| Abstract | We present a spatial model of group interaction in virtual environments. The model aims to provide flexible and natural support for managing conversations among large groups gathered in virtual space. However, it can also be used to control more general interactions among other kinds of objects inhabiting such spaces. The model defines the key abstractions of object aura, nimbus, focus and adapters to control mutual levels of awareness. Furthermore, these are defined in a sufficiently general way so as to apply to any CSCW system where a spatial metric can be identified - i.e. a way of measuring position and direction. Several examples are discussed, including virtual reality and text conferencing applications. Finally, the paper provides a more formal computational architecture for the spatial model by relating it to the object oriented modelling approach for distributed systems. cf. [Benford and Fahlén, 1993] Steve Benford and Lennart E. Fahlén. A spatial model of interaction in large virtual environments. In Proceedings of ECSCW ’93,September 1993. ftp://ftp.mrl.nott.ac.uk/pub/papers/ECSCW93.pdf. [Benford and Fahlén, 1994] Steve Benford and Lennart E. Fahlén. Viewpoints, actionpoints and spatial frames for collaborative user interfaces. In Pehrson and Skarbäck [1994], pages 225–231. [Benford and Greenhalgh, 1997] Steve Benford and Chris Greenhalgh. Introducing third party objects into the spatial model of interaction. In Hughes et al.[1997], pages 189–204. Benford and Fahlén, 1993, Benford et al., 1994a, Benford et al., 1995a] has turned out to be very fruitful for generating ideas for communication and interaction in shared virtual environments. The fundamental idea of the model is that every participant (avatar) is surrounded by three volumes, an aura, a nimbus and a focus. The nimbus of a participant can be seen as the projection of her presence into the environment; the focus is conversely her attention on the environment; the aura is a simple bounding volume that completely encloses the focus and nimbus, using it one can first do a simple collision test before doing potentially more complicated computations on the focus and nimbus volumes. The aura as such is not really necessary for the model to work, but is rather a way to improve the computational behaviour of the model. In my own work I have not used the aura as bounding volume, but rather have used it as a convenient collective term for focus and nimbus, at the risk of some confusion when reading papers by other users of SIM. An aura may be different for each communicative modality and does not necessarily have the same size or shape in the visual medium as it has in the aural medium, in the medium used for transmitting security credentials, etc; its size and extents can also be adjusted in run-time—possibly subject to system-enforced restrictions. When the auræ of the participants intersect, they can exchange information. When the focus of a participant intersects the nimbus of another, the former is considered to be aware of the latter. The awareness can be seen as a continuous function from 0 to 1, from no awareness to “full” awareness, determined from how large a proportion of the respective volumes intersect. It can also be seen that awareness is not necessarily symmetrical. The original SIM papers never defined awareness computation explicitly, but they could be interpreted to say that an aura would be a function f : R3 ! [0, 1], an awareness computation would then be some kind of integral of the product of focus and nimbus functions over the intersection of their respective definition volumes, the resulting awareness normalised to the interval [0, 1]. However, this definition has a number of problems, as it does not tell how awareness is affected if the volumes are of unequal sizes, should normalisation be done to the smallest possible volume, to the maximum possible value in the actual intersection or what? In any case such a computation is obviously infeasible to doin real time so I and other developers have rather used a simpler definition like awareness(a, b) = focus(a, pos(a, b)) · nimbus(b, pos(b, a)), where pos(o1, o2) computes the position of object o2 in relation to o1. That awareness can be computable is really the key concept of SIM. Once we have the awareness, we can use this to modify interaction between the actors1 in the environment. We can for example let the awareness value affect the audio quality of the audio stream from one user too another, or adjust the rendering level-of-detail, thus both simulating the lack of detail in our perception of items we do not attend to, as well as saving computer resources by not representing irrelevant items. The thing to note here is that while many graphics systems do support simplification of rendering, they will do this either based on a pure distance metric, or a metric based on the current computational load. An awareness computation within SIM however is context-dependent. The main drawback of SIM is that it is nontrivial to adjust one’s auræ, typical implementations allow switching between a small number of predefined volumes, but nothing near the rapid and effortless attention shifting we are capable of in real life [Coren et al., 1993] can be achieved. |
