Decision-making strategy for multi-agents using a probabilistic approach: application in soccer robotics

Abstract

The efficient coordination of soccer robots is a complex topic because there are numerous possible scenarios in the game, and the state of the robots can change rapidly. It requires the robot to be able to analyze and make decisions in a short time. In this article, we first use Dec-POMDP to describe the actions and states of the soccer robot team. Secondly, we introduce a probabilistic approach so that the robots can quickly make decisions corresponding to specific situations. More specifically, we present a method for calculating and evaluating the expected points corresponding to each particular action for each robot. The robots then choose the actions with the highest expected points. Finally, we have developed a simulator based on the digital twin approach to verify the proposed strategies on simulation models and implement these strategies rapidly on real robots.