Behavior Acquisition in RoboCup Middle Size League Domain
Yasutake Takahashi, Minoru As · 2007
This chapter briefly overviewed research activities, especially on behavior acquisition/ emergence based on machine learning techniques, in RoboCup middle size league. This research area has kept attracting people and been investigated not only in middle size league but also other ones such as simulation soccer and 4-legged leagues. Many results of applications of machine learning techniques to robots in the RoboCup domain show promising contributions to generate adaptive behaviors in real situations. On the other hand, many difficulties in practical use have also been unveiled so far. For example, selection of important features for purposeful behaviors, purposive behavior discovery through observation of others, self task decomposition and integration, rapid team strategy adaptation during a game, and so on, are to be investigated furthermore. One of the goals of RoboCup is "By the year 2050, develop a team of fully autonomous humanoid robots that can win against the human world soccer champion team."(Federation) A game of a middle size league robot team v.s. a human team was demonstrated in RoboCup2007 Atlanta USA. The human team showed much better performance than the robot team although it won the championship in the middle size league this year. Taking achievements in the past decade into consideration, however, we foresee that robots will play soccer with human players, learn many skills, cooperative/competitive behaviors, team coordination, positioning in the teams, fast adaptation of team strategy, and so on through interaction during games, and finally a robot team beats the human world soccer champion team.