An internet based, large scale virtual environment
Yoshiaki Araki
SONY Music Entertainment
April 7, 1996
Sony have been investigating a large scale, shared virtual environment
suitable for use in the Internet. The key focus of the project is an open
environment, suited to supporting large numbers of participants in a highly
interactive environment.
Our initial development, known as CyberPassage, is based on VRML1.0 but
extended to allow interactive behaviour, and multi-participant shared
worlds.
During the last several months we have participated in ongoing discussions
in the VRML community, contributing our experiences with CyberPassage to a
proposal for VRML2.0 co-authored with SGI and WorldMaker.
The current release of the CyberPassage system consists of a VRML1.0
browser extended to support behaviours and shared world. The sharing is
managed by a server process and relies on an IP based protocol which
communicates browser information and updates, via the server, to other
browsers sharing the world. Lastly, we provide an authoring tool,
CyberPassage Conductor, that allows uses to quickly add video, audio and
animation to VRML files using a drag and drop authoring metaphor.
The CyberPassage system has five distinguished conceptual aspects which we
feel are necessary for acheiving our goal of a highly interactive, usable
and scalable distributed VE.
1-1) High interaction rate support
To support high interaction rates with typical internet delays, shared
object behaviors are described by local scripts attached to objects.
Syncronization of shared object behavior are achieved using the client
server protocol (VSCP) which has been designed to minimize information
propogation and relies on 'dead reckoning' like techniques to reduce
communication.
1-2) Scalable network architecture
To support thousands of users in a shared world, the messages of shared
object behaviours are routed depending on the geometry of the 3D space. We
limit message traffic by defining a notion of bounded space, an Aura, which
describes the area of interest any object, including avatars have within
the space at any one time. We define visual, tactile and audiotory auras to
allow flexibility. We further reduce message traffic by defining differing
degrees of consistency requiring different message overheads.
1-3) Dynamically alterable world
To be practicable, any shared world must be dynamic. In addition to the
current model of dynacism via script execution, we also support the ability
to dynamically add objects by downloading from WWW servers, or a mechanism
to dynamically attach a network remote application via the server.
1-4) Configurable avatar
Users are represented by avatars in the virtual space. Avatars need to be
user selectable or designable. In addition we have experimented with
dynamic avatars for experession and feel that simple avatar expression
greatly augments the online experience.
1-5) Ownership
The issue of ownership is currently not addressed in existing VRML2.0
proposals. Implicit ownership in single user worlds will not scale when we
share worlds. As we begin to interconnect worlds it becomes critical to
deal with ownership to allow contol over change, and to support the ability
to pass ownership between users.
We are currently running shared servers in Japan and the US and have
regular visitors from among the 5000 or so people who have downloaded the
cyberpassage browser. Our current goal is to build a number of experimental
worlds in which we can experiment with interactive shared behaviours. There
are many areas we feel need further investigation, these include:
2-1) Message routing strategies for shared object behaviors
To archive distributed VEs, we need shared object behaviors among users
connecting in the same world. It is too hard to share all the information
with typical Internet delays. We need a set of smart strategies for
supporting consistency of the world. The message routing should be designed
to be configurable case by case, and be scalable to support thousands of
users in the same world.
2-2) Highly abstracted object library in distributed network architectures
For users to easily construct their own shared object behaviors and
interactions with object in VE, we need highly abstracted object library
like MOO database core. MOO is based on the old fashioned centeralized
network architecture. We should replace the rich concepts of MOO with a
distributed network architecture. The object constructions should be as
esay as HTML with authoring tools. The object descriptions should be
dynamically alterable as in MOO system.
D.K.S.