Designing Good Pursuit Problems as Testbeds for Distributed AI: A Novel Application of Genetic Algorithms

A basic N x M instance (game) of the Pursuit Problem is one in which N pursuing agents try to capture as many as possible of M prey agents by surrounding them, on a rectilinear grid. The 4 × 1 game has been considered as a testbed for comparing the effectiveness of different multiagent distributed architectures, and the 6 × 2 game has received a little attention. This paper reports a systematic exercise in evaluating the quality of pursuit games as potential testbeds for distributed artificial intelligence (DAI). Genetic algorithms (GAs) have been used both to optimise low-level architectural features of agents and to search the (N, M) space of games. The conclusion from experiments is that (M + 4) × M games have the right complexity to be good testbeds, provided that M > 4. Additionally, the paper demonstrates the usefulness of GAs as tools to help DAI designers, and argues that boredom is a concept that deserves consideration as a feature of general agent architectures.