The game’s bosses are designed to challenge the player, requiring them to strategize and master their skills in order to win. For game developers, the process of designing and balancing boss behavior is difficult and time consuming. The solution to this problem could be machine learning. This will allow developers to create bosses that are not only more challenging, but also more dynamic and adaptable to player behavior.
Traditional bosses are often pre-programmed with different attack patterns that are easily recognizable by experienced players. However, machine learning can be used to train the boss AI and use previous encounter data to improve its strategy. This could lead to dynamic and challenging battles, and players couldn’t rely on patterns to beat it.
Romain Trachel and Alexandre Peyrot are Eidos-Sherbrooke specialists in machine-learning. Eidos-Sherbrooke is a game development studio, which takes advantage of two nearby universities (University de Sherbrooke and Bishop’s University) and bring new computing technology from them into their games. At Unreal Fest 2022 Romain and Alexandre have demonstrated a game which combines machine learning with EQS (Environment Query System). EQS is a AI system in Unreal Engine 5 that collects data from the environment. EQS query can be used to make decision on how to proceed based on the results of developers’ tests. It can instruct AI characters to find the best possible location that will provide a line of sight to a player to attach, the nearest health or ammo pickup. In most games, behaviour trees decide about possibilities, but in this demo the EQS is providing information to AI character about its environment and machine-learning decides what is the best response.
The game demo is pretty simple, but yet shows a huge potential of this model. The player has to collect orbs across a map while being chased by enemy. The gameplay is almost the same as Pac-Man, but the chaser’s behaviours are not scripted. Here you can watch the whole presentation:
The use of AI in video games can sometimes result in gameplay that is not enjoyable for players. Romain and Alexandre experienced this in one of their game modes, where the enemy AI used a camping strategy that made the game unenjoyable. The goal of Romain and Alexandre is to find ways to incorporate machine learning into existing game without making the game too difficult or unfair for players.
In summary, machine learning can be a difficult technology to implement in video game development due to performance issues and animation challenges. Acquiring large amounts of training data is necessary for successful results, but this can be difficult to achieve in the open-world environments that players expect. Also modern game engines are not optimised for machine learning. Animation is another problem as AI driven bots may behave in unexpected ways that animation methods can’t accommodate.
https://www.wired.com/story/machine-learning-ai-game-development-bosses-enemies/
Using AI in video games could indeed make the game too hard, as AI learns and adapts very well. So, i believe it would be extremely difficult to find the golden mean and make the game challenging and exciting, but not impossible to win and enjoyable to play. Thank you for an interesting post!
Usage of the AI in the gaming industry is such a a cool idea !, it would improve the gaming expirence giving unlimited variations and possibilities for the people to enjoy. I think the greatest challenge here is to set some levels to it and make different difficulty levels. I think that over time, Ai will be a great solution, especially in RPG games.