Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
New methods for competitive coevolution
Evolutionary Computation
| Hi-index | 0.00 |
This paper evaluates a coevolutionary genetic algorithm's performance at generating competitive strategies in the initial stages of real-time strategy games. Specifically, we evaluate coevolution's performance against an exhaustive search of all possible build orders. Three hand coded strategies outside this exhaustive list provide a quantitative baseline for comparison with other strategy search algorithms. Earlier work had shown that a bit-setting hill-climber only finds the best strategies six percent of the time but takes significantly less time compared to a genetic algorithm that routinely finds the best strategies. Our results here show that coevolved strategies win or tie against hill-climber and genetic algorithm strategies eighty percent of the time but routinely lose to the three hand coded baselines. This work informs our research on improving coevolutionary approaches to real-time strategy game player design.