Genetic programming for life-time learning of a mobile robot

The paper deals with a mobile robot with structured intelligence. The robot interacts with a dynamic environment. The evaluation criteria or functions are the strategy for the behavior acquisition. Generally, it is difficult for human operators to describe the internal models of the robot because the organization of the robot is quite different from that of a human. In the optimization, the evaluation function is generally given by human operators beforehand. It is easy to give the evaluation functions if the environmental condition is easy and fixed, but the robot must interact with dynamic, uncertain and unknown environments or human operators. Therefore, the robot should generate the evaluation criteria by itself based on its embodiment. A human improves its behavior by using and changing its evaluation criteria as adaptive processes. The robot also has to acquire their evaluation criteria through life-time learning. Therefore, we apply genetic programming (GP) for generating evaluation functions. The result of computer simulation shows that GP can generate the evaluation function suitable for the applicable environments, the given tasks, and the robot.