Biologically Inspired Decentralized Adaptive Optimal Tracking Control For Large Scale Multi-Agent Systems with Input Constraint

In this paper, the online decentralized adaptive optimal tracking control problem has been investigated specifically for multi-agent system (MAS) with large population and input constraints. Most conventional MAS tracking control algorithms that deals with input constraints are suffering from the "curse of Dimensionality". Their effectiveness could be degraded significantly when the scale of MAS is relatively large. Therefore, in this paper, a novel online learning based decentralized optimal control strategy has been developed to tackle these challenges by engaging the emerging Mean Field Games (MFG) theory with a novel biologically inspired Actor-Critic-Mass (BI-ACM) learning scheme. Specifically, a biologically inspired Neural Network that mimics human brain is designed based on the Spiking Neural Network (SNN). The designed SNNs can effectively reduce the computation complexity through adaptively activating the necessary neurons based on real-time learning behavior. Similar to human brains, the BI-ACM includes three regions of neurons including 1) the reward region to approximate the optimal cost function which specially designed for the input constraints, 2) MAS population estimation region to predict the affects from other agents, and 3) action region to compute the optimal control. Moreover, a novel SNN weights update law based on gradient descent has been develop in this paper. Finally, the effectiveness of the proposed scheme have been validated through a series of numerical simulations.