Prediction of the optimal parameter values in reinforcement learning as a function of the environment

Reinforcement learning (RL), suitable for navigation of a mobile robot, has a difficulty in that parameter values can only be determined by trial and error. We proposed to use a genetic algorithm (GA) with inheritance to obtain optimal parameter values in RL, which drastically reduced the computational cost, and we proposed further reduction in our previous study. Our proposal, however, had to optimize the parameter values in RL for every environment, which was impractical due to the huge computational cost. In this paper, we propose to predict the optimal parameter values in RL as a function of measures for the complexity of the environment by multiple regression analysis, and we succeed in estimating the optimal parameter values in RL.