Controller comparisons for autonomous railway following with a fixed-wing UAV

The development of an on board vision-based path following system for a fixed-wing UAV is presented. The application is railway following and three path-following controllers have been investigated for their suitability for this task. The system is novel in that it includes all system components on board a lightweight and low-cost UAV. A vision system has been developed to characterise railway lines as first-order polynomials and relay this information to a flight controller. Its functionality was extended to provide information about the pixel distance between railway tracks, to be used to calculate the height of the plane above the railway. In simulation, the Non-Linear Guidance Law was found to be the controller that yielded the fastest settling times and least overshoot values; followed by the Vector Field method. A baseline PID method was the worst performing controller. Flight tests were performed to verify operation of the system but no conclusions could be drawn regarding controller performance.