A paper authored by two Purdue Polytechnic professors and two Ph.D. candidates has earned the 2023 Best Paper Award by ACM Transactions on Interactive Intelligent Systems (ACM TiiS), a journal focused on intelligent interactive systems research.
The authors studied collaborative gameplay in a virtual reality game to determine if artificial intelligence (AI) agents could evaluate the degree of collaboration required to finish a game level. A typical game design process can be tedious because the level of collaboration is manually built, but AI agents may be able to help expedite it.
Before the study, the authors were not aware of any available methods for evaluating the degree of collaboration—a necessary step in automating the building process for collaborative game levels. They developed a method that involved building playable levels, collecting data, and asking participants to rate their experiences playing the levels.
Their method was broken into three stages:
- 1) a game designer built playable chunks of a collaborative video game;
- 2) the researchers determined the level of player collaboration required in each chunk using two AI agents; and
- 3) they developed a method for automatically synthesizing game levels using those chunks based on a game designer’s specifications.
After the game chunks were created, undergraduate and graduate students were paired up to play three synthesized levels that required low, medium, and high collaboration. Their goal was to exit each level as quickly as possible by completing tasks like pushing buttons, distracting enemies, collecting items, pushing chests, and climbing ladders.
“Collaborating with a game designer to create game level chunks was an essential part of our process,” said Minsoo Choi, a co-author and Ph.D. candidate at the time of the study. “It allowed us to integrate expert insights into the design while ensuring each chunk offered a unique collaborative challenge. The designer's input was invaluable in shaping the overall experience, ultimately leading to a more fun and engaging environment for testing our AI-driven synthesis method.”
The team found that AI virtual agents could successfully characterize the collaborative behavior and that their unique method could be used to automatically produce levels of a multiplayer game. The degree of collaboration in their game levels also impacted the way participants collaborated in the application, and most people said they “considerably enjoyed” their experience.
They plan to continue their work by evaluating less structured game levels and evaluating whether collaborative games can be used as training tools for improving collaborative behavior in players.
“Winning this award is a tremendous honor and validates our approach to leveraging AI in virtual reality game design. It motivates us to push the boundaries even further,” Christos Mousas said. “Our future work will explore more complex and less structured game levels, as well as the potential of collaborative games to serve as training tools for improving teamwork skills in real-world scenarios. We are excited about the possibilities and look forward to contributing even more to the intersection of virtual reality and intelligent interactive systems.
Christos Mousas, Dominic Kao, Huimin Liu, and Minsoo Choi co-authored the paper, which was published in March 2023. Kao is an assistant professor and director of the Virtual Futures Lab. Mousas is an associate professor in Purdue Polytechnic’s Department of Computer Graphics Technology and director of the Virtual Reality Lab. Both Liu and Choi were Ph.D. candidates at the time of the study, and Liu has since earned her doctorate degree.
Additional information
-
Polytechnic founds new Applied AI Research Center, appoints founding directors (Purdue Polytechnic newsroom)
-
Kao wins NSF grant, studying methods to level playing field in virtual education (Purdue Polytechnic newsroom)