Research And Implementation Of Legged Robot Strategy System In RoboCup

In recent years, RoboCup becomes a keen research focus in the intemational AI,robotics and related fields. It provides a typical background and common issue for theresearch on MAS in dynamic unpredictable environment. Strategy system decides thelevel of robot intelligence. And it's vital to win the game when every functionalmodule in lower layer is well achieved. Aim at RoboCup legged robot soccercompetition, this paper discusses the designs of individual agent software architectureand hierarchy of strategy system. Moreover, this paper analyzes several key problemsof strategy system design by taking experiments on Sony ERS-7 ABIO robot, whichare the design of skills, roles and multi-agent coordination. On the base of that, thispaper presents the robot strategy system of Tongji University legged robot team(TJ-Ark), which was used in RoboCup2006 legged robot soccer competition.The design of robot strategy system has to take the architecture of robot systemand other modules in robot system into account. Aim at RoboCup competitionwhich is a dynamic unpredictable environment, this paper presents softwarearchitecture of legged robot, which consists of five subsystems, such as locomotion,vision, localization, communication and strategy. The software architecture fulfills therequirements of distributed multi-robot system in RoboCup. Besides, the design ofskills, the planning of robot role behavior, and the realization of dynamic roleassignment and coordination are vital problems in strategy design. According to theanalyses of reactive and deliberative agent architecture, strategy system adoptsreactive-deliberative hybrid architecture, and is realized as a couple of software layers,namely skill strategies and high-level strategies. The design of high-level strategiesadopts deliberative architecture, while the realization of skill strategies adoptsreactive architecture. Practices prove that this hybrid architecture can combineadvantages of both architectures. And then, robot can not only adapt to environmentquickly, but also has the ability of planning and dynamic task assignment.The design and implementation of skill strategies makes robot having basicintelligence and reactivity to environment. It's the foundation of entire strategy system. This paper analyzes the concepts and designs of five basic skills used incompetition, such as Ball Finding and Tracking, Ball Approaching, Ball Grabbing andActive Localization. The key to realize Ball Finding and Tracking skill is to controlrobot's legs and head together. The skill makes robot can find ball quickly and tracksteadily. Considering the cases in which the ball is near own goal or field edge, Getbehind Ball is designed for robot to avoid kicking the ball into its own goal orknocking it out of field incautiously. Ball Grabbing is implemented by controllingrobot's head, chin and forelegs together. By executing this skill, robot won't lose balleven if opponents jostle it. Ball Dribbling is designed for robot adjusting its positionwith ball to desired attack direction, and improves the chance to goal. ActiveLocalization is designed according to localization covariance. It improves theprecision of localization and ensures right decision making.Attacker role behavior is designed by marking off the field. The design ofgoalie role behavior adopts newly designed strategy structure and positioning method.On the basis of shared information, this paper presents attacker role bidding function,and applies the artificial potential field method to the positioning of supporter anddefender, and finally realizes dynamic role assignment and coordination in RoboCuplegged robot soccer competition. Experiment results demonstrate that multi-robotteam using our dynamic role assignment and coordination method outperforms anindividual robot and a team of individually skilled robots without coordination.According to the design concepts of skills and roles mentioned above, leggedrobot strategy system is realized basing on the hardware and software environmentprovided by Sony Corporation, and performs well in RoboCuo2006.