We're talking about planning.

And the new thing is that

our actions have add and delete lists.

And we're doing search at the description logic,

which means we have some kind of a logic for

describing the worlds we are in.

And then we do search.

So, and today we're going to go into

the specifics of planning search algorithms.

And kind of go through

a relatively detailed account of

how relaxed heuristics,

which we've learned in the beginning about,

actually perform in practice.

Yesterday, we mainly looked at planning complexity

and at a very concrete planning language,

this plan description language PDDL.

Which is essentially a language that uses

some form of first order logic written down in

Lisp form with lots of brackets,

which is a tradition in AI.

And where you can define both the domains.

The domains you basically specify.

The predicates you can use

to describe the world and the actions.

The actions have parameters,

which is something you don't

have in propositional logic.

Typically, you have kind of family of actions,

which is stack A on B,

stack B on C and so on.

And you want to write that as stack X on Y.

And then these can be instantiated

and make it easier to write down.

Things you have the preconditions

and you have the effects.

And you have both positive effects.

That's the at list in strips and negative effects,

which is the delete list in strips.

Pretty straightforward.

Something like that if you had gotten

the task of writing down a language,

you'd probably come up with something very similar.

So, this is a domain description that basically

tells you what the predicates and the actions are.

And you can instantiate that into it with

a problem description that tells you what the initial,

what the objects are that you're talking about,

what the initial state is and the goal state.

If you have a domain description