Collaborative building in Second Life

29 September 2008 | PlayOn authors archive

We recently completed our first study of collaborative building in Second Life. We asked groups of 2 or 3 people to work together on two building tasks and participate in a focus-group discussion. We observed people with all levels of expertise from novice to expert, in order both to capture cutting-edge practice as well as non-expert reactions to the user-interface. Some of our participants had not used SL before but were experienced with other virtual worlds or CAD software.

Each group carried out two building tasks. The first was to assemble a “house” out of parts we gave to them. A completed house is illustrated in the screen-shot above. This was effectively a 3d jigsaw puzzle. Groups used voice for communication, and could organize themselves any way they wanted. For the second task, one person was made group leader. They couldn’t use the editing tools, but instead had to direct their team-mates to build an outdoor scene of garden furniture arranged around the house. Only the leader saw the photo of how the garden should look. This arrangement forced groups to reference objects and places a lot, either verbally or by whatever means they could devise. We recorded the groups’ screen video and speech so we could examine their references to objects and their use of “alt-zoom”, Second Life’s free-moving camera. Read on to see what we found.

Prior research suggests problems for collaboration in 3d

Prior research from older Virtual Reality environments indicated that achieving common reference to objects was a major problem for collaborating groups: for example, pointing, and spatial deixis like “that thing over there”, is less likely to work in a virtual environment than it does in real life. To use deixis requires that peopleto know where their collaborator is looking. Small screen width makes this hard in any “desktop” (PC-based) virtual environment. In Second Life it is potentially impossible, because users can move their view-cameras independently of their avatar positions. In SL, you can see where someone’s avatar is looking, but you can’t see where they have placed their camera. We were interested in whether this was a problem for 3d collaborators and how they worked around it.

The out-of-avatar experience

First let’s look at how people used the “moving camera” feature. At any moment in SL, a user is either looking through their avatar’s eyes (usually from just behind their avatar’s head) or through the disembodied camera. We called these modes “in-avatar” and “in-camera”. (The latter term reminds us that camera locations are not visible to other users.) On average, users spent 43% of their time in-avatar and 57% in-camera. But this depends on expertise. We divided our users into groups based on experience and found that experts spent 90% of their time in-camera, while novices stayed within their avatars most of the time. This is illustrated in the graph on the right. Camera use didn’t differ significantly between the two tasks.

We wondered whether the movable camera changed SL users’ experience of the world. In most games and virtual worlds one’s camera is fixed to one’s avatar. When an SL user moves their camera away from their avatar, does it feel strange, like an out-of-body experience? Does it lessen the importance of one’s avatar as a place-holder in the world, or of adhering to norms of interpersonal distance, as Nick Yee and colleagues have observed in Second Life users?

We found that, by and large, SL users still identify with their avatar, even when their focus of attention is elsewhere. When asked “where are you now”, they usually nominated their avatar location, not their camera location. SL users are mindful that avatar locations are public and camera locations private, and will often enact real-world proxemics with their avatar (person-to-person distance, and mutual gaze) even while they move their camera elsewhere. However this does depend on activity: some experts reported that while building they considered their avatar to be irrelevant, or that it even got in the way. At times it seems as though Second Life is a combination of two user-interfaces: “Blender grafted onto a game” as one user put it.

Referring to places and things

Does Second Life’s free-moving camera mean that spatial deixis – utterances like “the wall to your left” – can’t work? Interestingly, users, including experts, used spatial deixis relative to a collaborator’s avatar position, even when they were aware that their collaborator had probably moved their camera and wasn’t looking from there. Furthermore, these references were usually interpreted correctly: users were able to translate them to their avatar’s frame of reference, by looking back to their avatar and/or by mental rotation. When this failed they sometimes interpreted deixis by returning their camera to their avatar. This requires only one keystroke in SL, however the camera position at which they were working is lost, and can take time to find again. This could be addressed by adding a “toggle” to switch between the two perspectives.

The graph on the right illustrates how often people used different styles of verbal reference to objects. As might be expected in an environment where it is not clear where a collaborator is looking, spatial deixis was not the most popular form of reference. More often, people described objects by naming or describing them, such as “the large brown rectangle”. But deixis was used regularly as well. Less popular was deixis with respect to the intrinsic frame of reference of another object: for example “in front of the house”. Extrinsic reference – using compass points such as “to the north” or distant landmarks such as “seaward” – was rarely used, although a map with compass is available in SL.

Second Life allows a few referencing tricks that would be impossible in the physical world. For example, if you place an object in edit-mode, your avatar points to it, and a line of dots connects you to the object. However this only works at a coarse-grained level, not for complex collections of objects, and our study participants rarely used pointing to refer to location. Another trick is to temporarily create a marker object at a point, to be deleted when no longer needed: this is like using pegs to mark out a patch of ground. A few users “wiggled” an object back and forth to indicate to a collaborator what they were referring to: “the one I’m moving now”. And on a number of occasions users marked an object simply by walking their avatar to it and referring to it as “beside me”.

In conclusion, although previous studies of 3d collaboration in virtual environments indicated that users had a great deal of trouble knowing to which objects each other were referring, this didn’t seem to bother our study participants too much, and most groups got their tasks done in the end. Where problems arose they were usually because of novice users difficulty with the build interface, especially having to move objects along Cartesian axes.

Division of virtual labour

We were struck by how many groups used the strategy of decomposing their build into sub-assemblies and assigning them to group members to complete individually. This strategy allowed people to work in parallel with less need for communication. Frequently one person assembled the roof while another put the walls in place. This strategy obviated the need to collaborate closely, as though people understood the difficulty of close collaboration in 3d and were avoiding it. Some of the groups did attempt to collaborate more closely, for example grabbing one wall each and moving them into place together. These groups talked more and probably had more fun, but got their house built more slowly. Our “garden” task was in effect a strategy by the researchers to force more collaboration and communication. The fact that we needed to force close collaboration may be telling. Expert users confirmed that decomposition was standard practice in Second Life, where large projects are usually divided into sub-assemblies that are completed asynchronously and joined at the final stage of the project. Alternatively tasks are divided by specialization: for example, one person might create a building’s “skin”, while another does furniture, another roads, and another works on textures.

When people build things together in real life, they frequently have to collaborate synchronously around individual objects: for example one person might hold two pieces of wood in place while another screws them together. In previous research, users were sometimes forced to collaborate in this way, by implementing gravity and friction into the virtual environment. But gravity and friction are usually not necessary in virtual worlds. Experienced Second Life builders make use of the fact that objects will stay hanging in the air, or even pass through other objects. Rather than force real-world modes of collaboration, it might be better to embrace the non-physicality of virtual worlds, allowing people to avoid the kind of synchronous 3d collaboration that the physical world forces upon them. Rather than tweak the environment to force real-world styles of collaboration, more might be gained by improving aspects of collaboration that users find troublesome, such as the permission system which currently prevents objects which have been made by different individuals from being linked together.

Comments (3)