Modelling Human Behaviour Using Partial Order Planning Based on Atomic Action Templates

Abstract

A problem in assisting users in intelligent environments is the detection of their long term intentions based on the current state. One approach to solving this problem is the employment of human behaviour models, such as CTML and PDDL. At present, most of these models are based on concrete domains or scenarios which makes them difficult or impossible to adapt for other use cases. To overcome this drawback, this paper introduces an approach that uses partial order planning for generating a user behaviour model. Furthermore, it proposes a generalization of human activities by introducing atomic action templates valid for activities from various domains. To illustrate the approach, a scenario from the elderly care domain is modelled. Overall, the paper provides an effective general approach for modelling human behaviour which can be embedded in the intention inference workflow.