So it's a good thing that the quiz starts automatically and you don't need me to be

here really. I've considered coming later but then... yeah. So there we go. So we've

just last week completed the chapter on knowledge representation which is really kind of a fine

tuning of logics for a certain kind of applications, namely relatively scalable world descriptions and

querying for properties of this world given an A box and a T box. And the idea was that we could

make logic agents. And the last chapter that we're going to look at we can think of as kind of

two... in two ways. One of them is essentially rather than having loads of little search

algorithms, uniform cost search and iterative deepening search and A star search and all of

those kind of things with all kinds of little heuristics and rather just basically what we

would like to have is just one algorithm to do all the search for us. That's one way of looking

at it. The other way of looking at planning is that we add time. In our agents we have

things that change. Knowledge about the world that is not monotonic. Knowledge about the world

like I am in Erlangen now. That changes as soon as I actually get on board a train to Munich.

I want to replace the fact that I'm in Erlangen now with the fact that I'm in Munich now. And the

surprising thing is that those two things are not only compatible but they're essentially the same

thing. And that's what we want to explore now. So if we remember our search algorithms we had

assumed a model-based agent, something that keeps track of the state of the world and has a transition

model that tells us how certain percepts and certain actions actually change the world. And

we already have this thing again. We change the knowledge about the world. Only for search it was

very very very easy because we have these unstructured states of the world. We had gazillions

of them. And so we really weren't focused on the fact that we were actually talking about time.

But if we think about using world description languages then we have to introduce time. Because

think about traveling in Romania. There's time even if it's just the next state the next time.

And we're going to look at that. We had states, we had actions, we had initial states, we had goal

states and we had solutions. And we instantiated that into different applications. Playing solitaire,

making train connections in the Indian railway system, whatever. And if you look at this ambition

of writing one program that kind of does all the search problems, then we already did something

like this in CSPs. We kind of had a world description language which is factored world

descriptions. And we were able to do the trick by essentially going to the world description level.

Only that CSPs don't really have a notion of time. Those were a kind of configuration problems. You

didn't need time for them. Rather than planning something in a changing environment, you're in

ARAT now and CBU and so on, we just basically had to configure something so that it actually fits.

But we still had this kind of a problem of a black box description language like in search and a,

in this case, factored world description level search that was much more efficient. And essentially

we went from kind of a black box description where we just had predicate that says, am I in an initial

state? Am I in a goal state? That you could call on a state. We went to a descriptive language where

you could actually see what's happening. Where you could take a description of the problem,

massage it into an equivalent tighter one until you could just read off the solution.

And that's what we saw. And something we saw here. And really it's all about these world

description languages. And in this chapter we're going to call those planning languages or planning

tasks. Okay. So that's kind of the same general idea, slightly different technical content.

Now we're switching over to the other kind of perspective. World description language,

that's what something we did with the Wampas already, right? How can we do that? How can we

go to search here? And that gives rise to a way of planning that we call logic based planning.

And it just basically follows the idea, namely, we'll take a logic and we add time to it. So

we can talk about the Wampas world in logics. We've seen it in propositional logic,

we've seen it in first order logic. And I'm sure you believe that we can do it in ALC or so as well.

Somewhere in the middle. We never quite did it, but you could do it if I gave it to you as a homework

or an exam question. Probably no big problem. Right? So if we kind of drill in on this idea that

we're the agent, it moves. It replaces knowledge about in which cell it is with knowledge about