Neural Network based Self and Mutual Concept Learning for Robot Cooperation

• Main Authors: Chun-HsienHou, 侯駿賢
• Other Authors: Tzuu-Hseng S. Li
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/qu7bch

碩士 === 國立成功大學 === 電機工程學系 === 103 === Multi-robot cooperation is an important issue in robotics. This thesis proposes a self and mutual concept learning algorithm based on Neural Network (NN) for robot cooperation. Robots learn a concept not only by themselves but also from each other, and they cooperate to complete a complicated task, that of Form Fitter. In the form fitter game, one robot explores the textures of shapes and grasps a shape on the desk to put on a box held by another robot. The Mutual Learning Neural Network (MLNN) system evolved from the Backpropagation Neural Network (BPNN). This visual system extracts the color and shape of objects by HSV and normalizes the images by bilinear interpolation. Robots utilize the TCP/IP communication system to communicate with each other and generate a series action of arm. The MLNN system updates both weights in the neural network system of each robot at the same time. The system compares the recognition results of both robots and chooses the better one. The robot which has greater accuracy will translate its learning weight to the other one to improve the performance of both robots. After learning many times, both robots can learn a concept. Finally, the proposed method is simulated by Matlab and demonstrated in two home service robots. The experimental results show the efficiency and feasibility of the MLNN system.