Reproduce “Augmenting Automated Game Testing with Deep Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Joakim Bergdahl, Camilo Gordillo, Konrad Tollmar, Linus Gisslén
Researchers at EA SEED developed a method to augment automated game testing using Deep Reinforcement Learning (DRL) agents, enabling them to discover game exploits and logical bugs in first-person shooter navigation. The DRL approach increased test coverage by exploring environments dynamically and provided a quantifiable metric for map difficulty based on agent training time, enhancing existing quality assurance pipelines.
Key insights
- DRL agents can effectively discover both game exploits (e.g., missing collision meshes allowing shortcuts) and logical bugs (e.g., areas where players can get stuck) by exploring behaviors beyond human-intended paths.
- Dynamic exploration by DRL agents during training significantly increases test coverage, identifying game states and paths that rigid scripted agents would typically miss.
- The training time required for DRL agents to master a specific navigation task serves as a quantifiable and objective metric for evaluating and comparing game map or scenario difficulty.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.