They address some of these problems.

Let me play this again.

Status Filter

Now let's look into particle filters, the subject of today's class, and it's really interesting to see the answers for particle filters.

Thrun This is a great segue to one of the most successful algorithms in artificial intelligence and robotics called particle filters.

It's pretty horrible if you want to search systematically.

Obviously the field is much, much bigger than what I just showed you, but I gave you examples of the key elements.

So, it's a cyclic world.

In fact, it is completely uncertain as to where it is.

This is the essence of particle filters.

The sum or set of all those vectors together form the belief space.

Check the following statements if they're true.

So for 100 dimensions it's hard to make work.

And only the first answer is correct.

Of course, we learned that the Kalman filter is unimodal, just a single bump, whereas histograms and particles can have multiple bumps.

Now, next is your homework assignment. I hope you can prove what you've learned and ace it.

The histogram method and now the Kalman method are really complex and involved.

You see a higher density of particles over here and over here, than you see, for example, over here.

We resample. We do forward prediction.

In question 3 I'd like to quiz your intuition.

But if you understand it, you can now go and implement particle filters for the great range of robotics applications.

Here is a localization quiz.

But there are extensions.

Over the past 10 years or so, I have worked on a number of different methods for buildings maps that are quite sophisticated,

It's much less likely.

Narrator And it turns out it's actually harder than you think, but I'm going to show you how to do it, and once you've done it, you can use the exact same code forever for any particle filter.

They have some deficiencies.

Welcome to homework assignment number 3.

Absolutely, yes.

I just ran it 20 times, and it said true for me every single time.

This geo engineering idea, in it's simplest form, is basically the following.

Thrun Wow! You learned a lot about hidden Markov models and particle filters.

I couldn't get the curser keys to work on my computer, but it's a really great basis to visualize.

So here's a multiple choice question for particle filters.

They're not independently drawn.

Next Class So, in the next class, I'm going to teach you our final state estimation algorithm.

And that's important. Many other filters can't.

We will talk today of localization.

The Kalman filter is the only filter that does scale exponentially, so it's very nice.

Okay. That's a great segue this question about switching between filters on the fly.

We generate a robot that finished with final location 20, 29, and this orientation over here.

But in other domains, mostly in tracking domains, they tend to scale much, much better, and I've not seen a good treatment yet of the complexity in practice for particle filters.

Stores the last status filter setting

There are ways to combine multiple filters types.

Now, in our final programming exercise,