Artform at the University of Huddersfield
Research into Planning and Scheduling in Artform of the University of Huddersfield has steadily increased over the last five years,
with projects supported
by the UK's research council and the National Air Traffic Services.
The projects involve both fundamental and applied research, and
encompass symbolic and subsymbolic AI, statistical and OR techniques.
Planform: An Open Environment for Building Planners
The Planform project is a major EPSRC-funded project which started in
December 1999 and which involves researchers
from the Universities of Huddersfield, Durham and Salford. The project is also
supported by the UK National Air Traffic Services Ltd and CogSys Ltd.
The Planform project aims to research, develop and evaluate a method and
supporting high level research platform for the systematic construction
of planner domain models and abstract specifications of planning algorithms,
and their automated synthesis into sound, efficient programs that generate
and execute plans.
Object-Centred Language: Specification of planning domain models
The Object-Centred Language (OCL) and more recently the hierarchical version OCLh development method forms a rigorous approach to capture
the functional requirements of classical planning domains.
Using OCL, one builds a model of the planning world, including valid
world states and operator schema modelling possible actions.
The operators' preconditions and postconditions are specified in terms of
the substate transitions of objects. This approach utilises the notion of
``lifting'' a world's representations from the level of the literal to the
level of a more abstract object-centred representation.
Object-centred Planning with TO and PO Planners
An investigation has been carried out into the use of domain models
written in OCL with three different object-centred planners.
The algorithm for an object-centred planner is consistent with the OCL
domain model in that it functions at the object level rather than the
predicate level. The first planner is an object-centred total-order
planner, based on McCluskey's FM. The planner calculates for each
object the difference, if any, between its current state and its goal
state. It then seeks to reduce this difference by finding steps that
achieve the object's goal state. If these steps are applicable in the
current planning state, then they are immediately applied, and planning
continues from this advanced state. The other 2 planners are both
object-centred partial-order planners. One is based on UCPOP and it
plans by finding actions that achieve a required transformation from
one object substate to another. These actions are partially ordered
with respect to each other. Planning stops once a state satisfying the
goal is reached. The second partial order planner focuses on one
object at a time, building a path of partially ordered actions from its
initial state to its goal state. Once a path has been constructed it is
the merged into the global plan. Both the total_order planner and the
partial-order path planner can make use of top-level goal orderings to
help improve their performance.
Object-Centred HTN Planning
The objective of the research is to extend the structures in OCL so
that it is expressive enough to be used in realistic applications with
HTN planner technology. An HTN planner called "EMS" has been
created that processes hierarchically defined domains in OCLh.
"EMS" can input problems posed as
combinations of goals, action schema and constraints,
and it outputs valid plans that solves these problems.
Planning Using an Object-Plan Graph
This ongoing work integrates two strands of
planning research - that of using plan
graphs to speed up planning, and that of using object representations
to better represent planning domain models.
To this end we have designed and implemented "OCL-graph", a
plan generator which builds and searches an object-centred plan graph,
extended to deal with conditional effects.
Our initial design and experimental results appear
to confirm our
conjectures that the extra information and structure
of OCL benefits plan generation efficiency and algorithmic clarity.
Knowledge-intensive conflict resolution planning
The research was sponsored by the UK NATS, and focused on the domain problem of conflict resolution
for the 4D multi-segment flight-plan clearances which figure in current
methods of air traffic control over the Atlantic. Development of a prototype
problem solver for conflict resolution of flight plans involved integration
of AI and OR techniques.
An investigation into ways of achieving more flexible and
adaptable, artificial intelligence based, nurse rostering systems
This research is centred round the use of multi-modal reasoning to
improve workforce scheduling, particularly in the nursing domain. The two
key paradigms being investigated are
Case-Based Reasoning (CBR) and Constraint Logic Programming (CLP).
The aim is to find ways of storing previous schedule solutions as cases held
as partially instantiated and generalised CLP descriptions. Similarity
measures will be used to extract all or
parts of these descriptions for use in the generation of new schedules.
The reason for this choice of multi-modal reasoning is to increase the
possibilities for adaptation and learning within the system. Of particular
interest is that ability of case-based reasoning to
reflect expertise or preferences that are not explicitly expressed.
Here are some recent publications.
Lee McCluskey - firstname.lastname@example.org
Ron Simpson - email@example.com