Semantic Virtual Environments

World Model

[Agent does not understand scene graph] The world model represents the contents of a virtual environment. It typically has the form of a scene graph, with lots of detail for realistic graphical rendering. A user can easily interpret the rendering via the human senses, to form a mental model of its meaning. An agent however does not have this cognitive ability, and will find it difficult to glean such information from the raw data structures. It might understand one particular world model, but then it cannot work in environments based on different world models - even if they belong to the same application domain.

[Agent understands SVE description] In contrast, the SVE description is an RDF graph that explicitly exposes the semantics of an environment - the entities it contains, their type and properties, as well as relationships among them. The SVE description may also include links to external information sources, such as additional entity descriptions, presentation models, and even related virtual environments. An agent can use the SVE description to quickly find task-relevant information. Since this happens on the level of application domain semantics rather than technical details, the agent can seamlessly work with any SVE belonging to the same domain.

Network Channel

[Agent does not understand raw protocol] The network channel of a multi-user environment disseminates interaction and change notifications to all participants, creating a shared dynamic world. When some event occurs in the world, a human user can easily recognize it by observing its effect on the rendering. However, an agent again will find it difficult to glean this information from the raw protocol messages. The agent might understand one particular protocol, but channels of other virtual environments will likely convey the same kind of events via entirely different protocols.

[Agent understands SVE events] An SVE event is a small RDF graph that describes an environment event explicitly - its type, properties, and reference to the world model. Most environments share a common event functionality for participation management, entity movement, and text chat, in addition to domain specific interaction. An SVE description can list the supported event types in its description of the environment's associated channel, along with the protocol and connection parameters. To access the SVE, an agent selects an appropriate channel adapter component, which can convert the raw messages of the protocol into SVE events and back. The agent then reacts to incoming events it understands, and emits its own events to interact with the environment. This again happens on the level of application domain semantics, so the agent can seamlessly work in any SVE of the same domain.

The SeVEn Framework

SeVEn is a framework for developing SVE agents from re-usable software components. It manages multiple SVE descriptions, related RDF schemas, and other information involved. SeVEn also dynamically configures event conversion chains, according to the needs of the agent logic and available environment channels. It is based on Java, the OSGi component framework, and the Jena RDF API.

More Information:
The SeVEn framework homepage.
Poster from the WWW 2005 conference (Japan), poster and short paper.
Presentation from the SVE'05 workshop (Switzerland), slides and paper.
Original presentation from the PhD Workshop 2003 (Berlin), slides and short paper.

Introduction

World Model

Network Channel

The SeVEn Framework