Evolving coordinated spatial tactics for autonomous entities using influence maps

摘要

We evolve tactical control for entity groups in a naval real-time strategy game. Since tactical maneuvering involves spatial reasoning, our evolutionary algorithm evolves a set of influence maps that help specify an entity's spatial objectives. The entity then uses the A* route finding algorithm to generate waypoints according to the influence map, and follows them to achieve spatial objectives. Using this representation, our evolutionary algorithm quickly evolves increasingly better capture-the-flag tactics on three increasingly difficult maps. These preliminary results indicate (1) the usefulness of our particular influence map encoding for representing spatially resolved tactics and (2) the potential for using co-evolution to generate increasingly complex and competent tactics in our game. More generally, this work represents another step in our ongoing effort to investigate the co-evolution of competent game players in a real-time, continuous, environment that does not assume complete knowledge of the game state.