Good morning and welcome to the last AIIns lecture.

We've been talking about planning.

Planning as a search procedure where we take special

consideration of the fact that the world can change.

It is something we didn't have to

do for regular search procedures because we had atomic states.

We couldn't look into the

states.

Now we're using world description languages and the world description languages

We've been looking into

logics don't do very well in describing time

describing change

especially describing things that become untrue at some point.

Logics have been, by and large,

developed to do monotone knowledge.

Things that at some point are true and will always stay true.

There's a couple of ways around.

There's a way to just basically give everything a time argument.

We looked into that.

That is not very attractive because you need frame axioms basically telling us

what remains true when the clock ticks.

There are special logics called modal logics, temporal logics,

that kind of have time being built in, in a much deeper level.

We don't have time to look at this

but that's an active research area

especially if you want to do hardware verification where basically

the hardware clock actually plays the role.

You use logic like that.

And for planning, we basically

do this by a trick, namely the delete lists.

The delete lists that just remove facts from the world description.

That's quite effective and that's kind of the framework we're exploring right now.

We've done

partial order planning.

We've done heuristic search of algorithms.

Both of these are active areas of research.

And now we want to kind of extend the framework

see whether we can extend it to real world situations.

And we've just started looking at the furniture coloring example

a very simple little example where we

can deal with unknown, with partial observability.

The partial observability here is that we have two cans of unknown color.

So there's information missing and we don't know what the color

the initial color of the chair is.

And we still have to do planning.

We just have to extend the algorithms for that.

Unfortunately, some of the algorithms can be extended rather well.

OK, one of the things is we can write it all down in PDL.

We might have to deal with the fact that there are unknowns.

And here