AI-based Adaptive Nonlinear MPC for Quadrotors

The UAV flight control tasks, especially their stabilities under large disturbances, are always complex and troublesome. In many application environments, drones are subject to both external interference and internal structural changes, which greatly affects the tracking accuracy. In this paper, an AI-based nonlinear control strategy was presented for unmanned aerial vehicles to achieve accurate tracking tasks. The basic idea is to combine neural network learning with traditional nonlinear model predictive control (NMPC). Moreover, some adaptive control algorithms are adopted for better robustness. First, existing MPC with L1 adaptive control is adopted as main controller for rejecting external disturbances. Second, considering path tracking performance, model reference adaptive control is added to the control policy to compensate internal structural changes. Third, to take advantage of the wealth of flight data, the neural network learning is adopted and the predictive error signal is fed back to the control system. Finally, the simulation studies are given to illustrate the effectiveness of proposed control strategy.