Abstract
The problem of office activity recognition is characterised by common occupant activities that may have composition relationships and involve expected objects with sensors providing both logical and numeric information. While this problem has been tackled using learning and ontological approaches, the associated disadvantages are obvious: a lot of training data is needed, the learned activities can be specific to the data, and ontologies suffer from flexibility and maintainability issues. We propose to address office activity recognition using Hierarchical Task Network (HTN) planning, a widely used Artificial Intelligence planning technique developed to model agents’ behaviour. We argue that HTN planning has the power to deal with all the characteristics of the office activity recognition problem while ensuring generality, flexibility, and maintainability. In addition, we suggest our approach be designed and implemented as a service not only to support the portability of off-the-shelf HTN planners but also to enable and encourage the integration of the approach in building automation systems. We present a use case to demonstrate the proposal’s feasibility.