Synergizing Code Coverage and Gameplay Intent: Coverage-Aware Game Playtesting with LLM-Guided Reinforcement Learning

The SMART (Structural Mapping for Augmented Reinforcement Testing) framework introduces a novel methodology for automating game quality assurance by bridging the gap between low-level structural code analysis and high-level functional gameplay intent. By leveraging large language models to interpret abstract syntax tree differences, the system decomposes game updates into actionable subgoals and structural anchors. This integration allows reinforcement learning agents to navigate complex game environments while simultaneously prioritizing the validation of newly modified code branches, effectively overcoming the limitations of traditional testing methods that often struggle to balance functional stability with deep code exploration.

Experimental evaluations conducted within the Overcooked and Minecraft environments demonstrate that this hybrid reward mechanism significantly outperforms standard reinforcement learning baselines. The framework achieves a 94% branch coverage rate and a 98% task completion rate, proving that agents can successfully target edge cases and deep code paths without sacrificing core gameplay objectives. Ablation studies confirm that the synergy between semantic subgoals and structural rewards is essential, as neither approach is sufficient in isolation to ensure both comprehensive coverage and functional reliability.

While the framework provides a robust, automated solution for modern game development, its efficacy remains contingent upon the accuracy of LLM interpretations and the precise tuning of reward weights. Current implementation strategies focus primarily on code-based logic, leaving non-code asset testing as a secondary consideration. Nevertheless, by automating the alignment of structural code changes with gameplay behavior, this approach offers a scalable path forward for game studios seeking to reduce manual testing burdens while maintaining high standards of software integrity across complex, evolving game architectures.