Emergent cooperative strategies for robot team sports

Automatic development of cooperative strategies far teams of distributed au tonomous robots, or software agents, is presented in this paper. It is show n that a team of robotic agents that initially plays a random game of simul ated soccer, can acquire winning strategies through successive generations, utilizing techniques of evolutionary computation. The concept of Tropism-b ased Control Architecture is introduced that not only allows for the evolut ion of cooperative strategies, but also obtains the acquired knowledge in a format that is easily comprehensible by humans. The advantage of this appr oach is that the cooperative strategies can then be transported onto a vari ety of platforms for testing and deployment. It is discussed as to why the game of robot soccer provides a good environment for this type of investiga tion, and how the presented concepts can have applications in multi-robot s ystem design. The proposed cognitive architecture has been inspired by biol ogical systems, and the paper includes a review of the related literature i n the field of evolution, both in the framework of animal societies, and mu lti-robot teams. The results of many generations of simulated evolution are presented, accompanied by the game results and fitness characteristics. A number of soccer game strategies which are developed through the experiment s are also described, where the obtained techniques for playing better team sports emerged as the result of evolutionary computation.