Testing is a major effort for the gaming industry, requiring significant budget and manpower. This case study focuses on the development version of the ice hockey game EA SPORTS NHL 26, specifically testing the goalie AI for behavioral exploits.
To reduce the effort of re-testing the goalie AI after every game or behavior modification in the development phase, we propose Reward-Adaptive Iterative Discovery (RAID), a novel approach that leverages iterative Reinforcement Learning (RL) to automatically discover exploits by training a population of goal-scoring agents.
While previous methods have successfully identified exploits, RL algorithms often overfit to a single solution. We introduce a simple extension on existing RL algorithms to find multiple diverse high-quality solutions.
In our first deployment of this approach, we were able to find six hockey scoring exploit strategies in a single experiment that were qualitatively similar to those discovered by human playtesters during hours-long manual testing sessions.
Blogger's Review: The RAID method proposed in this study significantly enhances the efficiency of game testing, particularly in discovering diverse strategies, which is of great practical importance. The innovation and application potential of this approach are noteworthy, and it may be adopted in more game development projects in the future.