Autorotation of an UAV helicopter via reinforcement learning

Abstract

The single-engine manned helicopters can face the undesired engine failure during their mission operation. In this case the pilots typically perform the autorotation maneuver properly and achieve the safe landing. For an unmanned air vehicle (UAV) helicopter, construction of the controller performing the autorotation maneuver autonomously is considered as a significant subject. It allows the user to save not only the UAV helicopter platform but the valuable sensors and data on board. However, the autorotation maneuver is a quite challenging problem that requires time-critical control inputs to accomplish a safe landing. Also, the dynamics and aerodynamics of the helicopter in autorotation are highly non-linear and complex. From these facts it is necessary to devise a controller which is able to obtain satisfactory performance in this field. Reinforcement learning is the feasible approach that can be applied in a complicated system. It is known that the reinforcement learning technique is based on the way of animal’s thinking, learning and acting. Contrary to supervised learning, the controller can adapt itself by the suggested reward function in a natural way. After several trials, the controller can find the solution to success the safe landing via reinforcement learning.