A new kind of learning algorithm with the mechanism of intrinsic motivation

Abstract

This paper presents a new kind of learning algorithm to help agents learn sensorimotor skills through their contact with the environment. The learning mechanism is based on Skinner's theory of operant conditioning (OC), which is improving the probabilities of good actions, and reducing the probabilities of bad actions. The concept of curiosity is introduced as the intrinsic motivation for an agent to learn. Using this algorithm to a robot with the task of right handshaking, the effectiveness of the algorithm is proved. The curiosity in this paper is new designed, and introduced as one factor that may affect the learning process. Its role and function are analyzed at the end of the paper.