AI in games is mainly used for:
- NPC behavior
- Custom/Procedural Content creation
The idea here is to work on a project that explores AI in games. We have identified the following potential approaches to this subject:
The use of academic AI in games is not that common. Usually, game developers try to focus on simpler solutions to mimic intelligent behavior -- such as FSMs and Behavior Trees. A few reasons behind this decision are:
- AI in games has to be "fun" and "beatable"
- Games have real-time restrictions and some methods simply require too much time
- Unpredictability of AI can sometimes be "bad" for gameplay and experience
However, a few games have already used academic AI methods and were succesful: Top 10 Most Influential AI Games Planning in Games Black & White Postmortem Paper on Black & White's AI
The goal here is to explore which "academic" AI methods have already been used -- or could be used -- in games and package them in some kind of AI Middleware. This would be benefitial because:
- Developers could explore different (and powerful) AI techniques
- Introduce dynamic behaviors, potentially removing the "scripted" feel of some games
As proof of concept, we would develop some sample games using the new framework.
Other useful links: Panorama of AI in Games Game AI Revisited Gamasutra Post Post on academic AI vs game AI Creating Interactive Characters with BDI Agents GDX-AI Framework (something like this but for academic AI) Artificial Intelligence in Video Games: Towards a Unified Framework
A good way to test AI solutions is using games. The goal here is to build a framework similar to:
That would allow AI developers to test new techniques of AI for a specific game genre.