Resumen de Intelligent generation and control of interactive virtual worlds
This thesis advocates the use of non-gaming virtual worlds as a significant future technology for the domains of e-* applications (e- learning, e-commerce, e-government) and social simulations. In such systems, a 3D virtual environment is often populated by a large number of inhabitants that can be either human-controlled avatars or intelligent virtual agents (avatars controlled by autonomous software entities) who engage in advanced interactions with their virtual environment and other participants.
One significant problem that impedes wide adoption of the virtual worlds technology for these problem domains is that virtual worlds in general are difficult to build, and significant effort has to be put into designing the 3D virtual environment and programming virtual agents; but even harder is to ensure the validity of participant interactions in such environments and enforce social norms on their inhabitants so that unauthorized behavior can be prevented.
To address this problem, we have developed a comprehensive technological solution that automates the design of such virtual worlds and its population with virtual agents. Our approach is based on the utilization of Virtual Institutions, which are virtual worlds with normative regulation of participant interactions.
The key focus of the thesis is on explaining how existing methods of formal specification of Virtual Institutions can be extended to automatically translate the institutional specification into an interactive 3D environment using the shape grammars approach and automatically populating such environments with virtual agents.
Shape grammars represent a powerful visual technique for creating procedural 2D and 3D designs, but existing work was not immediately suitable for our problem. Existing shape grammar solutions are normally restricted to very specific scenarios, do not normally address interactivity of the generated designs and rarely consider facilitating agent enactment of the generated environments as well as their normative regulation.
Thus, we have extended existing work and developed the Shape Grammar Interpreter framework, which addresses the limitations of existing solutions. This framework was further utilized for developing the concept of virtual world grammar, which is a sub-set of shape grammars targeting automatic generation of normative virtual worlds. As the result of this dissertation, Virtual Worlds Grammars constitute a strong formalization and a development environment not only enabling automatic generation of normative virtual worlds, but also their platform independent deployment (using the VIXEE infrastructure that has been developed as an important part of this dissertation).
Another significant contribution of this thesis is developing a mechanism of automatic population of the generated environments with an arbitrary number of software agents, which are capable of intelligent interactions with 3D objects placed in the environment. Moreover, these agents are able to collaborate with human-controlled avatars, facilitate their problem-solving and ensure that all agent actions strictly adhere to social norms of the given institution.
For this purpose we have developed a generic model of virtual agents, which enables an agent to be situated within any normative virtual world, generate its own goals based on its current physiological and psychological needs, as well as to dynamically generate plans for satisfying these goals using the underlying institutional specification. The institutional specification in this instance provides a high-level representation of an agent's interaction possibilities.
To illustrate the usefulness of methods and techniques presented in this thesis, we have applied it to the domain of historical simulation, re-enacting everyday life of ancient people in one of humanity's first cities, the city of Uruk. The existing design of the Uruk city was enriched with dynamically generated artefacts and a large crowd of virtual agents simulating ancient citizens from different classes of the Uruk society. We showed how our approach allows to create a desired number of visually and behaviorally diverse agents, as well as dynamically generating food, tools and other items that they can utilise to satisfy their goals, while acting in a historically authentic manner.
