Mimicking an Expert Team Through the Learning of Evaluation Functions from Action Sequences

Abstract

In the RoboCup Soccer Simulation 2D League, the performance of teams highly leans on the evaluation functions used for their decision making process. The aim of this paper is to propose a method that improves the performance of a team by mimicking a stronger one. For this purpose, a neural network is employed to model an expert team’s evaluation function. The neural network is trained by using positive and negative episodes of action sequences that are extracted from game logs. In our experiments, we successfully improved the performance (e.g., win rate, scored goal, and so on) of our team by mimicking the winner of RoboCup 2017 soccer simulation 2D league.