A Language-guided Acceleration Method for Smoke Testing of Game Quests

Game quest is a common play in a massively multiplayer online role-playing game (MMORPG), which requires a player to take a series of actions and achieve multiple goals to accomplish it. During the game development life cycle, game companies often create and maintain a large number of releases, updates and tweaks of the game. There is a critical need for efficient smoke testing techniques to swiftly verify the core functionalities of the game. However, due to the vast number of game quests, existing smoke testing techniques typically incur a lot of time and labor costs. In this paper, we present an automated testing framework, named LAnguage-Guided somkE Testing (LAGET), which combines online planning with natural language models to efficiently find optimal policies for accomplishing the quest so that the testing process can be significantly accelerated. First, the online planning is, essentially, a model-based approach, and it dynamically constructs a game environment model, and computes the optimal solution for the smoke testing. Second, LAGET leverages the text description of the quest and adopts natural language models to guide better exploration in the game. Experimental results in both an open-source mini-game and two real industrial MMORPGs demonstrate that our method outperforms baseline methods by around 30% to 60%.