AI Plan Representation in O-Plan2 _____________________________________________________________________________
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
This note explains a little about the model of activity being used in systems like O-Plan2, PLANIT, OPTIMUM-AIV, etc. and now being promoted as one basis for our work in process modelling. We are very keen to ensure our work fits closely with systems engineering work in train elsewhere - hence the comparison and linkage of terminology to IDEF. Its just as easy to link to other SADM type modelling (such as CORE or HOOD).
O-Plan2 Task Formalism (TF) Triangle Model of Activity ------------------------------------------------------
The O-Plan2 team at Edinburgh are actively revising O-Plan2 Task Formalism (TF) and in particular are trying to simplify some of the notions and to relate them better to existing software engineering and systems engineering requirements capture and modelling languages and methods (like IDEF, CORE, HOOD, etc). We have as a core an action decomposition (tasks descriptions, plans, partial plans, action schemas and other operators or primitive activities all being in exactly the same form as before in TF).
This is reflected in our "triangle" model of an activity. The vertical dimension reflects action (or plan or task) decomposition, the horizontal dimension reflects time. Inputs and Outputs are split into three principal categories (authority, teleology and resources). Arbitrarily complex modelling is possible in all dimensions. "Types" are used to further differentiate the inputs and outputs and and their semantics.
activity
^
|
|
|
/ \
/ \
authority / \ authority
pre-conditions --> / activity \ --> post-conditions
resources / decomposition \ resources
/ \
-----------------------
--- time --->
"Entry" to the model can be from any of the three points in the triangle model. From the top vertex to ask for action (or plan or task) decompositions, from the right to ask for actions (or plans or tasks) satisfying or providing the output requirement (a desired effect or "goal", a required resource, or a needed authority. These two sides are used mostly by our planners to date. The third side from the right can reflect triggering conditions for an action (or plan or task) and will be needed when improved independent processes are modelled as in our Excalibur system prototype.
The "intentions" or "rationale" behind the use of a particular activity in a process or plan can be related to the features of this triangle model. Normally causality or teleology via the pre-conditions/post-conditions has been used in AI planers for their plan rationale or Goal Structure (as we term it at Edinburgh). But in the richer model now in use in O-Plan2 for example, rationale in terms of resource usage or provision (e.g. this activity has been included in a plan only to provide a resource, its post-conditions may be a side efect for this PARTICULAR use). The same applies to authority provision.
Relationship to IDEF --------------------
Note that there is a direct mapping here to IDEF for example.
control
|
|
v
+---------------------+
| |
input ---> | decomposition | ---> output
| |
+---------------------+
^
|
|
mechanism
Where IDEF modellers usually use control for authority related triggers and mechanism to reflect resource availability. A criticism of IDEF if the lack of direct support for modelling the different types of output and their intended destination. Experience IDEF modellers use the arc labels, naming conventions and the "notes" system in an IDEF support "kit" to encode this information. The O-Plan2 TF triangle model more directly supports this and will allow better support tools. O-Plan2 TF typed conditions and typed resources are already employed to further differentiate inputs and outputs and how they may be legitimately manipulated in a plan. Further work is continuing on this and new work has now started on authority modelling along the same lines.