Showing Revision 1 created 11/28/2012 by Amara Bot.

1. Title:
   08-05 Sensorless Vacumm Cleaner Problem
2. Description:

   Unit 8 5 Sensorless Vacumm Cleaner Problem-

3. 0 0:00 - 0:03
   [Narrator] This is the belief state space
4. 3000 0:03 - 0:05
   for the sensor-less vacuum problem.
5. 5000 0:05 - 0:07
   So we started off here.
6. 7000 0:07 - 0:10
   We drew the circle around this belief state.
7. 10000 0:10 - 0:13
   So we don't anything about where we are,
8. 13000 0:13 - 0:15
   but the amazing thing is,
9. 15000 0:15 - 0:17
   if we execute actions, we can gain knowledge
10. 17000 0:17 - 0:20
    about the world even without sensing.
11. 20000 0:20 - 0:22
    So let's say we move right,
12. 22000 0:22 - 0:26
    then we'll know we're in the right-hand location.
13. 26000 0:26 - 0:28
    Either we were in the left, and we moved right
14. 28000 0:28 - 0:30
    and arrived there, or we were in the right
15. 30000 0:30 - 0:32
    to begin with, and we bumped against the wall
16. 32000 0:32 - 0:34
    and stayed there.
17. 34000 0:34 - 0:37
    So now we end up in this state.
18. 37000 0:37 - 0:40
    We now know more about the world.
19. 40000 0:40 - 0:43
    We're down to 4 possibilities rather than 8,
20. 43000 0:43 - 0:46
    even though we haven't observed anything,
21. 46000 0:46 - 0:48
    and now note something interesting,
22. 48000 0:48 - 0:50
    that in the real world, the operations
23. 50000 0:50 - 0:52
    of going left and going right are
24. 52000 0:52 - 0:54
    inverses of each other, but
25. 54000 0:54 - 0:56
    in the belief state world
26. 56000 0:56 - 0:59
    going right and going left are not inverses.
27. 59000 0:59 - 1:01
    If we go right, and then we go left,
28. 61000 1:01 - 1:03
    we don't end up back where we were
29. 63000 1:03 - 1:05
    in a state of total uncertainty, rather
30. 65000 1:05 - 1:08
    going left takes us over here
31. 68000 1:08 - 1:10
    where we still know we're in 1 of 4 states
32. 70000 1:10 - 1:13
    rather than in 1 of 8 states.
33. 73000 1:13 - 1:15
    Note that it's possible to form a plan that
34. 75000 1:15 - 1:18
    reaches a goal without ever observing the world.
35. 78000 1:18 - 1:21
    Plans like that are called conform-it plans.
36. 81000 1:21 - 1:23
    For example, if the goal is to be
37. 83000 1:23 - 1:25
    in a clean location
38. 85000 1:25 - 1:28
    all we have to do is suck.
39. 88000 1:28 - 1:30
    So we go from one of these 8 states
40. 90000 1:30 - 1:32
    to one of these 4 states and,
41. 92000 1:32 - 1:34
    every one of those 4,
42. 94000 1:34 - 1:36
    we're in a clean location.
43. 96000 1:36 - 1:38
    We don't know which of the 4 we're in,
44. 98000 1:38 - 1:41
    but we know we've achieved the goal.
45. 101000 1:41 - 1:43
    It's also possible to arrive
46. 103000 1:43 - 1:45
    at a completely known state.
47. 105000 1:45 - 1:47
    For example, if we start here,
48. 107000 1:47 - 1:50
    we go left; we suck up the dirt there.
49. 110000 1:50 - 1:53
    If we go right and suck up the dirt,
50. 113000 1:53 - 1:55
    now we're down to a belief state
51. 115000 1:55 - 1:57
    consisting of 1 single state that is
52. 117000 1:57 - 2:00
    we know exactly where we are.
53. 120000 2:00 - 2:02
    Here's a question for you:
54. 122000 2:02 - 2:04
    How do I get from the state where I know
55. 124000 2:04 - 2:06
    my current square is clean,
56. 126000 2:06 - 2:08
    but know nothing else, to the belief state
57. 128000 2:08 - 2:10
    where I know that I'm in the right-hand side
58. 130000 2:10 - 2:14
    location and that that location is clean?
59. 134000 2:14 - 2:16
    What I want you to do is click on the
60. 136000 2:16 - 2:18
    sequence of actions, left, right, or suck
61. 138000 2:18 -
    that will take us from that start to that goal.