Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← 04-41 Kalman Filter Prediction

04-41 Kalman Filter Prediction

Get Embed Code
3 Languages

Showing Revision 4 created 04/19/2017 by Udacity Robot.

  1. In Kalman filter land, we're going to build a 2-dimensional estimate.
  2. 1 for the location, and 1 for the velocity denoted x dot.
  3. The velocity can be zero. It can be negative, or it can be positive.
  4. If initially I know my location, but not my velocity,
  5. then I represent it with a Gaussian that's elevated around the correct location,
  6. but really, really broad in the space of velocities.
  7. Let's look at the prediction step.
  8. In the prediction step, I don't know my velocity,
  9. so I can't possibly predict for location. I'm going to assume.
  10. But miraculously, there'll be some interesting correlation.
  11. So let's for a second, just pick a point on this distribution over here.
  12. Let me assume my velocity is 0.
  13. Of course, in practice, I don't know the velocity,
  14. but let me assume for a moment the velocity is 0.
  15. Where would my posterior be after the prediction?
  16. Well, we know we started in location 1.
  17. The velocity is 0, so my location would likely be here.
  18. Now let's change my belief in velocity and pick a different one.
  19. Let's say the velocity is 1.
  20. Where would my prediction be 1 time step later starting at location 1 and velocity 1?
  21. I'll give you 3 choices.
  22. Here? Here? Or here?
  23. Please pick the one that makes the most sense.