Tip:
Highlight text to annotate it
X
We'll take a look at the very wide range of planning research areas and
techniques. There's been quite a lot of work in
domain modeling, domain description, and domain analysis, prior to using these
descriptions, these domain descriptions in planners themselves.
Then of course the core work in AI planning is being in the search methods,
and algorithms for creating plans, searching for plans.
People have looked up plan analysis because if you can constrain the types of
plan you're interested in through analysis, that can limit the search
spaces involved. It could be useful to pause here and to
read through this list of areas and techniques.
How many have you already heard of? Check out or ask about any acronym you
don't know. Continue on when you're ready.
Then people concerned with realistic and practical planning problems have often
had to deal with the hue of repairing plans when things go wrong or when the
circumstances change. A number of people have looked at planned
generalization, the reuse of generalized plan fragments.
The sport of the user involved in planning via a suitable user interface
is, advice interfaces and the ability to generate plans in a mixed issued fashion
between the automated system and the human are important.
In recent years, planners have been made available as web services in order that
they can be used as a component of larger scale systems.
And plans can be used in a range of other areas, such as natural language
generation, dialogue management, the sharing of plans between humans and so
on. Pause again, to look at the research
areas and techniques listed. So as you can see here, we're dealing
with the whole life-cycle plans, right through from, creating domain models,
to making use of those plans in productive situations.
The problem is that we need to make sense of all of this.
We need to find a way of fitting in the very many techniques that are available
now and that will become available through further research and development.
What's needed is more collaborative planning framework.
At the alta level, we've got to be able to relate to the humans involved in the
planning process, where they can present their objectives,
their issues, they can make sense of the situation.
They can give multiple options and advice.
They can argue about those options, discuss them, outline plans, and so on.
Then we want detailed planners, search engines, constraints, all that analyzers
and simulators, that can act within that outdoor frame, and work in an
understandable way, and use it to provide feasibility checks,
detailed constraints, and guidance. Want to be able to share processes and
information about process products between humans and systems,
and want to look at the current status, the context we're in, the environment
we're in, you know, to be sensitive to that.
We need a link between informal and unstructured planning and more structured
planning and methods for optimization. And that's what we've been trying to do
with our I X I plan work, just as an example of how some of this integration
is going on in practical A I planning systems.
First of all, we base it on an underlying, intelligible, easily
communicated, easily extended, conceptual model for objectives, processes,
procedures and plans. Just based on four components, a set of
issues to be addressed, a set of nodes, which are activities to be included in
the plan, a set of constraints which have got to be respected and a set of
annotations on the entities that are involved there.
We call that model INCA. Then we want to communicate the dynamic
status and presence of the agents involved.
The collaborative processes that process products, and what they, what they're
able to do in terms of their capabilities.
Want to be sensitive to the current context so the presentation of options
for action are those which are suitable for the context we're in.
And we want to do intelligent activity planning, execution and monitoring
repair, and plan repair, and we do that via the iPlan planner and what we call
the IX process panels, which are the user interface element in the IX technology.
So IXs aim is a planning workflow and task messaging catch all.
We designed it in order that it can deal with the wide range of problems that can
be addressed in planning and activity management.
It's meant to be able to take any requirement to handle an issue,
perform an activity, respect a constraint, or note an
annotation. So that's the underlying computational
model of the I-X platform. They can deal with these by a manual
activity, so you can basically, just take items
off, for instance, on checklist. That's a perfectly good way of
representing the fact that you've performed an activity in some systems.
But also, by inbuilt internal capabilities in the system,
by external capabilities you know about, or by rerouting or delegating to other
panels and agents, which is where communication and collaboration and the
models of those, those other agents comes in.
And all we can plan next year to compensate to these capabilities and we
can use the planner inside the system itself, to plan that kind of work flow.
And then the system receives reports and interprets them to understand the current
state it's in, and to be able to start to handle that
situation and help the user control the situation.
And the idea is it can cope with just partial knowledge of the process and
organization involved, and be able to fit in to an environment
where it isn't the only agent that's working on plans and processes within an
organization. I-x, as we've said before, has been
applied to emerge in serious bonds, and we've used this kind of process panel or
user interface that's what the task. But it's in a context where there is a
lot of verbal tools. The main editors, links to mapping tools,
the planner itself can have a pop up window, its like you look at the options
its generating and guide them and give them advice and so on and you can link to
messaging and communication tools to the systems.
So that concludes this series of presentations on AI planning and context.
We've been looking at the context of practical systems, and we've shown you
some practical planners in use. We've looked at the context of task
assignment, and execution, the fact that we're often working with multiple agents,
multiple systems and multiple services, and we've looked at the context of
planned representation itself, where rich plan representations can be used for many
kinds of communication and collaboration in activity and, and planning situations.
And finally, we looked at Planning++, to try to make sense of the wide range of
techniques being developed in this field, and looked at a model for bringing some
of these techniques together in a productive way.