Adaptive Neural Optimal Backstepping Control for Heterogeneous Multiagent Systems With Noncooperative Target via Identifier–Critic–Actor Algorithm

This article proposes a control strategy that integrates simplified optimized backstepping control with RBF neural network (NN) observer to address the issue of heterogeneous multiagent cooperative control. Under the identifier–critic–actor framework, the optimized backstepping control method shapes every regression virtual and actual control to be the most effective solutions of the respective subsystems. Meanwhile, the NN observer approximates the unknown nonlinear equations of the system. The identifier is tasked with estimating unknown dynamics. The critic is responsible for assessing the performance of the system. The actor is charged with carrying out the necessary control actions. The rules for updating the reinforcement learning algorithm stem from the negative gradient of a straightforward, positive-definite function. At the same time, this design scheme can break through the limitation of the persistent excitation condition required by most existing optimal control methods. Ultimately, the efficacy of the approach is confirmed both through theoretical analysis and simulation studies. The method proposed in this article will provide support for related research on embodied artificial intelligence (AI) agent systems conducting self-organized collaboration.