Evolutionary group robots for collective world modeling

This paper is concerned with collective problem-solving with a group of autonomous agents. The major challenges addressed are (1) how to enable the distributed agents to dynamically acquire their goal-directed cooperative behaviors in performing a certain task, and (2) how to apply the methodology of group learning to handle ill-defined problems, e.g., problems without complete representations. In response to those challenges, we have developed an evolutionary self-organization approach to collective problem-solving, and furthermore, demonstrated the implemented approach in tackling a specific problem in intelligent robotics; namely, how to collectively build a global spatial representation, e.g., an artificial potential field map, of an unknown environment. In the paper, we shall present both the detailed formulation of our approach and the results of empirical validation in the context of world modeling with a group of distributed autonomous robots.