Model-Based Control of Destructable Flying Objects in Computer Games

Destructable environments and objects in computer games make it difficult to use traditional game AI techniques for computer controlled agents. This paper presents a method to control the velocity of destructable flying objects with the help of system identification and model-based action extraction. A training scheme is developed that makes it possible to identify all parameters in isolation using single variable linear regression. The experimental results indicate that this method works well in terms of control performance although some of the performance must be sacrificed in order to create a good visual appearance and to construct a more robust controller when faced with damages.