Neural Network Based Model Predictive Control for a Quadrotor UAV

A dynamic model that considers both linear and complex nonlinear effects extensively benefits the model-based controller development. However, predicting a detailed aerodynamic with good accuracy for unmanned aerial vehicles (UAVs) is challenging due to their irregular shape low Reynolds number behavior. This work proposes an approach full translational dynamics of quadrotor UAV by feedforward neural network, which adopted as prediction in predictive (MPC) precise position control. The raw flight data are collected tracking various pre-designed trajectories PX4 autopilot. network trained predict accelerations from log. network-based then implemented automatic control optimization toolkit (ACADO) achieve real-time online optimization. Software loop (SITL) simulation indoor experiments conducted verify performance. results indicate proposed leads 40% reduction average trajectory error compared traditional PID controller.