A Micromanagement Task Allocation System for Real-Time Strategy Games

Real-time strategy (RTS) game play is a combination of strategy and micromanagement. While strategy is clearly important, the success of a strategy can depend greatly on effective micromanagement. Recent years have seen an increase in work focusing on micromanagement in RTS AI, but the great majority of these works have focused on policies for individual units or very specific situations, while very little work has aimed to address the need for a broadly applicable structure for unit group coordination. This paper conceptualizes RTS group level micromanagement as a multiagent task allocation (TA) problem, and proposes the micromanagement task allocation system (MTAS) as a framework to bridge the gap between strategy and unit level micromanagement. MTAS fits into the common layered RTS AI structure, and can thus, in principle, be integrated into any existing system based on this layered structure. We describe the integration of MTAS into E323AI (an existing AI player for the spring RTS engine), and provide empirical results demonstrating that MTAS leads to statistically significant improvement in performance on the popular RTS game Balanced Annihilation.