Drawing on the fields of robotics, computer games and usability testing, this thesis contextualises the work of cyberneticist Valentino Braitenberg within the current field of game AI and demonstrates how there remain aspects of his propositions that cannot yet be adequately captured theoretically, or tested empirically or conclusively, in simulations of complex biologically inspired behaviours.
Justification for this stems from the view that believable, simulated characters are increasingly prevalent and exposed to mass-market scrutiny in visual entertainment media, such as visual effects in films and in particular the growing video games business, but will also have a growing, significant role in serious games and educational applications.
This thesis combines analytical, synthetic and empirical methods and discusses the contributions and limitations of each to the evaluation of believability in the context of human computer interaction scenarios. It first analyses Braitenberg’s more intuitive architectures and demonstrates how these can be adapted, using a synthetic approach, to function as controllers for believable virtual agents, in a series of typical gaming scenarios. It then presents a set of empirical studies that were performed to develop a method for evaluating agents for their ability to elicit suspension of disbelief in the user. The resulting method, which is user centred and combines a qualitative content analysis approach with believability metrics for virtual agents sourced from the literature, was used to evaluate a series of increasingly complex agent models in a simple game scenario. The data was then compared to the results of the analysis of the underlying agent architecture to determine the correlation between AI design intent and the resulting user reactions and observations gathered from the empirical study.
|Date of Award||Dec 2015|