Interactive trajectory generation using evolutionary programming for a partner robot

Abstract

This paper proposes an integrated method for generating a human-friendly trajectory. First of all, the robot detects the position of the facing human, and then, the robot generates the trajectory realizing a hand-to-hand behavior by using evolutionary programming. Basically, human evaluation is very important for generating robotic behavior, but the structure of human evaluation is not clear beforehand. Therefore, a fuzzy state-value function is used for estimating the structure of human evaluation. We apply a profit sharing plan using the human evaluation to update the fuzzy state-value function. Furthermore, we propose a temperature scheduling method of a Boltzmann selection dependent on the time-series of human evaluation in the interactive evolutionary programming. Several experimental results show the proposed method can generate a human-friendly trajectory with few human evaluation times.