Co-Evolving Influence MapTreeBasedStrategy GamePlayers

We investigate theuseofgenetic algorithms to evolve Alplayers forreal-time strategy games.Toovercome theknowledge acquisition bottleneck foundinusing traditional expertsystems, scripts, ordecision treeswe evolve players through co-evolution. Ourgameplayers areimplemented as resource allocation systems. Influence map treesareusedto analyze thegame-state anddetermine promising places toattack, defend, etc.Thesespatial objectives arechained tonon-spatial objectives (train units, buildbuildings, gather resources) ina dependency graph. Players areencoded within theindividuals ofagenetic algorithm andco-evolved against eachother, with results showing theproduction ofstrategies thatareinnovative, robust, andcapable ofdefeating asuite ofhand-coded opponents.