Double inverted pendulum control by linear quadratic regulator and reinforcement learning

The paper gives an original combination of linear quadratic regulator and reinforcement learning dedicated to the position control of a double inverted pendulum system. An agent based on a modified Sarsa algorithm is applied to swing up the pendulum. The linear quadratic regulator is applied to the linearized mathematical model of the process in the vicinity of upright position. Digital simulation results show the performance of the new approach.