Evolutionary approach for designing the behavior generator of communication robot

Our goal is to create the robot system which interacts with human users keeping their interest during a long period. We focus on the Interactive Evolutionary Computation (IEC) technique to achieve this goal. Although the IEC enables users to design various systems which reflect their subjective preferences, it forces users to evaluate a huge number of individuals in the genetic pool during the evolution period. To solve this problem, we propose a refined IEC technique, named Human-Machine Hybrid Evaluation (HMHE), which selects the representative genes for user evaluation and estimates the evaluation results of the other genes. It can increase the population size without increasing the users' evaluation processes. We carried out some simulations where a humanoid robot with our method interacted with a user. The experimental results demonstrated that the HMHE could continue to generate the various robot behaviors by adapting to the transition of user's subjective preferences.