📄 Paper

Modeling engineering systems is time-consuming and labor-intensive, since system parameters drift with temperature, component aging, and other factors. We propose a data-driven, model-free optimal controller based on Deep Deterministic Policy Gradient (DDPG) for continuous-time leader-following multi-agent consensus:

• No system model required. The controller reaches consensus using only the consensus error, without any initial admissible policy.
• Neural function approximation. Two neural networks fit the state and action spaces, avoiding the dimensional explosion of the time-consuming state-iteration process.
• Self-learning & energy-efficient. It adapts in real time as system parameters change, achieving consensus with minimal energy consumption.

Convergence and stability are proven theoretically and verified through simulation experiments.