Manipulation Conflicts

As humans, we do not only meet to talk and gesture, but also to explore the world together and adapt it to our needs. This is no different in virtual environments fro collaboration. Joint object manipulation has received particular interest in re- search on collaborative user interfaces – often due to observed coordination issues. If several participants are provided with equal means to manipulate objects in a shared environment, their actions may cause mutual interference rather than sup- port (e.g. [235, 262, 282]). Such conflicts seem to be particularly pronounced if the input and output spaces (person space and task space) are not overlapping and par- ticipants may intrude each other’s personal territories [149, 156, 268] (Figure 5.4). In real-world settings, similar situations occur – and in case of many games this strug- gle for control over a particular item is often the whole point of the action (consider ball games as an example). However, we can generally build on experiences with similar objects, in similar setting, and under the same physical laws, which provide us with an implicit understanding of each other’s action capabilities. We can antici- pate the goals and actions of others, since objects must be in reach for manipulation and we can observe movements towards the areas of interest (see Section 4.4).

Figure 5.4: Research on multi-user collaboration in virtual environments often

emphasized the issue of manipulation conflicts and suggested techniques to support the negotiation of access. In real-world settings people successfully co- ordinate their actions based on mutual observation and synchronization. The spatial dissociation of user input and its effects, as it often occurs in computer applications, hinders the necessary awareness. If the person space and the task space cannot be perceived as a coherent whole, people may unconsciously vio- late each other’s interaction territories.

Some of these cues can also be exploited in collaborative computer applications, if the person space and the task space are consistent, i.e., if the spatial relations of user actions and their effects are directly related and observable. With multitouch dis- plays, for example, input can be directly applied to the visible application content, while movements of a mouse pointer are spatially separated from the operated input device. Hornecker et al. compared these two paradigms for multi-user interaction on a tabletop display [149]. They observed more concurrent interactions in the touch condition which allowed for mutual support but also increased the frequency of con- flicting actions. With multiple mice, instead, users tended to act one after another, in order to permit the monitoring of each other’s actions and avoid potential inter- ferences. The higher mutual awareness in the touch condition facilitated negotiation and conflict solving, which allowed them to take more risk and engage simultane- ously in the cooperative task.

Collaboration often requires to anticipate the actions of others. In real-world set- tings, we can build on our experience of possible actions under invariant physical constraints. This expectation, however, does not hold in the context of computer in- terfaces. Objects can be manipulated from any distance and any type of user action can be mapped to the various effects. The manipulation of virtual objects is not con-

Multi-User Cooperation 73 strained by physical laws and the consequences of concurrent user actions need to be explicitly considered. Furthermore, distributed systems for remote collaboration suffer from synchronization issues: Users at different locations may want to affect the same attributes of an object simultaneously without being aware of their concurrent actions. Greenberg and Marwood identified this concurrency control problem in dis- tributed groupware and showed that neither locking, nor merging, or forced serial- ization of user input can solve all of the resulting problems [111]. They discussed the tradeoffs related to various implementations of these concurrency control schemes and proposed to apply user-centered design methods to find the best compromise for each specific use case.

In collocated groupware, typically system consistency is maintained by managing the application state through a single application and the physical presence of in- volved users improves their mutual awareness. However, Morris et al. found that social protocols are not always sufficient to prevent conflicts [235]. Unexpected side effects of interaction methods are a part of the problem. Ease of use can become an- other issue. Consider common actions like maximizing a GUI element or panning the workspace. Direct access to global transformations is clearly beneficial for single user workplaces, but in the context of collaborative applications, one must keep in mind that the results affect others too. Gutwin and Greenberg found that the usability re- quirements for individual users and groups can be conflicting [120]. In the context of remote groupware developments, they suggested to enable individual views of the shared interaction space and explicit indicators for mutual awareness to alleviate these issues.

Morris et al. also observed conflicts in collocated settings and discussed the design space for application-controlled coordination policies both for global changes and in- dividual object transformations [235]. They suggested a distinction of policies based on the source of the coordination initiative. Proactive coordination relies completely on the decisions of the user who initiated the conflicting action. Reactive policies consider only the situation of everybody else and mixed-initiative coordination tech- niques combine aspects of both policies. Their systematic analysis of the topic pro- vides a good starting point for specific groupware developments. Following Green- berg and Marwood’s considerations on distributed groupware [111], they argued that there is no ideal solution, but the suitability of various approaches depends on the users and application scenarios.