A Gene Expression Programming Inspired Evolution Symbiont Agent for Real-Time Strategy Generation

Abstract

AdaptiveSGA is a method for achieving Adaptive Game Artificial Intelligence-Based Dynamic Difficulty Balancing through the Symbiotic Game Agent Model. Previous work has shown that AdaptiveSGA can achieve Dynamic Difficulty Balancing in simulated soccer by effectively changing a team's strategy based on the opponent's performance. AdaptiveSGA pre-existing strategies and switches between them during runtime to increase the game's replayability by adapting the challenge it poses to the human player. Although this method works, its limitation is that if the human player surpasses the most intelligent strategy of the computer opponent, there is no way for the model to generate a new strategy during runtime that can potentially overcome the human player. AdaptiveSGA can only maintain engagement with the human player if the human player has not overcome the best strategy for the pool of pre-existing strategies. Current work addresses this limitation by introducing an Evolution Symbiont Agent whose purpose is to generate new strategies in real-time (during gameplay) through evolutionary mechanisms using Gene Expression Programming. Experimental results show that the presence of the evolution symbiont agent can use Gene Expression Programming to generate strategies capable of outperforming an opposing strategy.