Although it is generally desirable that users in a collaborative virtual environment should perceive it in the same way we argue that in some situations it may be useful to allow users' views of the environment to diverge, allowing each user to tailor their view to one that best suits their needs whilst still allowing some form of collaboration. We refer to such environments as subjective environments. In this paper we describe an initial prototype of a subjective visualization system and the techniques used in its implementation.
In the field of abstract information visualization it is possible to imagine several techniques for visualization of the same data set. Since the data set has no intrinsic ``natural'' representation (unlike a 3D CAD model for example) the choice of a particular technique will depend on the task at hand or on the preferences of the user. It is therefore natural for visualization systems to present the user with a number of choices which govern the nature of the visualization presented to them. Since people need to work together, there is some interest in the field of Populated Information Terrains (PITs) [Benford'95a] in which both users and information are embodied in a Collaborative Virtual Environment (CVE).
However, using most current systems all users are forced to use the same representation of the information and thereby trade flexibility for the ability to collaborate in the use of the information. In this paper we describe a prototype environment that both allows multiple users to collaborate and communicate in a shared space and allows the environment to contain viewer dependent features. We refer to this as a subjective [Snowdon'95] environment.
Section 2 will introduce the concept of Populated Information Terrains and in section 3 we will provide a brief justification for extending the PITs concept to include subjectivity. Sections 4 & 5 will describe how we implemented a subjective PIT. Finally, in section 6, we shall conclude with a brief discussion of the trade-offs and potential problems with our current implementation.
The concept of Populated Information Terrains [Benford'95a] combines ideas from the fields of Computer Supported Cooperative Work (CSCW), virtual reality and databases to create multi-user virtual environments which support visualization of, and cooperative work within, shared data. The underlying philosophy of PITs is that they should support people in working together within data as opposed to merely with data. In other words users are explicitly embodied in the virtual environment and not relegated to the status of external agents whose presence is merely implied as a side effect of their actions. There are several reasons why we believe PITs are useful, these include:
Possible applications include browsing large databases, browsing library catalogues, navigating through large hypertext systems such as the World Wide Web or supporting teams of programmers developing large pieces of software. Section 2.1 will describe VR-VIBE an existing PIT developed using the SICS DIVE [Hagsand'96, Carlsson'93] multi-user virtual environment.
VR-VIBE [Benford'95b] is a multi-user 3D visualization of a collection of documents or document references. The visualization is structured using a 3D spatial framework of keywords, called Points of Interest or POIs. The spatial position of an document icon indicates the relative attraction of a document to the different POIs, where attraction is estimated in terms of thematic similarity. Thus, an icon equally spaced between two POIs is equally relevant to both, while an icon close to a particular POI is relevant to that POI only.
Absolute relevance is depicted by the size and brightness of the document's representation; the more relevant the document the bigger the icon and the brighter the colour. As spatial location only determines relatedness to each POI in the current query, relative icon size and brightness thus differentiates documents which are slightly relevant to all POIs from documents which are highly relevant to all POIs.
To determine these relative and overall measures, VR-VIBE employs a simple text matching algorithm (i.e. counting the number of occurrences of keywords in titles, abstracts and the body of the document) to compute a match between each document in the store and each POI. This computation results in a normalised score which is translated into the appropriate visual representation.
Figure 1 illustrates VR-VIBE at work. Here 5 POIs are specified, represented by green octahedrons. A white sphere above an octahedron indicates the POI is currently 'active'. Blue blocks represent documents. Also visible in the picture are three other users, all with differing styles of embodiment ranging from the simple green T-shaped "blocky" on the right of the 3D view to the more complex "cartoon" embodiment on the left. Users can navigate within the 3-D space, select individual documents, control the display according to the dynamic relevance threshold and "drag" POIs to new locations. Dragging alters the shape of the document icon space; icon location changes can be used to gain further information about relative attraction to different POIs. In addition, new searches can be specified by creation of new POIs and/or by specifying keywords. Selecting a document icon causes some summary information to be displayed. If a document is available via the World Wide Web [Berners-Lee'92] then VR-VIBE can invoke a web browser to display the entire document contents
Figure 1: The world of documents in VR-VIBE
We chose VR-VIBE as our test application because we had access to the source code and therefore could adapt it for subjective operation and because it already supports a number of visualization styles allowing the creation of subjective views that may differ substantially
Most current multi-user virtual reality systems provide a highly objective virtual environment. That is, all users see the environment in the same way, albeit from different viewpoints and all users see same objects in the same places with the same appearances. However, experiences with the strictly objective WYSIWIS (What You See Is What I see) paradigm for 2D interfaces suggest that collaborative applications in general require some degree of subjectivity, leading to variants such as ``Relaxed WYSIWIS'' [Stefik'87]
.
Figure 2: Alternative views of the same dataset shown in figure 1.
Visualizations of abstract information, such as those provided by VR-VIBE, can have a great deal of flexibility in the choice of visualization style since the source data has no intrinsic appearance. This is illustrated by Figure 2, which shows alternative VR-VIBE visualizations of the same dataset as that shown in Figure 1. Given this freedom of choice it is likely that users will form their own preferences for the display of particular datasets. If the virtual environment does not support subjective views then the users are forced to agree on a common (possibly non-optimal) visualization style. However, if the virtual environment is capable of supporting subjective views then users are free to choose their own preferred visualization style. We therefore hypothesize that the PITs concept would benefit if it were extended to allow subjective views of the data and other users. In order to achieve this we need to be able to do the following:
The next sections will show how we attempt to solve these problems.
In order to support subjective visualizations VR-VIBE needs to be capable of generating a different visualization for each user. It does this by explicitly separating objective and subjective state information. The objective state contains the content of the document store and other information that is not dependent on the nature of the generated visualization. There is exactly one copy of the objective state information. The subjective state contains the objects used to create a specific visualization and any other parameters that are dependent on the nature of the visualization (configuration options etc.). There is an instance of the subjective state for each subjective view generated by VR-VIBE.
The latest versions of DIVE allow a virtual world to be hierarchically partitioned between several multicast groups. Associating a multicast group with each user provides an efficient mechanism for implementing subjective (viewer-dependent) views of parts of a DIVE world as each user-client is only sent the updates relating to the top-level (objective) portion of the world and the portion contained in their own multicast group.
VR-VIBE constructs a subjective environment in the following manner: The main DIVE world (which also has a multicast group associated with it) contains only invisible objects containing objective state information. Each user is required to use an embodiment with a subjectivity flag set, which upon entry in the world causes the creation of a multicast subgroup unique to that user. When VR-VIBE detects that a new user has entered, it creates copies of each object and places them in the new group. They, as well as the userís embodiment are thus invisible for everyone except the user in question. The next section will explain how users retain the ability to communicate.
Since the artifacts in one user's view may be in very different locations in another user's view we cannot place user embodiments at the same world coordinates in each view. We must therefore consider some other mechanism for positioning user embodiments in subjective views. We observe that an empty space conveys no useful information; it is the artifacts, other users in the space and their actions in relation to one another that provides us with information. We therefore consider the position and orientation of users in terms of the artifacts they are accessing rather than in terms of their location in world coordinates. We do this using a technique, which we term artifact-centred coordinates, which uses the artifacts the user is aware of to determine the position and orientation of that user in other subjective views.
The basic concept behind artifact-centred coordinates is to compute a user's awareness of a set of artifacts, find which of these artifacts exist in the subjective view we want to represent the user in and place the representation of the user in a position and orientation determined by the location of the artifacts that the user is aware of in the target view.
A translator process is located in the same main world as VR-VIBE, and thus sees the same subjective views being created and subscribes to these. For each new user that enters it will create a "pseudobody" for this user in every other private view; this is necessary since, as we noted above, firstly the user's "true" body is invisible to all others, and secondly, the position of that body is meaningless in the reorganised space of another user.
To position the pseudobodies, the translator finds the set of artifacts that the user is considered to be aware of by computing a "volume of interest'' around the user which encompasses all the artifacts that the user is currently capable of interacting with. (We assume that the user has some degree of awareness of all artifacts which intersect their volume of interest.) An awareness function is then applied to these artifacts to determine the degree of awareness of each artifact. This awareness function could simply use the distance between the user and artifact and how close the artifact is to the user's line of sight, alternatively, an arbitrarily complex function could be used which took into account e.g. the number of accesses of the artifact and the type of the artifact itself.
The translator will then for all other views find the objects that correspond to the set of objects the user is aware of in her private view, and then adjust the position of the pseudobodies in those views, according to some inverse of the awareness function, such that the pseudobodies in all other views end up close to the same objects the "true" embodiment was close to in the user's own view. It may be that that the objects the user is aware of do not exist in a particular subjective view, in which case we may decide either to move the pseudobody to a neutral position or not to move it at all.
We are still experimenting with awareness functions and their inverses, but our experience up to this point indicates that the awareness function should have a cutoff such that less than ten objects are likely to be within it, since this will minimise spurious wobbling of the pseudobodies when the user moves. Whether spurious motion indeed is detrimental will have to be determined by further user studies..
It seems obvious that if usersí subjective views diverge beyond a certain point then meaningful collaboration will become difficult or impossible; at the same time we theorize that if users are forced to share the same objective view then collaboration may occasionally become awkward if users have to continually negotiate an agreed common view ó this was found to be the case for strictly WYSIWIS 2D shared editors and lead to the development of the Relaxed WYSIWIS concept. This has lead us to believe that a subjective environment that allows users to adjust their subjective views to suit individual needs, but still retain awareness of others and the ability to communicate, may be a useful tool.
We would imagine that over the course of a session users' views may converge and diverge depending on the situation. We will therefore need to develop techniques that allow user to switch to a common view or to find information on how divergent their view has become from other users. It may also be useful for the system to monitor the parameters for each user's subjective view and allow users to switch to the currently most popular view.
It is worth noting that our current implementation allows objective applications to run in the same world without modification. It an objective application belongs to the world's multicast group then it will be available to all users in the world. Alternatively, objective applications might belong to a given user's multicast group and thereby be completely private to that user. This would allow users to still use tools such as the DIVE whiteboard whilst also using a subjective visualization.
In this paper we have argued that in some situations providing subjective or viewer-dependent features in an environment my actually aid collaborative information visualization by allowing each user to tailor their view while still maintaining contact with other users. We have described a prototype information visualization based on this ideas. Since our implementation is based on a collection of applications it would be relatively easy to replace some components to provide a different of features (e.g. to visualize other sorts of abstract information).
Many thanks to Lennart Fahlén and the Swedish Institute of Computer Science (SICS) for funding Dave Snowdon as a guest researcher at SICS for the period of this work. We also thank Olof Hagsand, Mårten Stenius and Emmanuel Frécon for DIVE help and John Bowers for enlightening discussions.
[Benford'95a] Benford, S., Snowdon, D. and Mariani, J. (1995), "Populated Information Terrains: First Steps", in Virtual Reality Applications, Academic Press Ltd., Earnshaw, R.A., Vince, J.A. & Jones, H. (eds)., pgs 27-39.
[Benford'95b] Benford, S., Snowdon, D., Greenhalgh, C., Inrgam, R., Knox, I., and Brown, C. (1995), "VR-VIBE: A Virtual Environment for Co-operative Information Retrieval", Computer Graphics Forum 14(3) (proceedings of Eurographics'95), 30th August - 1st September, NCC Blackwell, pp 349-360.
[Berners-Lee'92] Tim Berners-Lee, Robert Cailliau, Jean-Francois Groff and Bernard Pollerman (1992), World -wide web: The Information universe, in Electronic Networking: Research, Applications and Policy, 1(2).
[Carlsson'93] Christer Carlsson and Olof Hagsand, "DIVE - a Platform for Multi-user Virtual Environments", Computers and Graphics, 17(6), 1993, pgs 663-669.
[Hagsand'96] Olof Hagsand, "Interactive Multiuser VEs in the DIVE System", IEEE Multimedia, 3(1), 1996, pgs 30-39.
[Snowdon'95] Snowdon, Dave, Greenhalgh, Chris, Benford, Steve (1995), "What You See Is Not What I See: Subjectivity in Virtual Environments", Framework for Immersive Virtual Environments (FIVE'95), QMW, London, UK.
[Snowdon'96] Dave Snowdon and Kai-Mikael Jää-Aro (1996) "Body-Centred Configuration in Collaborative Virtual Environments", FIVE'96 Framework for Immersive Virtual Environments II, December 19-20, Pisa, Italy.
Mark Stefik, Daniel G. Bobrow, Gregg Foster, Stan
Lanning and Deborah Tatar (1987), "WYSIWIS Revised: Early
Experiences with Multi-User Interfaces, ACM Transactions on Office
Information Systems 5(2), pgs 147-167.