Reinforcement learning in continuous systems: Wavelet networks approach

Abstract

In this paper we integrate the self-adjustment environment advantages of reinforcement learning into the adaptive wavelet network controller. The novel approach is called adaptive wavelet reinforcement learning control, which uses wavelet to approximate a continuous Q-function, in order to obtain a optimal control policy. The simulations of applying the proposed method to an inverted pendulum and Pendubot system are performed to demonstrate the properties of the adaptive wavelet network controller. © 2007 Springer-Verlag Berlin Heidelberg.