Research On Mini-UAV Hovering Controller Algorithm

Miniature unmanned aerial vehicle (mini-UAV) is the contractible type of manned large helicopter. Although it has small cubage, its effect for military and civil use was increasingly extended because of its sensitivity. Besides its easy operation, mini helicopter can also complete some challenge tasks that large ones can not do, such as using in survey, monitoring and so on. The mini-UAV can economizes manpower and thing power. So the research of mini-UAV has great significance. Since the mini-UAV has the special characters such as high coupling, nonlinearity, asymmetry, the control of it remains in an initial stage and the based technique of UAV such as aerobatic flight, still need artificial supervision. So the design of controller of the mini-helicopter and implementation of full-automatic control, have great significance on complicated work. This article is mainly about the controller design of mini-UAV in the state of hovering. First I use LQR/LQG method to design the controller, which is easy to calculate, this algorithm used in helicopter can make the helicopter hover rapidly and maintain the state. Secondly,I use the controller based on neural network which not only can simulate nonlinearity but also has fine learning capacity and high degree of variation, so it is very suitable for the mini-UAV system that has complex degrees of freedom. In order to adjust the complex parameters of the controller, I adopted a random search method which combines the hill climbing algorithm and SPSA (Simultaneously Perturbation Stochastic Approximation) to insure multiple parameters can be changed simultaneously, at the same time, the reward function of reinforcement learning algorithm can grow in the increment directionBased on above theories the article also makes some simulation experiments, which can prove that the controller system that use algorithms proposed in the paper can greatly improve the performance of the system, thus these algorithms is effective.