Behavioral Design by Operating Underlying Evaluation Functions

Conventional robotic behaviors are directly programmed depending on the designer’s personal experience. With this method, it is difficult to create new behavioral designs. In this study, the authors aim to establish a new method of designing robotic behaviors by operating underlying evaluation functions in mathematical forms that can generates behaviors in computer calculations. The characteristics of the method lie in the process of obtaining the evaluation function from a model behavior, which is opposite to the conventional process in which a predefined evaluation function generates the optimum behavior. This method is composed of two stages: (1) acquiring the evaluation function and (2) synthesizing the acquired evaluation functions and creating a new behavioral pattern from them. Our simulation shows that some evaluation functions of generating behaviors are synthesized using this method.