Deep Belief Network–Based Learning Algorithm for Humanoid Robot in a Pitching Game

Abstract

A cognition learning algorithm based on a deep belief network and inertia weight Particle Swarm Optimization (PSO) is presented and examined in a humanoid robot. The psychology concepts were adopted from Thinking, Fast and Slow by Daniel Kahneman. The human brain comprises two systems, System 1 and System 2. Based on their characteristics, System 1 and System 2 handle different tasks during cerebration. In this study, Deep Belief Network (DBN) is trained to construct the function of System 1 for the rapid reaction. On the other hand, PSO is applied to build System 2 for the slow and complicated brain behavior. Through the cooperation of System 1 and System 2, the proposed cognition learning algorithm can apply the psychology theories to allow the humanoid robot for learning the suitable pitching postures autonomously. In the experiments conducted in this study, the robot was trained for only five selected points and was then asked to throw precisely to nine points. The proposed algorithm provided 100% accuracy in the robot pitching game. The feasibility of the proposed algorithm was thus verified.