Formation Control of Multiple Aerial Robots Using LSTM-based Model Predictive Control

This paper proposes a Long Short-Term Memory- (LSTM-) based Model Predictive Control (MPC) design for the formation of multi-agent Unmanned Aerial Vehicles (UAVs). The quadrotor model is considered nonlinear, and there are no simplifying assumptions; therefore, the results are more reliable, and the proposed method is applicable. A cascade controller with a position controller as the main loop and an attitude controller as the inner loop is presented to address the formation control problem. An LSTM network is used as the estimator to predict the future states of each quadrotor, and the Differential Evolution (DE) finds the optimal value of inputs to satisfy constraints and decrease tracking errors. In addition to the online learning ability, the proposed controller could overcome environmental or structural disturbances, uncertainties, and noises. Furthermore, closed-loop stability analysis is studied. In the end, simulation examples are given to prove the acceptability of the presented control structure.