Diversity voting and its application in real-time strategy games multi-objective optimization decision-making behavior modeling

In this paper, we present a novel approach based on diversity and voting for studying the Multi-Objective Optimization (MOO) problems in Real-Time Strategy game (RTS). MOO is one of the most important research topics in Decision-Making Behavior Modeling (DMBM) of RTS. It includes but is not limited to Tactical Position Selection (TPS), Force and Fire Allocation, Targets Selection and Path Planing, etc. The computing efficiency of multi-objective optimization algorithms is of vital importance to the believable representation of RTS agents decision-making behaviors. Currently however, most MOO techniques used in Game AI decision-making behavior modeling cannot satisfy the great demand in future RTS game AI in aspect of time efficiency. According to above problems and taking the TPS decision-making behavior in RTS game AI as research background, the paper proposed a novel MOO-oriented Diversity Voting model based on Group Diversity and Social Choice theory. Through forming diverse problem solving team, agents compute and further propose individual solutions and using voting method for group opinion aggregation. Contrary to traditional Diversity Voting models, the paper fully considers the feature of multiple Pareto results in MOO problem solving and establishes our model based on Approval Voting. More importantly, we change the traditional parameter-based voting into solution-based method during opinion aggregation among problem solvers. We formalize the TPS problem in MOO manner, then carry out the problem solving tests based respectively on single MOO algorithm and on our Diversity Voting model. Further, we take advantages of little interactions among problem solvers and test the problem solving ability of model under parallelized conditions.