DARPA Gamebreaker AI Exploration Program
The Gamebreaker program seeks to develop and apply Artificial Intelligence (AI) to existing open-world video games to quantitatively assess game balance, identify parameters that significantly contribute to balance, and explore new capabilities, tactics, and rule modifications that are most destabilizing to the game.
– Blue Wave was given top marks by DARPA, placing in the top 3 in 4 categories.
– A Genetic Algorithm guided our autonomous game playing.
– Model interpretability techniques such as SHAP and LIME illuminated the “why’s” of gaming wins, losses, and draws.
– The impacts of weapon capabilities, costs, and commander strategies were explored and explained.
– Methods to maximally disadvantage opponents were developed.
– Our trained ANNs were able to predict outcomes with a 97% accuracy.
– We were able to prove that our method of modeling is extensible from a complex RTS game to a completely different and even more complex RTS game. Furthermore, we were able to leverage the data from the first game by using Transfer Learning.
– We were able to explore the action space using automated ‘Bot’ players of different skill levels and discover the relationship between military unit and weapon advancement and the training required to take advantage of them.
– We were able to develop an infrastructure capable of simulating 5000 years of two humans playing head-to-head 24/7/365 and gather a myriad of data for training Artificial Neural Networks (ANNs):
High-level architecture for running 5000+ years of simulations guided by a Genetic Algorithm