Enhancing Automated Video Game Regression Testing through Behavior-Driven Development and Imitation Learning

The evolution of video game development has led to increasingly complex game environments, presenting significant challenges for automated testing. Traditional manual testing methods struggle to keep pace with modern game development’s dynamic nature. We propose an automated testing approach integrating Behavior-Driven Development (BDD) with Reinforcement Learning (RL) and Imitation Learning (IL). Our framework leverages natural language specifications from BDD to define expected game behaviors, guiding RL-based agents in automated testing. By coupling BDD’s scenario definition with RL’s adaptive exploration, our approach facilitates scalable test creation while reducing reliance on handcrafted scripts. To accelerate training and improve performance, we incorporate IL, enabling agents to learn from expert demonstrations before transitioning to RL for fine-tuning. We validate our methodology through a case study using a Super Mario Bros clone developed in the Godot engine. Our evaluation highlights practical advantages such as reduced test development time, enhanced test coverage, and detection of complex game regressions. Despite these advantages, key challenges persist, including the design of effective reward functions and the computational demands associated with training reinforcement learning agents. Our findings demonstrate that combining BDD with RL and IL offers a promising solution for automating game testing, promising efficient testing pipelines and higher game quality.