Okay

welcome everybody.

After constraint satisfaction problems in the last week

where we basically looked at factored

representations

we started a new part of the course

the last but one

where we're

graduating, if you want, to structured representations.

That's the kind of most general representation form that we're going to look.

The idea is that we're going to have a world description language

a formal language that

describes the world state.

We've kind of shied away from that for the factored representations.

We could have done it there

but we haven't

just to leave the new step here.

The idea is that we have a structured language where we fix in the language what we can say

about the world.

But it's not just key value pairs, what we can say about the world.

If you think about factored representations

it's just basically we have a couple of functions

that give us values about the world.

Here we also allow other parts of world descriptions

like how are two objects related

not just

what is this one feature of an object.

And so we're going to work with the following examples.

The Wumpus world, we've looked at it in detail.

It's a little toy world in which we have objects like the Wumpus

the gold

the pits

the agent

and so on.

And they're in relations to each other

and there are also percepts

and the percepts are

also in relation to certain things.

For instance

if you have a pit in one cell

there's a breeze in the adjoining cells.

And so we need something which can describe the world in a little bit more detail than

just a factored representation could.

A factored representation might say things like

oh

there are three pits

and one Wumpus

and one gold

but not where the breezes and so on are in relation to other objects.

So the way the agent works in the world is actually by an inference process.