Modular synthesis of intention analysis

Researcher: Christoph Burghardt

Runtime: 11/2006-10/2008

An ambient environment tries to assist its users in a pro-active way. For inferring the intentions of the user, the system has to have a notion about the possible and probable user behaviour. We have developed the Team Intention Tracker to assist teams during a meeting in a smart meeting room. This is a hierarchical model that aggregates sensor information to derive the intentions of user's and teams. The behaviour of the team is described as a Markov Model. The development of such a behaviour model is a difficult task because the number of states grows exponentially with the number of devices, persons, and history steps. To broaden the use of the intention analysis, we research new ways of adapting such a model to new scenarios, different team sizes and different device ensembles.

We employ planning techniques to automatically generate the possible team behaviour. The descriptions on how to operate a device comes from the devices themselves in form of an electronic manual. This enables us to build the plan dynamically, making the model independent from the number of persons involved, or the devices available.

Key publication

Christoph Burghardt, Martin Giersich, and Thomas Kirste. Synthesizing probabilistic models for team activities using partial order planning. In KI'2007 Workshop: Towards Ambient Intelligence: Methods for Cooperating Ensembles in Ubiquitous Environments (AIM-CU), September 10 2007.