Switching Learning-Based Cooperative Control with Its Application to Connected Automated Vehicles

Most existing model-based cooperative tracking control approaches heavily rely on precise dynamic models and overlook the transient performance, ultimately resulting in the designed controller falling far short of optimal control effectiveness. To address this issue, this paper proposes a novel switching learning algorithm to optimize the optimal formation tracking control in unknown heterogeneous multi-agent systems. Firstly, a distributed observer-based decentralized formation control protocol is proposed to provide the state estimation of the leader, where the estimated leader's state facilitates that each agent maintains formation distance between the leader and the formation process. Secondly, a decentralized control policy is considered and learned by the iterative solution of the Bellman equation to achieve the optimal formation tracking control for each agent. However, the aforementioned obtained results heavily rely on the system dynamics and an initial stabilizing control policy. To relax these limitations, a data-based switching learning algorithm is proposed, and it consists of a model-free matrix updating learning algorithm and a data-based policy iteration algorithm. In contrast to existing algorithms for similar studies, the proposed algorithms eliminate the system model and initial stabilizing requirements, but also ensure the formation control in an optimal control way. Finally, a practical connected automated vehicles example is given to verify the theoretical analysis.