Cooperative Path Tracking-Based Learning Control for Unknown Multi-Agent Systems via Dynamic Event-Triggered Mechanisms

This paper investigates the event-triggered output path-tracking control of networked heterogeneous multi-vehicle (agent) systems with unknown model dynamics. Different from most existing distributed observer methods to estimate the leader vehicle’s state matrix and state, these state-based observer approaches raise the disadvantages of high dimensionality and high frequency of data exchange of state. To address this, in this paper, we propose a novel adaptive distributed output observer (ADOO) that estimates the coefficients of the minimal polynomial instead of requiring knowledge of all the entries of the leader vehicles system matrix. Moreover, our proposed ADOO is model-free without relying on the leader’s accurate system, unlike the model-based way in existing works. Meanwhile, an asynchronous dynamic event-triggered control strategy is developed to reduce the communication load among neighboring vehicles. Then, a decentralized path-tracking controller is learned via a model-free matrix updating learning technique to achieve optimal path-tracking control without requiring an initial stabilizing control policy. By rigorous mathematical analysis shows that our proposed algorithms not only can greatly reduce the dimension of existing observer methods and the frequency of information exchange among neighboring vehicles, but also exclude the Zeno phenomenon. Finally, the numerical simulation is used to validate the efficiency of the theoretical algorithms under investigation. Note to Practitioners—This paper investigates the learning-based distributed optimal path-tracking control for heterogeneous vehicle systems with event-triggered communications. The proposed distributed control algorithms can be applied to multiple ground vehicles, air vehicles, and underwater vehicles. Unlike existing results on distributed state observer methods, the obtained results rely on accurate system dynamics and ignore the transient performance, which makes the designed controller far from optimal in potential applications. Moreover, the optimal learning algorithm strictly relies on the initial stabilizing control policy related to accurate system dynamics, and it may be helpless when the system model is completely unknown. To overcome those issues, we developed a model-free matrix updating learning technique to study distributed output path tracking control by collecting system data instead of using accurate system dynamics. Our algorithm not only overcomes the initial stabilizing assumption but also ensures the algorithm convergence in a model-free fashion. Besides, our proposed distributed observer can greatly lower the dimension and the frequency of information exchange of existing observers. Potential applications of the proposed control algorithms include cooperative formation control of heterogeneous unmanned systems.