Dynamic behavior-based control and world-embedded knowledge for interactive artificial intelligence

Video game designers depend on artificial intelligence to drive player experience in modern games. Therefore it is critical that AI not only be fast and computationally inexpensive, but also easy to incorporate with the design process. We address the problem of building computationally inexpensive AI that eases the game design process and provides strategic and tactical behavior comparable with current industry-standard techniques. Our central hypothesis is that behavior-based characters in games can exhibit effective strategy and coordinate in teams through the use of knowledge embedded in the world and a new dynamic approach to behavior-based control that enables characters to transfer behavioral knowledge. We use dynamic extensions for behavior-based subsumption and world-embedded knowledge to simplify and enhance game character intelligence. We find that the use of extended affordances to embed knowledge in the world can greatly reduce the effort required to build characters and AI engines while increasing the effectiveness of the behavior controllers. In addition, we find that the technique of multi-character affordances can provide a simple mechanism for enabling team coordination. We also show that reactive teaming, enabled by dynamic extensions to the subsumption architecture, is effective in creating large adaptable teams of characters. Finally, we show that the command policy for reactive teaming can be used to improve performance of reactive teams for tactical situations.