Adaptive Learning-Based Observer with Dynamic Inversion for the Autonomous Flight of an Unmanned Helicopter

This study presents the development of an autonomous flight control system for a small-scale unmanned helicopter based on an online adaptive learning-based observer and model predictive control (MPC). The adaptive learning-based observer provides an approximate dynamic inversion for solving the coupled dynamic problem. The discrete-time model predictive control (DMPC) uses a prediction process to obtain a stable model. The Laguerre function, a state observer, and a recursive learning algorithm were integrated into DMPC, the core of the flight control system. Exponential data weighting was adopted to increase the stability of the designed flight control system. By integrating these approaches, an adaptive learning-based flight control system, which consists of the pitch, roll, yaw, and heave controllers, is presented to evaluate the stability and performance of the proposed autonomous flight control system of an unmanned helicopter. Moreover, an adaptive controller based on neural approximation for yaw motion is used to improve the performance of yawing control of the helicopter. In order to validate the feasibility of the proposed method, a hardware-in-the-loop (HIL) simulation was performed in real time. The system included an embedded system, a self-developed users interface (UI), and the X-Plane flight simulator. The simulation results showed that the flight control system was able to maintain the position and the attitude of the helicopter in hover mode under the influence of wind gust and disturbance. The proposed method had better performance than a robust control system, in both ideal and disturbed environments.