Optimal Consensus of Heterogeneous Multi-Agent System with Unknown Dynamics
Abstract
This paper addresses the optimal consensus problem for heterogeneous multi-agent systems (MASs) with unknown dynamics. To eliminate dependence on accurate system models, we design distinct performance index functions for first-order and second-order agents, separating their tracking tasks. We employ a reinforcement learning algorithm to estimate the optimal performance index function and control policies. To enhance training stability, a novel training mode and a state-constrained iterative algorithm are proposed, with their convergence and optimality rigorously analyzed. Furthermore, a new error calculation method is introduced to accurately describe agent-neighbor distances, accelerating convergence and reducing energy consumption. The effectiveness of the proposed methods is validated through simulations.