On the use of reinforcement learning for testing game mechanics
Aaron Snoswell, Centaine L. Snoswell · QUT ePrints (Queensland University of Technology) · 2018
Researchers in Artificial Intelligence routinely use simulated environments as benchmarks for testing algorithms. We propose that this paradigm could be flipped whereby reinforcement learning “reward hacking” could provide a mechanism for testing game mechanics. Numerous literature examples demonstrate game exploits and implementation problems that have been discovered through optimization based techniques, often incidentally. Harnessing the power of reinforcement learning for computer game testing has the potential to optimize gameplay and reduce game defects or exploits that are released onto the market unintentionally, mitigating the economic impacts if they need to be patched once the game is live.