A self-learning sensorimotor model based on operant conditioning theory

Abstract

This paper presents a self-learning model to help agents learn the sensorimotor skills. The model includes the sensory part, the motorial part, the sensorimotor map and the learning mechanism. At every learning step, the agent senses its states in its internal environment, executes motions based on the sensorimotor map, and at the same time gets a reward from the external environment as the result of its behavior. Then the sensorimotor map is tuned according to the learning mechanism which is designed based on the theory of Skinner operant conditioning. The convergence of learning mechanism is proved. To show the model's ability of self-learning, the paper first simulated the famous Skinner pigeon experiment, and then used the model to a robot with the task of right handshake. Both of the results show that the model designed is intelligent and can help agents learn the sensorimotor skills.