An event-triggered formation control strategy is proposed for a multi-agent system (MAS) suffered from unknown disturbances. In identifier-critic-actor neural networks (NNs), the strategy only needs to calculate the negative gradient of an approximated Hamilton-Jacobi-Bellman (HJB) equation, instead of the gradient descent method associated with Bellman residual errors. This simplified method removes the requirement for a complicated gradient calculation process of residual square of HJB equation. The weights in critic-actor NNs only update as the triggered condition is violated, and the computational burdens caused by frequent updates are thus reduced. Without dynamics information in prior, a disturbance observer is also constructed to approximate disturbances in an MAS. From stability analysis, it is proven that all signals are bounded. Two numerical examples are illustrated to verify that the proposed control strategy is effective.

中文翻译：

---

通过简化强化学习对具有未知扰动的多智能体系统进行事件触发近似优化的编队控制


针对未知扰动的多智能体系统 （MAS），提出了一种事件触发编队控制策略。在标识符-批评者-参与者神经网络 （NN） 中，该策略只需要计算近似的 Hamilton-Jacobi-Bellman （HJB） 方程的负梯度，而不是与 Bellman 残差相关的梯度下降方法。这种简化的方法消除了对 HJB 方程残差平方的复杂梯度计算过程的要求。批评者行为者 NN 中的权重仅在违反触发条件时更新，因此减少了频繁更新造成的计算负担。在没有先验动力学信息的情况下，也会构建一个干扰观测器来近似 MAS 中的干扰。从稳定性分析中可以证明所有信号都是有界的。通过两个数值算例验证了所提出的控制策略是有效的。