Interactive Training for Synthetic Characters

摘要: Compelling synthetic characters must behave in ways that reflect their past experience and thus allow for individual personalization. We therefore need a method allows to learn. But simply adding traditional machine learning algorithms without considering the characters’ own motivations desires will break illusion of life. Intentional require interactive learning. In this paper, we present results Sydney K9.0, project based on Synthetic Characters creature kernel framework. Inspired by pet training, have implemented character can be trained using “clicker training” technique. Clicker training utilizes natural an animal employs operant conditioning procedures shaping behavior. The necessary plasticity system interconnections shaped associations rewards is required clicker was integrated into also includes module named DogEar designed collecting real-world acoustic data, such as human voice commands, kernel’s perception system. This provides seamless interface between simulated real worlds. Detailed implementation interaction are presented.