Evolutionary construction of behavior arbitration mechanisms based on dynamically-rearranging neural networks

Recently, the evolutionary robotics (ER) approach has been attracting lots of concern in the fields of robotics and artificial life, since it can automatically synthesize controllers by taking the embodiment and the interaction dynamics between the robot and its environment. However, the ER approach still has serious problems that have to be solved. In this study, we particularly focus on one of the critical problems in the ER: plasticity vs. stability dilemma. In order to alleviate this problem, we investigate the effectiveness of the dynamically-rearranging neural networks by taking a peg-collecting task, which requires appropriate sequence of behavior to accomplish the task, as a practical example.