A motor learning model based on the basal ganglia in operant conditioning

Abstract

In this paper, a motor learning model based on the basal ganglia (BG) in operant conditioning (OC) using actor-critic (AC) learning is presented. The model has three networks for realizing action selecting function of the BS, such as actor network, critic network and explorer network. Actor network uses a probabilistic fuzzy controller for action selection which is enhanced by the introduction of a probability measure into the learning structure based on OC learning. A tropism mechanism is designed for describing intrinsic motivation which is a key factor for animal learning and it can direct the orientation of the agent learning. Critic network is composed of a multi-layer feedforward network and the learning is enhanced by TD(λ) algorithm and gradient descent algorithm. Explorer network is to settle the conflict between exploration and exploitation. Through the experiments of cognitive experiment, the method endows the mobile robot with the capabilities of learning obstacle avoidance and finding the target actively.