Model-free linear quadratic tracking control for unmanned helicopters using reinforcement learning

This paper addresses the autonomous flight control system of an unmanned helicopter. We adopt a model-free discrete linear quadratic tracking (LQT) control architecture based on reinforcement learning algorithm by rewriting the Q-learning approach. From input and output data, the linear quadratic optimal gain is directly found without system identification procedure. Least square method is adopted in order to estimate the Q-value and the parameters related to optimal control gain. This methodology does not access to an exact model of the system and can be applied to full flight envelop maneuvering from hovering to aggressive flight with small modification. We constructed numerical simulations to evaluate the proposed algorithm with a discrete linear model of the unmanned helicopter.