The Mapping of Courses of Action Derived from Cognitive Work Analysis to Agent Behaviours

In a previous study[1], the feasibility of controlling entities in the Onesaf Test Bed simulation using a command and control hierarchy represented as teams of agents was demonstrated. The scenario used for the demonstration was the preparatory phase for a company attack on a platoon and it highlighted a number of issues requiring further research. The most important of these was the issue of behaviour capture. In the previous work, this knowledge was elicited from domain experts via a series of interviews. While this process proved satisfactory for a demonstration of feasibility, it was recognised that even with such a constrained scenario, a principled method is required in order to capture a complete and consistent set of behaviours. Cognitive Work Analysis (CWA)[2] was suggested as a candidate for such a method and a separate study was initiated to assess the usefulness of CWA for our scenario [3]. A particular attraction of CWA is that the output of an analysis can be specified in terms of behaviours and preconditions for their execution. While the actual execution model presupposed by CWA is different to that of our agents (which use the BeliefDesire-Intent (BDI) model) the level of behavioural specification is the same. Consequently, it was decided to use the results of a more extensive CWA analysis to provide our teams of agents with much richer and more realistic behaviour models. The purpose of this paper is to describe a mapping of Courses Of Action (COA) derived from CWA to agent behaviours.