Vision-Based Trajectory Planning And Robust Control For Autonomous Landing Of Unmanned Aerial Vehicles

For vision-based autonomous landing of fixed-wing unmanned aerial vehicles(UAV) which are incapable of hovering, the trajectory planning and robust control has been studied in this paper. The trajectory generation algorithm can provide optimized path which guarantees the target being field of view of camera and reduce the negative effects of vision-based estimation error. The robust controller enables the UAV to obtain robust performance with unreliable vision-based navigation information though the analysis of vision-based estimation error.Firstly, a strategy for vision-based autonomous landing is designed according to the relative pose between UAV and runway and the operating state of vision-based navigation system. The Go-Ground is introduced to handle the failure to detect and track target. Moreover, a segmented hierarchical path planning method is proposed and the off-line planning is implemented on a special one.As for on-line dynamic planning, constrains and performance index are firstly designed to guarantees the target being field of view of camera, and the gauss pseudospectral methods is introduced to transcript trajectory planning into a nonlinear programming. Also, a Dubins path planning algorithm is studied to improve the efficiency of planning. Finally, the receding horizon control is realized to reduce the negative effects of vision-based measurement error.A vision-based measurement model is developed to help elaborately analyze the sensitivity of vision-based position estimation to different model uncertainties. Meanwhile, a full-coverage nonlinear model of UAV is constructed and separated into two loops to make the best of information on trajectory planning. Then, the outer loop with the above model uncertainties is shaped by mixed H2/H∞ robust control to ensure robust performance.In the end, all studied technologies are verified by numerical simulation. Specially, the analysis defines the acceptable range of uncertainties of vision-based measurement model, such as calibration error, measurement error and measurement delay, which can guide the research on vision-based pose estimation.