Geoffrey F. Miller, Dave Cliff
Animals often chase each other about. These pursuit-evasion contests require the continuous dynamical control of complex sensory-motor behavior, and give rise to some of the most common and challenging co-evolutionary arms races in nature. This paper argues for the importance and fruitfulness of studying pursuit-evasion scenarios using evolutionary simulation methods, and reviews the relevant literatures; in a companion paper, our simulation methods and results will be presented. We first review the biological ubiquity of pursuit-evasion contests, the protean (adaptively unpredictable) behavior that often evolves in evasion strategies (e.g. when prey zig-zag to evade predators), and the relevant neuroethology of capture by predators and escape by prey. The differential game theory relevant to analyzing pursuit-evasion games is then reviewed, including the proven optimality of mixed strategies (corresponding to protean behavior) in many such games, and the difficulty of deriving analytical solutions in realistically complex games. Previous work on evolving pursuit-evasion tactics in simulation and in real robots is then reviewed, and directions for further research are identified. We conclude with some possible engineering applications and scientific implications of pursuit, evasion, and their co-evolution. Keywords: pursuit-evasion contests; co-evolution; protean behavior; differential game theory, mixed strategies; neuroethology.
Download compressed postscript file