Optimal Event-Triggered Consensus for Unknown Input-Constrained MASs via Cooperative Regression Reinforcement Learning

This paper investigates the consensus problem of unknown input-constrained nonlinear heterogeneous multi-agent systems (MASs) and proposes an optimal dynamic event-triggered control protocol based on a cooperative regression reinforcement learning (CRRL) algorithm. First, two neural networks (NNs) for each agent are used to approximate the system's dynamics model information. The designed identifier incorporates a compensation dynamic parameter and a compensation matrix, resulting in faster identification speed. Additionally, using an actor-critic structure, the event-triggered control and dynamic triggering conditions are designed, which does not exist in Zeno behavior. Subsequently, a CRRL algorithm is developed to ensure that the constant weight updating error is uniformly ultimately bounded and the consensus error asymptotically converges to zero. Notably, this algorithm relaxes the persistent excitation condition. Finally, the effectiveness and superiority of the proposed theoretical algorithm are validated through inverted pendulum systems and 2-DOF robots.