Minimizing Computational Overhead while Retaining Gameplay Effectiveness in Starcraft Bots
David Kipnis, Wesley Deneke, Matthew Mitchell, Nils Rys-Recker · 2022
Over the past decade, RTS games like Starcraft have emerged as testbeds for developing complex artificial intelligence. While several bots have been successful in their objective of beating opponents, no research has sought to minimize their computational resource usage. Reducing a models' computational complexity has the potential to expand the range of applications for such AI, which is the primary objective that VikingBot was designed to explore. By combining a heterogeneous agent model with a machine learning approach, VikingBot demonstrates that a Starcraft bot can minimize resource usage while retaining good gameplay effectiveness.