Digital Key Technologies

There is a long tradition of the development and application of artificial intelligence (AI) in the games industry. AI makes game worlds increasingly realistic. Virtual opponents and co-players learn to adapt to the course of the game and react comprehensibly to the actions of the players in front of the PC or the game console. In addition, developers are increasingly relying on AI when testing games software. Whereas in the past, all sorts of situations had to be simulated by humans, today, an essential part of this work can be done by AI.

One downside to this is that the more powerful the algorithms become, the more difficult it becomes for games developers to understand the strategies of virtual agents. However, understanding the test results is essential to improving the games. The research group of Prof. Fabian Beck is working on a new approach to solving this problem. The scientists want to use suitable visualizations to make the behavior of the artificial agents understandable. In doing so, they are concentrating on literally rolling out the temporal behavior that takes place on the screen during a game and making it readable. With this help, games developers could, for example, better balance difficulties in games or improve the artificial agents themselves.

The researchers have already achieved initial successes. Doctoral student Shivam Agarwal, together with Dr. Günter Wallner from the Eindhoven University of Technology, has developed a visualization for analyzing Pommerman games. Pommerman serves a test environment for machine learning researchers to train the cooperative and enemy behavior of AI-controlled agents. In a survey of AI and games experts, the visualization tool was found to be useful for understanding agent behavior and the underlying AI algorithms and improving their performance.

The scientists now want to make their approach applicable for further games. Suitable visualizations could enable games developers to get more knowledge out of their test data in the future. This could benefit more than just the games industry, which is known to be a driver and important test field for AI research.

Further Information

Contact

Shivam Agarwal, E-Mail: shivam.agarwal (at) paluno.uni-due.de, Tel.: +49 201 183 4654

Publications

Agarwal, Shivam; Herrmann, Christian; Wallner, Günter; Beck, Fabian: Visualizing AI Playtesting Data of 2D Side-scrolling Games. In: Proceedings of IEEE Conference on Games (to appear). 2020.

Agarwal, Shivam; Wallner, Günter; Beck, Fabian: Bombalytics: Visualization of Competition and Collaboration Strategies of Players in a Bomb Laying Game. In: Computer Graphics Forum, Jg. 39 (2020) Nr. 3, S. 89-100. doi:10.1111/cgf.13965