OK, so welcome to the last AI-1 lecture.

We've been talking about what I called planning and searching

in the real world.

The idea there is that in the last hours of this lecture,

we basically list some of the somewhat unrealistic,

simplifying assumptions about the environment.

So we had, most importantly, assumed

that the environment is fully observable,

deterministic, and static, and all of these kind of things

that make life easier.

And now I just want to give you kind

of a preview of next semester, where we actually

do this in earnest.

But kind of what we need to do, see

what we need to do to our agent architectures

to incorporate some of these things.

So essentially, we can subsume all of this

under the heading of world models of uncertainty,

where we do not know what state the world is in,

where the agent model, we're still

thinking in broad terms about model-based agent,

needs to account for the fact that it's

uncertain about the world state.

And in a nutshell, our solution was to basically say, oh, yeah,

but we have to have a set of possible worlds.

The states the world might be in,

and we'll kind of that set we're going

to call the belief state.

And everything we do, we kind of map to the world state.

And we need something that we call a sensor model,

the sensor model being something that

allows us to make prediction about the world given

certain percepts.

We can also incorporate a sensor fault model there.

We know that sometimes sensors misbehave,

and that might be in there.

There's a fog, and I can only see half

of what I would normally see.

And then I might have artifacts in there.

And we have a transition model which

becomes interesting and non-trivial whenever

we have non-deterministic actions.

Unreliable or partial percepts give us

kind of a branching of what the world, what

the state of the world might actually be.

The same thing with unreliable actions.

Even if we know we're in state A,

we might have a whole set of successor states.

And the question, of course, is how do we deal with this?

The answer for all the things we've done before is not at all.