Proposal of new framework of body image - Acquisition of the body image and distinction of tools by reinforcement learning

The thing that enables animals to move adaptively is called body image. In the field of psychology, it is considered that the body image is expanded when human beings use tools, and it becomes possible to use the tools effectively. On the other hand, in the field of control engineering, it is considered that the function which enables to control its body at will is realized by inverse model of the body, and how to obtain the inverse model was studied. However, to obtain the inverse model by learning is very difficult because usually the model is nonlinear and there is mutual interference. And moreover, robots cannot distinguish tools from useless objects autonomously. In this paper, we propose new definition of the body image, and we demonstrate that robots can obtain the body image by reinforcement learning and can distinguish tools from useless objects autonomously by using the obtained body image.