My Nonlin planner and the NOAH planer from SRI (both about 1975) were able to generate plans from hieracrchical descriptions of actions and plan components (processes). The output was a hierarchical PERT chart or petri-net representation of the plan. Nonlin specifically used a representation of the underlying reasons for the specifically chosen actions in a plan. Later version of Nonlin with a plan revision capability (based on a decision graph similar to early versions of facilities now available in reason/truth maintenance systems).
Over the years more flexible plan representations have been developed. We worked on a "club" project as part of the UK ALvey Programme called PLANIT with around 25 UK organisations to develop an AI plan representation based tool that could integrate information about process planning (not in the business process sense - I mean here the manufacturing process for components), project planning and job shop scheduling. The user of the KEE based prototype tool could perform constrained editing of the rationale based linked structures in an enterprise (the example being a company building fuel tanker trucks) to adjust the processes, plans and schedules as circumstances or requirements changed or as an alternative manufacturing process was being considered. It was early days (1984-5) but is similar to what we are trying to do with more sophisticated representations now.
We have used this background to input to recent work to define a shared ontology for plan entities which is being published via the DARPA/Rome Lab Planning Initiative as KRSL.
Our most recent planner (O-Plan) takes this some way forward but is still based on underlying goal structures to capture the rationale for activities chosen for a specific process or plan. As well as applications in mission control, manufacturing, logistics, etc, we are now using the same concepts in business process modelling. Our approach gives the capability to re-engineer processes after modelling and analysis points out areas for improvement. This can then be done by constrained editing out parts of the current process (replacing these with "flaws" based on the rationale structure), and then using constrained editing or constraint propogation and search based synthesis planning methods to propose or make improvements to the process.
Hope these notes serve to fill you all in on our interests. Austin Tate
Prof. Austin Tate AI Applications Institute University of Edinburgh 80 South Bridge Edinburgh EH1 1HN United Kingdom
email A.Tate@ed.ac.uk
tel UK +31 650 2732 fax UK +31 650 6513