Situated Interaction on Spatial Topics

In his PhD thesis (Situated Interaction on Spatial Topics. Christian Kray (2003). PhD. thesis, DISKI series vol. 274, AKA Verlag, Berlin), Christian Kray presents a model and an implementation to handle situational interactions on spatial topics as well as several adaptation strategies to cope with common problems in real-world applications.

The interaction on spatial topics is highly important no only in the context of mobile and situated systems but also in other fields such as natural language access to maps or user interfaces. Therefore, one goal of this thesis is to develop a generic model for situated interaction on spatial topics that can be used to build real world applications. [...] In addition, we analyze typical problems that arise in the context of mobile real world applications, and point out strategies for coping with them. [...] The implementation realizing the model corresponds very closely to the structure of the model, and was put to test in a scenario of a mobile tourist guide. [...] From a user’s perspective, a mobile assistance system is not very useful, if it fails when faced with less than expected information quality.

It is an interesting work for my thesis because its seeks adaptive strategies for real world applications for spatial interaction on spatial topics:

Since the world that we live in is not a perfect one, it is quite possible that information needed in the reasoning process is simply not available, or at a lower precision than what the model requires. Consequently, building a truly helpful system for real world use means to take these issues into account. Therefore, we designed several adaptation strategies for common problems in spatial interaction on spatial topics. [...]

A frequent problem arising in the context of human-computer interaction is the unreliability of information
sources: often, sensors will not return sufficiently precise information or no information at all, network connections will fail disabling access to remote databases, and some information may be immeasurable and can only be derived over time (e. g. the user’s interests).

In addition to resource restrictions on the cognitive level (e.g. driving a car), we have seen that technical resources may also be restricted. On an abstract level, the lack of lack of information ca be addressed in several general ways:

• ignoring missing information
• accessing alternative sources
• using default values
• inferring missing information
• adapting computation
• requesting information from the user

A dead recknoning algorithm can be used in case of inferring missing positioning information, The result is a set of potential positions under the assumption that the user did not change his speed and direction.

Determination of the user’s current position can be a mix of measurements, inferences and user inputs:

Chris Kray also addresses two topics of interests: self positioning and positional information visualization:

Self positioning

In case of self localization task, the user’s goal is to learn about her current location in such a way that she is able to position herself within her current model of the world. Frequently, this is achieved by means of a you-are-here map [Richter, 2001], on which not only the user’s current position is marked (by an arrow and/or a cross) but also familiar landmarks.

Position visualization

When the precision of the positional information decreases, one way to compensate is to use other graphical means to mark the user’s position such as circles that grow with the imprecision instead of crosses (see, for example, LoL@ and REAL in 3.6 and 3.7).

The choice of implementing a real world mobile tourist guide is explained as follow:

The way we selected for our model was to build a system that realizes the theories underlying our model. This has the advantage of proving the practical relevance of our approach while allowing for a later empirical evaluation of selected parts of the system in a real-world setting.

A further reason why we favored an implementation over a purely empirical or mathematical validation was the breadth of the model. Instead of just analyzing a small subset of situated interaction on spatial topics, our goal was to design a model that covers most tasks in this realm.