Learning Evaluation Function for RoboCup Soccer Simulation using Humans' Choice
Hidehisa Akiyama, Masahi Fukuyado, Toshihiro Gochou, Shigeto Aramaki · 2018
An online search approach to perform cooperative actions among simulated soccer players have shown practical performance. However, it is still not easy to design an appropriate evaluation function for the search process. In order to solve this problem, this paper proposes a method that applies learning to rank and a method to compose training data with smaller operational cost using clickthrough data. Experimental result shows our method successfully improved the performance of simulated soccer players in the small subtask.