Simulated Environment in Robot Soccer

In the last two decades the concept of multi-agent mobile systems has been observed in many computer simulations, laboratory examples and in some practical applications. Among such systems robot soccer has shown to be a very popular research game and has served as a perfect example of multi-agent systems in the last few years (Ferber, 1999; Moss & Davidsson, 2002; Stone & Veloso, 2000). In this work the mathematical background of the developed robot soccer simulator is presented. The main purpose of the simulator design procedure is to obtain a realistic simulator which would be used as a tool in the process of strategy and control algorithms design for real world robot soccer as well as for other mobile-robotics related topics. To assure transferability to the real system the obtained strategy algorithms have to be designed on a realistic simulator. The main motivation for robot soccer simulator development was to design and study multi-agent control and strategy algorithms in FIRA Middle or Large League MiroSot category (5 against 5 or 11 against 11 robots). However, on FIRA’s (Federation of International Robot Soccer Association) official website (www.fira.net) there exists a simulator for SimuroSot league, which could only be used in Middle League MiroSot (5 against 5 robots). A similar simulator was built by (Liang & Liu, 2002) where robot motion is simulated by dynamic model, collisions remaining oversimplified. There also exist a number of other simulator applications but not many papers are available. An important part of every realistic robot soccer simulator is collision modelling and simulation. Good mathematical background in rigid body collisions modelling and simulation could be found in (Baraf, 1997). Another useful contribution in the field of robotic simulator is (Larsen, 2001) where collisions are treated by spring-dumper approach rather than by impulse force only. The use of spring-dumper linkage in collisions makes velocities changes continuous, which is less problematic for simulation than discontinuous change of velocities (Fremond, 1995) obtained by impulse usage. However, spring and dumper coefficients are not easy to identify. Moreover, when observed from macroscopic time scale (as it is in simulation) collisions are indeed discontinuous events. Simulated robots should have a realistic shape, which should not be represented simply with a square (the real shape of the robot is not a square) otherwise the simulation of ball guidance and other collisions becomes unrealistic. Furthermore, some of the available robot soccer simulators do not treat collisions well, especially the collisions among robots (robot corners), collisions between robot and boundary and situations where the ball is in-between