Developing AI to Improve the Efficiency of Quality Assurance in Game Development

game

Project Summary

Araya's Autonomous Agent team is working to improve the efficiency of QA (Quality Assurance) in game development by applying machine learning and reinforcement learning technologies.
[2020.5.19] Contents updated.

background

Improving the efficiency of QA is an urgent issue in game development, which is becoming increasingly large and complex, and various efforts to automate test play have begun in the game industry.
To address this issue, ARAYA is aiming to realize more advanced and efficient QA methods by utilizing "autonomous agents" that behave actively based on machine learning and reinforcement learning technologies for game test play.

Benefits of Using Autonomous Agents for Test Play

An "autonomous agent" is an AI that can judge situations and perform tasks on its own. It learns game operations through trial and error by playing games on its own (reinforcement learning) or by learning from model data given by humans (imitation learning).
Conventional automation technology requires detailed writing down of procedures in the form of scripts and macros, but with the use of autonomous agents, this process is no longer necessary, and we believe that it will be possible to conduct a variety of tests more easily.

GAME AI

Application example 1: Automation of play through

Conventional QA efficiency/automation technologies include writing and executing test scripts such as RPA, but due to the randomness of the game and disturbances such as communication delays, scripts created once may no longer work effectively.
Autonomous agents based on reinforcement learning technology are expected to be highly robust against randomness and disturbances, since they constantly observe their own state and select the optimal action according to the state. For this reason, we believe that autonomous agents will be highly effective for test items that require complex and lengthy operations, such as play-throughs.

autonomous agent

Application example 2: Support for balance adjustment

In games that require frequent updates, such as social games and online games, the number of content combinations snowballs, making balance adjustments increasingly difficult.
Therefore, by having autonomous agents automatically and repeatedly test-play games, we can help evaluate whether the balance and difficulty of new content is appropriate. This will shorten the time required for balance adjustment and also reduce the number of adjustment errors.

GAME AI

Use Case 3: Creating a Clone Player

In the case of online games, if the developer/operator has a log of player operations, it is possible to create a clone that mimics the behavior of the player from which it was learned by imitating and learning the agent based on the operation log. For example, the cloned player can be used in the following ways
Create clones for each player's skill level and use them in test play to help balance the game.
Clone players can be used as part of the in-game content to fill in when there are not enough players in a competitive or cooperative game.

For the future

ARAYA IS WORKING TO VALIDATE AI/REINFORCEMENT LEARNING TECHNOLOGY. WE ARE ALSO LOOKING FOR PARTNER COMPANIES IN THE GAME INDUSTRY TO SOLVE SPECIFIC ISSUES IN THE GAME DEVELOPMENT FIELD.

Please also see the following conference presentations on this project.
Yusuke Tanimura, Masahiro Yasumoto: "A Study on Practical Application of Reinforcement Learning Technology for QA Efficiency in Game Development", Proceedings of the 10th Annual Conference of Digital Game Research Association of Japan (DiGRA JAPAN), pp.84-87, 2020 .
The 10th Annual Conference of Digital Game Research Association of Japan Web page: http://digrajapan.org/conf10th/