%0 Journal Article
%T Evolving Neural Network Controllers for a Team of Self-Organizing Robots
%A Istv¨˘n Feh¨¦rv¨˘ri
%A Wilfried Elmenreich
%J Journal of Robotics
%D 2010
%I Hindawi Publishing Corporation
%R 10.1155/2010/841286
%X Self-organizing systems obtain a global system behavior via typically simple local interactions among a number of components or agents, respectively. The emergent service often displays properties like adaptability, robustness, and scalability, which makes the self-organizing paradigm interesting for technical applications like cooperative autonomous robots. The behavior for the local interactions is usually simple, but it is often difficult to define the right set of interaction rules in order to achieve a desired global behavior. In this paper, we describe a novel design approach using an evolutionary algorithm and artificial neural networks to automatize the part of the design process that requires most of the effort. A simulated robot soccer game was implemented to test and evaluate the proposed method. A new approach in evolving competitive behavior is also introduced using Swiss System instead of the full tournament to cut down the number of necessary simulations. 1. Introduction The concept of systems consisting of multiple autonomous mobile robots is attractive for several reasons [1]. Multiple cooperative robots might be able to achieve a task with better performance or with lower cost. Moreover, loosely coupled distributed systems tend to be more robust, yet more flexible than a single powerful robot performing the same task. A benefit of the collaborative interaction of mobile robots can be an emergent service, that is, a progressive result that is more than the sum of the individual efforts [2]. A swarm of robots can thus build a self-organizing system [3]. The continuous technical development in robotics during the last decades has provided us with the hardware for swarms of small, cheap autonomous devices [4¨C6]. However, designing the behavior and interactions among the robots remains a very complex task. Using a standard top-down design approach with fixed task decompositions and allocation typically leads to systems working only for a small set of parameters. On the other hand, effects like changing environments or breakdowns and faults of hardware require a robust and flexible solution that provides a useful service for many possible system states. An alternative to the classical design approach is to organize the robots as a self-organizing system performing the intended task. Thus, the robots achieve a global system behavior via simple local interactions without centralized control [7]. As shown by many examples in nature, simple rules for interactions can emerge to quite complex behavior while being scalable and robust against
%U http://www.hindawi.com/journals/jr/2010/841286/