PhD. thesis



General informations


Jury


Abstract


Many tools for creating virtual worlds focus on the interaction between the users (represented by avatars), but a few of them let non-programmers (such as video game scriptwriters or biology researchers) easily describe the behaviours of the autonomous entities that populate virtual worlds. Our aim is to provide such tools, as a toolkit named InViWo (Intuitive Virtual Worlds); this PhD. thesis lays the foundations of the InViWo toolkit.

We consider a virtual world as an homogeneous multi-agent system, in which some agents (called avatars) can be partly controlled by users. Any object of an InViWo world is thus an agent, which is able to perceive its environment in order to react to events by executing actions into the world, depending on its internal state and its own motivations. We therefore propose a generic, dynamic architecture for virtual agents. An InViWo agent is made of characteristic attributes, sensors to continually perceive the modifications of the environment and the internal state, a decision process, and effectors to realize internal and external actions.

We have furthermore designed a distributed architecture for action selection, which is based on both reactive behavioural modules and a generic mechanism for arbitration that combines the decisions made by the concurrent modules. The dynamic network formed by the behavioural modules works according to the reactive synchronous model. We propose the Marvin language, strongly inspired by the Esterel language, in order to define behavioural modules; an InViWo agent can be entirely described in Marvin. We developed an execution platform, which has been tested on several examples specified in Marvin.