Activity recognition for an agent-oriented personal health system

We present a knowledge representation framework that allows an agent situated in an environment to recognise complex activities, reason about their progress and take action to avoid or support their successful completion. Activities are understood as parameterised templates whose parameters consist of a unique name labelling the activity to be recognised, a set of participants co-involved in the carrying out of the activity and a goal revealing the desired outcome the participants seek to bring about. The novelty of the work is the identification of an activity lifecycle where activities are temporal fluents that can be started, interrupted, suspended, resumed, or completed over time. The framework also specifies activity goals and their associated lifecycle, as with activities, and shows how the state of such goals aids the recognition of significant transitions within and between activities. We implement the resulting recognition capability in the Event Calculus and we illustrate how an agent using this capability recognises activities in a personal health system monitoring diabetic patients.