The AIAI Seminar Series in association with the SSP group of DAI
January 14th, Wednesday, 3pm-4pm
Department of Artificial Intelligence, 80 South Bridge, Room F13


 

Integrating Hierarchical Task Network and Model-Based Planning Technologies

Peter Jarvis
AIAI

Model-Based Planning (MBP) is characterised by the generation of plans from a central model of an organisation's products or services. The technology has developed independently from HTN planning through an effort concerned with providing implemented applications for industrial clients. Whilst both HTN and MBP technologies have achieved successful commercial application, their independence has prohibited the cross fertilisation techniques between them. This seminar resolves this independence by presenting an integrated architecture designed to exploit the relative strengths of each technology.

A KADS model set conceptualising the planning task in the Construction Industry is used to identify the complementary strengths and limitations.

The integrated architecture uses object-oriented constructs for representing domain-specific knowledge. A MBP applies knowledge encoded in the constructs to determine actions and ordering constraints. A MBP / HTN interface compiles the actions and ordering constraints into HTN schemata. A HTN planner then assembles the schemata into a complete and interaction free plan.

The architecture is currently being developed in two directions. First, it is being characterised as a problem-solving method suitable for storage in a library of such methods (e.g. TinA and Par-KAP). This work is characterising the types of knowledge that must be available in a domain for the architecture to be applicable. Second, in "action" oriented domains where plan synthesis is not underpinned by a product model, a product model may be of use underpinning a support tool for the domain writer. This direction has been proposed as a DAI MSc project.