An Approach of Real-Time Team Behavior Control in Games

The design of NPC (non-player character) is an analytic process. It is relying on assumptions of human game players´ behavior. In practice, however, different PCs (player characters) often exhibit variable behavior, making them difficult to predicate and complicating the design process. In this paper, we describe an approach for team AI planning and learning. This approach is based on procedural knowledge and a layered multi-agent architecture. We implement real-time transfer learning and adaptive mechanism for the team of NPCs. The team can react to the human player with the tactical awareness of seasoned team behavior. Results indicate that the approach of using the hybrid of transfer learning and adaptive mechanism can improve NPCs´ overall performance in real-time.