Abstract

The game balance as a whole and the combat balance, in particular, is crucial in the video games development process since an unbalanced system might prevent players from continuing to play a certain game. In simple words, the game can be too complicated or too simple for the players. In this thesis, we explore the benefits of applying agent-based modeling (ABM) as the design method for combat balancing in video games. We used ABM for prototyping and playtesting. Considering that the balancing process is not fast and relatively complicated, we assumed that ABM can speed up iterations and facilitate combat balancing in video game development. Herewith, we reviewed the definition of combat balancing with its common aspects. We also analyzed the iterative design in game development and the ABM features. Then, considering their characteristics and game balancing difficulties, we prototyped a boss battle between a player and enemies and playtested it. As a result of running the implemented 3D prototype many times, we generated a multiparametric data set. By investigating the outcome of the modeling and how the relationship between game elements affect it, we present how the balancing process of the combat system could be fine-tuned to achieve a desirable level of combat parameters.

Project Description

We assessed how ABM fits into the prototyping process of iterative game design. After that we concluded that ABM allows us to balance a combat system from a technical point of view, for instance, toward such constraints as the number of hits and the combat time limit. Such a method can be useful for balancing the combat system in RPGs with a lot of boss fights and player characters. Plots and visualizations of parameter values should help understand relationships between resources and interactions between agents. The implemented in this work ABM-based prototype with possibilities to turn on/off some mechanics demonstrates the tuning process for balance. The derived dataset consists of data points from 1000 simulations. To request more information about this project, please contact via email: margarita.kukhtevich@tum.de

Implementation

The prototype in Unity models the arena battle of a single player with one or five bosses. Randomness in actions and movements of the player and the bosses enables different scenarios. We can set the number of simulations and enemies, agents attributes, and time scale in the inspector window. The plot scene shows the downfall of HP and the win rate. 

Results

The simulations were carried out 1000 times each, with a different number of bosses and a different type of attack. For instance, when fighting the single boss, the player win rate was approximately 3 times less than the enemies, and the battle lasted about 2 minutes. Adding another four bosses and still using the single-target attack slightly increased the chances of winning. Combining the single-target attack with the AoE attack doubled the player win rate because the AoE attack was twice as effective in this situation. Such factors as how frequently enemies cluster and how long they keep clustering influenced the comparison of the effectiveness of different types of attacks. An example of the last situation is given (with 2 performed simulations).

example2simMP4.MP4

Conclusion

ABM is a low-cost method, fast and accurate to a certain degree compared to human playtesting. The topic has the potential to be further researched. Simple randomness in the player behavior can be replaced with the more complex and human-like behavior, which will give more accurate battle results (combination of ABM with other techniques e.g. machine learning). Besides, the automatic turning of parameter values could be researched.