Effective Strategies for Improving Mobility Efficiency and Keeping Numerical Superiority in AI Worldcup

Abstract

AI technologies have been a revolution for computer vision. With AlphaGo winning the Go game against human champion in 2016, people have expected AI technology to be applied to various computer games. In response to these expectations, the first AI world cup competition was held at KAIST last year. Like a real soccer game, five robots in each team play against for scoring higher goals for winning. However, it is not a human-to-computer game like AlphaGo, but some intelligent algorithms plays the game according to the learned strategy. In AI soccer, there are some factors that are slightly different from other soccer games, so effective strategies are needed. In this paper, we propose three effective strategies for AI world cup competition; effective role-change strategy, prediction of ball trajectories, and "deadlock" avoidance. We compared our method with different cases and got the best result against the default rule-based code. We added further works that can improve the performance in the appendix.