Neural Network Controller for Drones

Neural networks have gained prominence in control systems due to their ability to approximate complex nonlinear mappings and adapt to uncertain environments. The model predictive control (MPC) problem for a quadcopter is a complex one, rooted in the fact that the dynamics model of the quadcopter is inherently complex and requires a suitable controller. This study proposes a design for an MPC controller to land a quadcopter by training a neural network. It also integrates a linear dynamic model to achieve a smooth descent of the quadcopter. The loss function considers the impact of path planning, deviations from the target state, and operational costs. Experiments demonstrate the advantages of the constructed model by varying the number of hidden layers, and the results show that it achieves good control performance.