Host: Japan Society for Fuzzy Theory and Intelligent Informatics (SOFT)
Agent2d is used as a base program by many teams that participate in the RoboCup 2D Simulation League. Agent2d prepares a tree search using a series of actions called a chain action and a state evaluation function for a player agent to plan and determine the next action when the agent gets the ball. This paper realizes a return pass between a passer and a receiver in the framework of a chain action by producing better edges and nodes in the search tree and improving an evaluation function that evaluates the nodes. The new function evaluates a node, which represents a set of positions of players and the ball, from the standpoint that the passer can score if it kicks the ball towards the opponent goal at the current position. Our experimental results show the effectiveness of our proposed method.