Published 27 August 2020

DC Labs and IGGI at CoG 2020: Automated balancing in multi-agent games

Researchers from the Centre for Doctoral Training in Intelligent Games and Game Intelligence (IGGI) and DC Labs have presented the first steps towards a robust tool for automated balancing in multi-player games.

The study was presented at the online 2020 IEEE Conference on Games.

Achieving game balance in videogames (where character strengths are balanced by a corresponding weakness in another area) is a major concern for game designers. Largely a manual process of trial and error, game balancing is especially problematic in multi-player games where perceived fairness has a drastic impact on player experience.

Game balancing takes a lot of time and resources. It is reliant on human intuition and expert knowledge to estimate how changes in game mechanics affect emergent gameplay. Human play testing as part of this process is time consuming and requires many human testers for long-play sessions, which grow longer with more complex games.

While in-house tools can be built for the adjustment and authoring of individual game elements, balancing and adjusting meta-game elements presents a wider challenge. For the purpose of this study, meta-game components are defined as high-level strategies that are separate and additional to core game play.

The application of data-analytics presents one approach to balancing meta-game components, however it is only possible for games with large quantities of player data. Furthermore, analytics can only discover balance issues in content that is already live, and by that point the issues may have negatively impacted on player experience.

As part of the study, the team developed an algorithm to auto balance a game, as requested by a designer. The approach requires a designer to construct an intuitive graphical representation of their meta-game target, representing the relative scores that high-level strategies should experience.

Combining concepts from AI for game playing, optimisation, game and graph theory, researchers demonstrated the capabilities of their tool on examples based on a simple Rock-Paper-Scissors game, and on a more complex asymmetric fighting game.

The contributions of this research could be transformed into the “backend” of an actual tool, with a user friendly “frontend” developed to make it practical for non-technical users.

Link to academic paper - http://eprints.whiterose.ac.uk/162107/1/Autobalancing_CoG_paper_3_.pdf