Reducing computational time on evolution under the real environment using fitness estimation

Evolutionary computations (ECs) are widely applied to optimization problems. Although they show good performance in many cases, they are generally time-consuming because of their trial-and-error-based characteristics. One of the main advantages of ECs is their flexible fitness definitions. Recently, interactive ECs (IECs), which are derived by human evaluation, have shown their performance in the field of design schemes. In IECs, there is a serious problem that the load for the human evaluator is very high. In this paper, we apply a fitness estimation method to lighten the load of the human evaluator. Evolutions of robots are examined using LEGO Mindstorms to show the feasibility of the proposed method.