RNN-Based Model Predictive Control of Multi-agent System Using Switching Topologies

In this paper, we presented the RNN-based MPC strategy for improving the coordination and decision making of the MAS for the consensus control problem. A recurrent neural network-based optimization reduces the computational burden over MPC for the nonlinear multi-agent systems (MAS) thereby improving the system’s ability to overcome the external disturbances in the system. Our approach involves predicting the cost objective function to be minimized over finite horizon using MPC and optimizing the control objective using recurrent neural network (RNN). The algorithm is evaluated through simulation studies on the quadrotor fleet MAS. Overall, this research highlights the potential of combining RNNs and MPC to enhance the performance of the multi-agent systems.