MULTIMODAL SENSORY MOTOR COORDINATION BY LEARNING IN DYNAMICAL SELF-ORGANIZING RELATIONSHIP NETWORK

作者：Sakaguchi Shunsuke^*; Horio Keiichi; Yamakawa Takeshi

来源：International Journal of Innovative Computing Information and Control, 2011, 7(10): 6097-6114.

• Multimodal sensor fusion
• Sensory motor coordination
• Dynamical self-organizing relationship network (DSOR)
• Adaptation
• Evaluation
• Fuzzy inference
• Trial-and-error
• Foveation
• Sound localization
• Knowledge acquisition

Action learning of autonomous robots is an important issue for development of complex robot systems. In the learning, robots learn by sensory motor coordination. The difficulties of this issue are that they should learn by trial-and-error and that they should adapt their actions to changes of environments. Foveation, which is to move their body and their sensors in order to capture a target object on the center of the visual field, is an essential example of the sensory motor coordination. Although several learning methods have been proposed for autonomous foveation learning, these have several crucial problems of efficiency and adaptability. In order to overcome these problems, in this paper, we propose a new learning method called Dynamical Self-Organizing Relationship Network (DSOR). It can learn an input/output relationship between multiple sensory information and motor action by trial-and-error, and acts as a knowledge acquisition tool and also as a fuzzy inference engine. It can also adapt to the dynamic change of I/O relationship. We examine the DSOR by applying it to a foveation learning problem in the computer simulation.

• 出版日期2011-10