Design and Implementation of Cognition Learning Algorithm for Humanoid Robot Playing 3 by 3 Square Baseball Game Using DBN and PSO

104

This thesis aims to design a cognition learning algorithm that allows the robot to learn the posture of playing 3 by 3 square baseball game. The robot can hit the designated grid area accurately with this proposed algorithm. The overall system proposed in this thesis includes image processing algorithm and learning algorithm. In the image processing system, a CMOS webcam sensor is used on the robot as the eye. In order to catch the ball location efficiently, two internet protocol cameras are installed on the top and side of the 3 by 3 square. To recognize and track the objects, a simple searching algorithm is developed for the issue. Then, a novel learning algorithm is motivated by a human thinking conception proposed in “Thinking, Fast and Slow” by Daniel Kahneman. He is a psychologist who won the Nobel Memorial Prize in Economic Science 2002. This algorithm is based on cognitive psychology, which divides human thinking into two modes, fast and slow. The fast mode favors intuitive thinking while the slow mode favors rational thinking. Furthermore, we establish the fast mode by Deep Belief Network and the slow mode by Inertia Weight Particle Swarm Optimization Algorithm in the developed cognition learning algorithm. The proposed algorithm is implemented and applied on the robot, and then the robot performs the fast and slow mode in 3 by 3 square baseball game. Finally, experimental results demonstrate that the performance of the cognition learning method is very efficient. In other words, this learning algorithm also verifies that the thinking mode of the human being is reasonable and available on the robot.