Synchronization For Heterogeneous Multi-Agent Systems Via Adaptive Dynamic Programming

The idea of distributed cooperation among multi-agent systems has garnered significant attention in various research fields,including biology,physics,robotics,control engineering,and social science.This approach has practical applications in controlling the formation of unmanned aerial vehicles,coordinating mobile robots,and designing distributed sensor networks.In practical applications,the design of control algorithms for multi-agent systems may face various limiting factors or mixed factors,thereby requiring consideration of a range of constraints such as resource and input constraints.Furthermore,due to the heterogeneity and unknowability of such systems,higher design requirements are necessary.For instance,in military networks,a multi-agent system consisting of diverse and unknown agents including unmanned aerial vehicles,unmanned underwater vehicles,and unmanned ground vehicles must effectively collaborate to achieve complex missions.The objective of this paper is to address the optimal cooperative control problem of unknown,high-order heterogeneous multi-agent systems that face resource and input constraints in a directed graph setting.An adaptive dynamic programming algorithm is proposed to tackle two key cooperative tasks:tracking synchronization and bipartite tracking synchronization.The main result of the dissertation is given as follows.Firstly,the problem of bipartite optimal output tracking synchronization for high-order linear unknown heterogeneous multi-agent systems with adversarial inputs is investigated.A state space transformation method is proposed that converts the bipartite output tracking synchronization problem into an output tracking synchronization problem,followed by the design of an H∞ optimal bipartite controller to achieve asymptotic convergence of the synchronization errors.The bipartite optimal control and the solution of the algebraic Riccati equation are obtained using the model-free adaptive dynamic programming algorithm,which does not require the dynamics information of agents.Secondly,the bipartite optimal state tracking synchronization problem for highorder linear unknown heterogeneous multi-agent systems is investigated with adversarial inputs.A fully distributed observer-base H∞ bipartite optimal controller is designed to realize synchronization between agents under the existence of a bounded control of the leader.The optimal control gain and the solution of the inhomogeneous algebraic Riccati equation can be obtained by using the model-free adaptive dynamic programming algorithm without requiring dynamics information or global information from the agents.Thirdly,this study focuses on solving the event-triggered optimal output tracking synchronization problem for high-order linear unknown heterogeneous multiagent systems.A model-free optimal event-triggered controller and triggering condition are proposed for a realistic system with resource constraints,such as communication and computing burden.This controller uses the optimal control gain to achieve aperiodic sampling and output synchronization among agents.It also proves that the designed event-triggered controller does not exhibit Zeno behavior,thereby confirming the feasibility of the proposed algorithm.Fourth,the optimal output tracking synchronization problem of high-order nonlinear unknown heterogeneous multi-agent systems that have asymmetric input constraints is studied.The follower undergoes a system transformation to convert the synchronization problem with asymmetric input constraints into a synchronization problem with symmetric input constraints.An optimal input-constrained controller based on the distributed observer is proposed.The optimal control factors and input-constrained HJB equations are obtained by using the improved modelfree adaptive dynamic programming algorithm,which the algorithm eliminates the need for system dynamics information required by the policy iteration algorithm.Furthermore,the convergence of the proposed algorithm and the asymptotic convergence of the synchronization error are proved.Finally,an optimal solution to the problem of dynamic event-triggered state tracking synchronization for high-order nonlinear unknown heterogeneous multiagent systems is developed.The optimal dynamic event-triggered controller and triggering mechanism that uses a fully distributed observer are proposed for a realistic system with resource constraints.An adaptive dynamic programming algorithm based on an identification-actor-critic network structure is used to obtain the optimal dynamic event-triggered control,which does not require the dynamics information of the follower and global information related to the directed communication topological graph.Additionally,it proves that the dynamic triggering mechanism performs better than the static triggering mechanism in excluding Zeno behavior,and further gives the feasibility analysis of the proposed algorithm.