WEBVTT

1
00:00:05.100 --> 00:00:09.700
<00:00:05.100>You <00:00:06.500>are

2
00:00:08.700 --> 00:00:12.200
<00:00:08.700>right. <00:00:10.100>Good

3
00:00:11.200 --> 00:00:14.500
<00:00:11.200>morning <00:00:11.375>everyone <00:00:11.549>and <00:00:11.725>thank <00:00:11.899>you <00:00:12.075>for <00:00:12.250>the <00:00:12.424>introduction. <00:00:12.600>So

4
00:00:14.500 --> 00:00:17.300
<00:00:14.500>my <00:00:14.640>name <00:00:14.780>is <00:00:14.920>Liang <00:00:15.060>Yu <00:00:15.200>I <00:00:15.340>am <00:00:15.480>a <00:00:15.620>data <00:00:15.760>scientist <00:00:15.900>and

5
00:00:17.300 --> 00:00:20.700
<00:00:17.300>Tech <00:00:17.533>development <00:00:17.766>lead <00:00:18.000>at <00:00:18.233>cariad. <00:00:18.466>And <00:00:18.700>today

6
00:00:20.700 --> 00:00:23.400
<00:00:20.700>I <00:00:20.875>would <00:00:21.050>like <00:00:21.224>to <00:00:21.400>talk <00:00:21.575>about <00:00:21.750>a <00:00:21.925>topic <00:00:22.099>that

7
00:00:23.400 --> 00:00:26.300
<00:00:23.400>is <00:00:23.555>of <00:00:23.711>Paramount <00:00:23.866>importance <00:00:24.022>in <00:00:24.177>the <00:00:24.333>field <00:00:24.488>of <00:00:24.644>autonomous <00:00:24.799>driving.

8
00:00:26.300 --> 00:00:30.200
<00:00:26.300>That <00:00:26.475>is <00:00:26.650>how <00:00:26.825>do <00:00:27.000>we <00:00:27.175>find <00:00:27.350>the <00:00:27.525>unknown <00:00:27.700>unknowns?

9
00:00:31.200 --> 00:00:34.300
<00:00:31.200>How <00:00:31.316>do <00:00:31.433>we <00:00:31.550>know <00:00:31.666>when <00:00:31.783>our <00:00:31.900>perception <00:00:32.016>model <00:00:32.133>is <00:00:32.250>failing <00:00:32.366>on <00:00:32.483>the <00:00:32.600>road?

10
00:00:35.200 --> 00:00:38.100
<00:00:35.200>Even <00:00:35.375>after <00:00:35.550>driving <00:00:35.725>millions <00:00:35.900>of <00:00:36.075>miles, <00:00:36.250>they <00:00:36.425>could <00:00:36.600>still

11
00:00:38.100 --> 00:00:41.400
<00:00:38.100>be <00:00:38.300>anomalous <00:00:38.500>scenarios <00:00:38.700>or <00:00:38.900>Corner <00:00:39.100>cases <00:00:39.300>that <00:00:39.500>a

12
00:00:41.400 --> 00:00:42.800
<00:00:41.400>self-driving <00:00:41.610>car <00:00:41.820>doesn't <00:00:42.030>know <00:00:42.239>how <00:00:42.449>to <00:00:42.660>handle.

13
00:00:43.600 --> 00:00:46.500
<00:00:43.600>So <00:00:43.716>it's <00:00:43.833>very <00:00:43.950>important <00:00:44.066>that <00:00:44.183>we <00:00:44.300>know <00:00:44.416>when <00:00:44.533>these <00:00:44.650>happen <00:00:44.766>and <00:00:44.883>can <00:00:45.000>identify

14
00:00:46.500 --> 00:00:49.200
<00:00:46.500>them. <00:00:46.655>So <00:00:46.811>today <00:00:46.966>I <00:00:47.122>like <00:00:47.277>to <00:00:47.433>talk <00:00:47.588>to <00:00:47.744>you <00:00:47.900>about

15
00:00:49.200 --> 00:00:52.100
<00:00:49.200>how <00:00:49.375>my <00:00:49.550>team <00:00:49.725>at <00:00:49.900>cariad <00:00:50.075>is <00:00:50.250>taking <00:00:50.425>this <00:00:50.600>problem.

16
00:00:54.600 --> 00:00:57.400
<00:00:54.600>And <00:00:54.775>first <00:00:54.950>of <00:00:55.125>all, <00:00:55.300>what <00:00:55.475>is <00:00:55.650>carried <00:00:55.825>and <00:00:56.000>who

17
00:00:57.400 --> 00:01:01.100
<00:00:57.400>are <00:00:57.633>we <00:00:57.866>caread <00:00:58.100>is <00:00:58.333>an <00:00:58.566>automotive <00:00:58.800>software

18
00:01:00.100 --> 00:01:03.300
<00:01:00.100>and <00:01:00.380>technology <00:01:00.660>company <00:01:00.940>and <00:01:01.220>it's <00:01:01.500>a

19
00:01:03.300 --> 00:01:05.100
<00:01:03.300>100% <00:01:04.000>Volkswagen <00:01:04.700>subsidiary.

20
00:01:05.900 --> 00:01:08.900
<00:01:05.900>So <00:01:06.100>our <00:01:06.300>mission <00:01:06.500>is <00:01:06.700>to <00:01:06.900>create <00:01:07.100>one <00:01:07.300>unified

21
00:01:08.900 --> 00:01:11.300
<00:01:08.900>software <00:01:09.250>platform <00:01:09.600>and <00:01:09.950>techstack <00:01:10.300>for

22
00:01:11.300 --> 00:01:12.500
<00:01:11.300>all <00:01:11.840>Volkswagen <00:01:12.380>brands.

23
00:01:14.200 --> 00:01:17.700
<00:01:14.200>In <00:01:14.340>order <00:01:14.480>to <00:01:14.620>speed <00:01:14.760>up <00:01:14.900>the <00:01:15.040>Innovation <00:01:15.180>speed <00:01:15.320>of <00:01:15.460>Volkswagen <00:01:15.600>group

24
00:01:17.700 --> 00:01:20.900
<00:01:17.700>and <00:01:17.875>transform <00:01:18.050>it <00:01:18.225>to <00:01:18.400>a <00:01:18.575>truly <00:01:18.750>digital <00:01:18.925>and <00:01:19.100>software

25
00:01:20.900 --> 00:01:22.200
<00:01:20.900>driven <00:01:21.485>Mobility <00:01:22.070>provider.

26
00:01:24.600 --> 00:01:28.400
<00:01:24.600>And <00:01:24.755>we <00:01:24.911>are <00:01:25.066>currently <00:01:25.222>working <00:01:25.377>on <00:01:25.533>the <00:01:25.688>next <00:01:25.844>gen <00:01:26.000>uniform.

27
00:01:29.100 --> 00:01:31.400
<00:01:29.100>platform <00:01:29.800>called <00:01:30.500>the

28
00:01:32.500 --> 00:01:35.500
<00:01:32.500>EQ <00:01:32.780>2.0. <00:01:33.060>So <00:01:33.340>this <00:01:33.620>consists <00:01:33.900>of

29
00:01:35.500 --> 00:01:38.600
<00:01:35.500>three <00:01:35.640>different <00:01:35.780>layers <00:01:35.920>one <00:01:36.060>of <00:01:36.200>which <00:01:36.340>is <00:01:36.480>the <00:01:36.620>hard <00:01:36.760>way <00:01:36.900>layer

30
00:01:38.600 --> 00:01:41.500
<00:01:38.600>which <00:01:38.774>is <00:01:38.949>in <00:01:39.125>the <00:01:39.300>car <00:01:39.475>and <00:01:39.649>the <00:01:39.824>software <00:01:40.000>layer

31
00:01:41.500 --> 00:01:42.500
<00:01:41.500>and <00:01:41.800>the <00:01:42.100>operating <00:01:42.400>system.

32
00:01:43.400 --> 00:01:46.300
<00:01:43.400>And <00:01:43.540>also <00:01:43.680>the <00:01:43.820>automotive <00:01:43.960>cloud <00:01:44.100>with <00:01:44.240>all <00:01:44.380>of <00:01:44.520>its <00:01:44.660>data <00:01:44.800>centers

33
00:01:46.300 --> 00:01:49.400
<00:01:46.300>and <00:01:46.427>this <00:01:46.554>will <00:01:46.681>be <00:01:46.809>rolled <00:01:46.936>out <00:01:47.063>in <00:01:47.190>folks. <00:01:47.318>Welcome <00:01:47.445>brand <00:01:47.572>cost <00:01:47.700>starting

34
00:01:49.400 --> 00:01:50.500
<00:01:49.400>from <00:01:49.598>the <00:01:49.796>middle <00:01:49.994>of <00:01:50.192>this <00:01:50.390>decade.

35
00:01:52.300 --> 00:01:55.500
<00:01:52.300>And <00:01:52.440>using <00:01:52.580>the <00:01:52.720>system <00:01:52.860>we <00:01:53.000>are <00:01:53.140>trying <00:01:53.280>to <00:01:53.420>develop <00:01:53.560>test <00:01:53.700>and

36
00:01:55.500 --> 00:01:58.600
<00:01:55.500>deploy <00:01:55.780>new <00:01:56.060>software <00:01:56.340>for <00:01:56.620>autonomous <00:01:56.900>driving

37
00:01:58.600 --> 00:01:59.000
functions.

38
00:02:01.300 --> 00:02:02.500
<00:02:01.300>and <00:02:01.660>within <00:02:02.020>this <00:02:02.380>domain

39
00:02:03.300 --> 00:02:06.500
<00:02:03.300>we <00:02:03.416>are <00:02:03.533>trying <00:02:03.649>to <00:02:03.766>dedicate <00:02:03.883>a <00:02:04.000>lot <00:02:04.116>of <00:02:04.233>effort <00:02:04.350>and <00:02:04.466>energy <00:02:04.583>to <00:02:04.700>data

40
00:02:06.500 --> 00:02:09.500
<00:02:06.500>acquisition. <00:02:06.675>And <00:02:06.850>the <00:02:07.025>reason <00:02:07.200>why <00:02:07.375>we <00:02:07.550>do <00:02:07.725>this <00:02:07.900>is

41
00:02:09.500 --> 00:02:12.400
<00:02:09.500>because <00:02:09.699>start <00:02:09.900>acquisition <00:02:10.100>is <00:02:10.300>incredibly <00:02:10.500>important <00:02:10.699>for <00:02:10.900>a

42
00:02:12.400 --> 00:02:15.100
<00:02:12.400>lot <00:02:12.540>of <00:02:12.680>different <00:02:12.820>functions <00:02:12.960>that <00:02:13.100>we're <00:02:13.240>trying <00:02:13.380>to <00:02:13.520>perform <00:02:13.660>if <00:02:13.800>we

43
00:02:15.100 --> 00:02:18.500
<00:02:15.100>want <00:02:15.227>to <00:02:15.354>test <00:02:15.481>the <00:02:15.609>performance <00:02:15.736>of <00:02:15.863>a <00:02:15.990>piece <00:02:16.118>of <00:02:16.245>software <00:02:16.372>or <00:02:16.500>if

44
00:02:18.500 --> 00:02:21.500
<00:02:18.500>we <00:02:18.655>want <00:02:18.811>to <00:02:18.966>develop <00:02:19.122>a <00:02:19.277>new <00:02:19.433>perception <00:02:19.588>model <00:02:19.744>if <00:02:19.900>we

45
00:02:21.500 --> 00:02:25.300
<00:02:21.500>want <00:02:21.700>to <00:02:21.900>train <00:02:22.100>a <00:02:22.300>new <00:02:22.500>model <00:02:22.700>all <00:02:22.900>these

46
00:02:24.300 --> 00:02:26.600
<00:02:24.300>we <00:02:24.580>need <00:02:24.860>vast <00:02:25.140>amounts <00:02:25.420>of <00:02:25.700>data.

47
00:02:27.600 --> 00:02:30.600
<00:02:27.600>And <00:02:27.880>traditionally <00:02:28.160>data <00:02:28.440>acquisition <00:02:28.720>is <00:02:29.000>done

48
00:02:30.600 --> 00:02:33.600
<00:02:30.600>by <00:02:30.833>recording <00:02:31.066>every <00:02:31.299>piece <00:02:31.533>of <00:02:31.766>information <00:02:32.000>that

49
00:02:33.600 --> 00:02:36.200
<00:02:33.600>comes <00:02:33.739>through <00:02:33.880>the <00:02:34.019>sensors <00:02:34.160>onto <00:02:34.299>some <00:02:34.440>local <00:02:34.579>storage <00:02:34.720>device <00:02:34.859>in <00:02:35.000>the

50
00:02:36.200 --> 00:02:36.500
car.

51
00:02:37.700 --> 00:02:41.100
<00:02:37.700>And <00:02:37.816>then <00:02:37.933>at <00:02:38.049>the <00:02:38.166>end <00:02:38.283>of <00:02:38.399>day <00:02:38.516>this <00:02:38.633>has <00:02:38.750>to <00:02:38.866>be <00:02:38.983>manually <00:02:39.100>transferred

52
00:02:40.100 --> 00:02:42.800
<00:02:40.100>to <00:02:40.450>some <00:02:40.799>Central <00:02:41.150>storage <00:02:41.500>area.

53
00:02:43.600 --> 00:02:46.600
<00:02:43.600>And <00:02:43.775>of <00:02:43.950>course, <00:02:44.125>this <00:02:44.299>is <00:02:44.475>incredibly <00:02:44.650>expensive <00:02:44.825>in <00:02:45.000>terms

54
00:02:46.600 --> 00:02:49.300
<00:02:46.600>of <00:02:46.880>both <00:02:47.160>labor <00:02:47.440>and <00:02:47.720>storage <00:02:48.000>capacity.

55
00:02:50.200 --> 00:02:51.300
<00:02:50.200>It's <00:02:50.530>also <00:02:50.860>very <00:02:51.190>inefficient.

56
00:02:52.400 --> 00:02:52.600
because

57
00:02:53.800 --> 00:02:56.100
<00:02:53.800>in <00:02:54.000>addition <00:02:54.200>to <00:02:54.400>the <00:02:54.600>useful <00:02:54.800>information <00:02:55.000>it <00:02:55.200>also

58
00:02:56.100 --> 00:02:59.300
<00:02:56.100>provides <00:02:56.275>our <00:02:56.450>developers <00:02:56.625>with <00:02:56.799>a <00:02:56.975>large <00:02:57.150>amount <00:02:57.325>of <00:02:57.500>junk.

59
00:03:00.200 --> 00:03:01.900
<00:03:00.200>So <00:03:00.399>at <00:03:00.600>carryout <00:03:00.799>our <00:03:01.000>solution <00:03:01.200>to <00:03:01.399>this <00:03:01.600>problem.

60
00:03:02.900 --> 00:03:05.100
<00:03:02.900>Is <00:03:03.133>called <00:03:03.366>the <00:03:03.600>big <00:03:03.833>loop. <00:03:04.066>It <00:03:04.300>is

61
00:03:05.100 --> 00:03:08.600
<00:03:05.100>an <00:03:05.333>intelligent <00:03:05.566>data <00:03:05.799>aggregation <00:03:06.033>system <00:03:06.266>that's <00:03:06.500>capable

62
00:03:08.600 --> 00:03:11.900
<00:03:08.600>of <00:03:08.799>dramatically <00:03:09.000>reducing <00:03:09.200>the <00:03:09.400>amount <00:03:09.600>of <00:03:09.799>information <00:03:10.000>recorded.

63
00:03:12.700 --> 00:03:15.400
<00:03:12.700>By <00:03:12.875>only <00:03:13.049>capturing <00:03:13.225>the <00:03:13.399>useful <00:03:13.575>information <00:03:13.750>and <00:03:13.924>filtering <00:03:14.100>out

64
00:03:15.400 --> 00:03:16.200
<00:03:15.400>all <00:03:15.760>the <00:03:16.120>junk.

65
00:03:17.200 --> 00:03:20.800
<00:03:17.200>It <00:03:17.433>also <00:03:17.666>provides <00:03:17.899>concurrent <00:03:18.133>access <00:03:18.366>to <00:03:18.600>all

66
00:03:20.800 --> 00:03:23.100
<00:03:20.800>the <00:03:20.955>data <00:03:21.111>that <00:03:21.266>we <00:03:21.422>collected <00:03:21.577>for <00:03:21.733>our <00:03:21.888>developers <00:03:22.044>to <00:03:22.200>look

67
00:03:23.100 --> 00:03:23.500
<00:03:23.100>at <00:03:23.460>them.

68
00:03:25.400 --> 00:03:28.500
<00:03:25.400>And <00:03:25.575>the <00:03:25.750>hardware <00:03:25.925>for <00:03:26.100>the <00:03:26.275>big <00:03:26.450>loop <00:03:26.625>resides <00:03:26.800>inside

69
00:03:28.500 --> 00:03:31.200
<00:03:28.500>the <00:03:28.780>scalable <00:03:29.060>compute <00:03:29.340>platform <00:03:29.620>in <00:03:29.900>the

70
00:03:31.200 --> 00:03:32.700
<00:03:31.200>car. <00:03:31.424>It's <00:03:31.649>a <00:03:31.875>size <00:03:32.100>of <00:03:32.325>a <00:03:32.549>shoebox.

71
00:03:33.900 --> 00:03:36.400
<00:03:33.900>So <00:03:34.133>here <00:03:34.366>you <00:03:34.600>see <00:03:34.833>the <00:03:35.066>Adas <00:03:35.300>blade.

72
00:03:37.200 --> 00:03:37.900
<00:03:37.200>And <00:03:37.410>this <00:03:37.620>is <00:03:37.830>where?

73
00:03:39.100 --> 00:03:42.800
<00:03:39.100>This <00:03:39.275>is <00:03:39.450>the <00:03:39.625>software <00:03:39.799>that <00:03:39.975>takes <00:03:40.150>care <00:03:40.325>of <00:03:40.500>all

74
00:03:42.800 --> 00:03:45.800
<00:03:42.800>the <00:03:43.033>Adas <00:03:43.266>full <00:03:43.500>driving <00:03:43.733>stack <00:03:43.966>capabilities <00:03:44.200>and

75
00:03:45.800 --> 00:03:48.100
<00:03:45.800>right <00:03:45.955>next <00:03:46.111>to <00:03:46.266>it. <00:03:46.422>We <00:03:46.577>have <00:03:46.733>the <00:03:46.888>protected <00:03:47.044>area <00:03:47.200>blade.

76
00:03:48.900 --> 00:03:51.300
<00:03:48.900>So <00:03:49.100>this <00:03:49.300>protected <00:03:49.500>area <00:03:49.700>blade <00:03:49.900>is <00:03:50.100>isolated <00:03:50.300>from

77
00:03:51.300 --> 00:03:54.000
<00:03:51.300>interference <00:03:51.580>or <00:03:51.860>any <00:03:52.140>interaction <00:03:52.420>with <00:03:52.700>the

78
00:03:54.300 --> 00:03:57.300
<00:03:54.300>rest <00:03:54.416>of <00:03:54.533>the <00:03:54.650>vehicle. <00:03:54.766>So <00:03:54.883>here <00:03:55.000>we <00:03:55.116>can <00:03:55.233>safely <00:03:55.350>deploy <00:03:55.466>and <00:03:55.583>Test <00:03:55.700>new

79
00:03:57.300 --> 00:03:57.800
software.

80
00:03:58.700 --> 00:04:01.400
<00:03:58.700>While <00:03:58.899>ensuring <00:03:59.100>the <00:03:59.299>safety <00:03:59.500>of <00:03:59.700>passengers <00:03:59.899>inside <00:04:00.100>the

81
00:04:01.400 --> 00:04:01.600
car.

82
00:04:04.400 --> 00:04:07.700
<00:04:04.400>Okay, <00:04:04.540>so <00:04:04.680>the <00:04:04.820>protected <00:04:04.960>area <00:04:05.100>blade <00:04:05.240>is <00:04:05.380>the <00:04:05.520>hot <00:04:05.660>of <00:04:05.800>the

83
00:04:07.700 --> 00:04:10.300
<00:04:07.700>big <00:04:07.875>loop <00:04:08.049>and <00:04:08.225>my <00:04:08.399>team <00:04:08.575>at <00:04:08.750>cariad <00:04:08.924>also <00:04:09.100>known

84
00:04:10.300 --> 00:04:11.300
<00:04:10.300>as <00:04:11.200>ingenuity.

85
00:04:11.900 --> 00:04:14.500
<00:04:11.900>And <00:04:12.027>we <00:04:12.154>use <00:04:12.281>the <00:04:12.409>protected <00:04:12.536>area <00:04:12.663>blade <00:04:12.790>for <00:04:12.918>two <00:04:13.045>functions <00:04:13.172>one <00:04:13.300>of

86
00:04:14.500 --> 00:04:17.200
<00:04:14.500>which <00:04:14.675>is <00:04:14.850>shadowing <00:04:15.025>so <00:04:15.200>that <00:04:15.375>we <00:04:15.550>can <00:04:15.725>monitor <00:04:15.900>the

87
00:04:17.200 --> 00:04:20.900
<00:04:17.200>performance <00:04:17.399>of <00:04:17.599>software <00:04:17.800>in <00:04:18.000>real-world <00:04:18.200>conditions. <00:04:18.399>And <00:04:18.599>the

88
00:04:20.900 --> 00:04:23.200
<00:04:20.899>second <00:04:21.179>one <00:04:21.459>is <00:04:21.739>intelligent <00:04:22.019>data <00:04:22.299>collection

89
00:04:23.200 --> 00:04:26.600
<00:04:23.200>by <00:04:23.375>leveraging <00:04:23.550>the <00:04:23.724>compute <00:04:23.900>power <00:04:24.075>of <00:04:24.250>this <00:04:24.425>protect <00:04:24.599>area

90
00:04:26.600 --> 00:04:26.800
blade.

91
00:04:28.700 --> 00:04:31.400
<00:04:28.700>And <00:04:28.855>the <00:04:29.011>big <00:04:29.166>look <00:04:29.322>in <00:04:29.477>encompasses <00:04:29.633>many <00:04:29.788>different <00:04:29.944>components <00:04:30.099>as

92
00:04:31.400 --> 00:04:32.900
<00:04:31.400>you <00:04:31.625>can <00:04:31.849>see <00:04:32.075>here <00:04:32.299>in <00:04:32.525>this <00:04:32.750>image.

93
00:04:34.100 --> 00:04:37.200
<00:04:34.100>Many <00:04:34.240>different <00:04:34.380>functions. <00:04:34.520>But <00:04:34.660>today <00:04:34.800>I <00:04:34.940>want <00:04:35.080>to <00:04:35.220>just <00:04:35.360>concentrate <00:04:35.500>on

94
00:04:37.200 --> 00:04:40.500
<00:04:37.200>one <00:04:37.340>particular <00:04:37.479>use <00:04:37.620>case <00:04:37.760>and <00:04:37.900>that <00:04:38.039>is <00:04:38.180>related <00:04:38.320>to <00:04:38.460>the <00:04:38.599>question

95
00:04:40.500 --> 00:04:43.200
<00:04:40.500>that <00:04:40.640>I <00:04:40.780>post <00:04:40.920>at <00:04:41.060>the <00:04:41.200>beginning <00:04:41.340>of <00:04:41.480>the <00:04:41.620>presentation, <00:04:41.760>which <00:04:41.900>is

96
00:04:43.200 --> 00:04:45.500
<00:04:43.200>how <00:04:43.433>do <00:04:43.666>we <00:04:43.900>find <00:04:44.133>the <00:04:44.366>unknown <00:04:44.599>unknowns?

97
00:04:46.300 --> 00:04:49.300
<00:04:46.300>How <00:04:46.427>do <00:04:46.554>we <00:04:46.681>know <00:04:46.809>whale <00:04:46.936>models <00:04:47.063>failing <00:04:47.190>on <00:04:47.318>the <00:04:47.445>road? <00:04:47.572>And <00:04:47.700>how

98
00:04:49.300 --> 00:04:52.500
<00:04:49.300>do <00:04:49.416>we <00:04:49.533>collect <00:04:49.650>data <00:04:49.766>on <00:04:49.883>these <00:04:50.000>Corner <00:04:50.116>cases <00:04:50.233>to <00:04:50.350>improve <00:04:50.466>the <00:04:50.583>model's <00:04:50.700>performance

99
00:04:52.500 --> 00:04:52.800
<00:04:52.500>in <00:04:52.635>the <00:04:52.770>future?

100
00:04:56.400 --> 00:04:59.100
<00:04:56.400>Okay, <00:04:56.507>so <00:04:56.615>I <00:04:56.723>want <00:04:56.830>to <00:04:56.938>explain <00:04:57.046>a <00:04:57.153>little <00:04:57.261>bit <00:04:57.369>more <00:04:57.476>detail <00:04:57.584>what <00:04:57.692>I <00:04:57.799>mean

101
00:04:59.100 --> 00:05:00.300
<00:04:59.100>by <00:04:59.640>unknown <00:05:00.180>unknowns.

102
00:05:01.200 --> 00:05:04.300
<00:05:01.200>So <00:05:01.340>if <00:05:01.479>we <00:05:01.620>plot <00:05:01.760>the <00:05:01.900>frequency, <00:05:02.039>the <00:05:02.180>probability <00:05:02.320>of <00:05:02.460>occurrence <00:05:02.599>of

103
00:05:04.300 --> 00:05:07.500
<00:05:04.300>all <00:05:04.455>kinds <00:05:04.611>of <00:05:04.766>events <00:05:04.922>that <00:05:05.077>a <00:05:05.233>self-driving <00:05:05.388>car <00:05:05.544>might <00:05:05.700>encounter

104
00:05:07.500 --> 00:05:08.500
<00:05:07.500>on <00:05:07.628>the <00:05:07.757>road. <00:05:07.885>It <00:05:08.014>looks <00:05:08.142>something <00:05:08.271>like <00:05:08.400>this.

105
00:05:10.200 --> 00:05:11.500
<00:05:10.200>So <00:05:10.492>on <00:05:10.784>the <00:05:11.077>one <00:05:11.370>end.

106
00:05:12.400 --> 00:05:13.700
<00:05:12.400>We <00:05:12.789>have <00:05:13.180>all <00:05:13.570>these.

107
00:05:14.600 --> 00:05:17.300
<00:05:14.600>Events <00:05:14.727>that <00:05:14.854>we <00:05:14.981>don't <00:05:15.109>care <00:05:15.236>about <00:05:15.363>and <00:05:15.490>that <00:05:15.618>makes <00:05:15.745>up <00:05:15.872>the <00:05:16.000>vast

108
00:05:17.300 --> 00:05:20.700
<00:05:17.300>majority <00:05:17.427>of <00:05:17.554>events <00:05:17.681>that <00:05:17.809>we <00:05:17.936>see <00:05:18.063>on <00:05:18.190>the <00:05:18.318>road <00:05:18.445>because <00:05:18.572>traffic <00:05:18.700>conditions

109
00:05:20.700 --> 00:05:21.500
<00:05:20.700>are <00:05:21.060>just <00:05:21.420>normal.

110
00:05:22.500 --> 00:05:25.200
<00:05:22.500>We <00:05:22.607>already <00:05:22.715>know <00:05:22.823>how <00:05:22.930>to <00:05:23.038>handle <00:05:23.146>these <00:05:23.253>situations <00:05:23.361>very <00:05:23.469>well. <00:05:23.576>We <00:05:23.684>don't <00:05:23.792>need <00:05:23.900>to

111
00:05:25.200 --> 00:05:28.300
<00:05:25.200>collect <00:05:25.316>more <00:05:25.433>data <00:05:25.550>on <00:05:25.666>those. <00:05:25.783>But <00:05:25.900>what <00:05:26.016>we <00:05:26.133>really <00:05:26.250>do <00:05:26.366>care <00:05:26.483>about <00:05:26.599>is

112
00:05:28.300 --> 00:05:31.400
<00:05:28.300>the <00:05:28.455>tail <00:05:28.611>end <00:05:28.766>of <00:05:28.922>the <00:05:29.077>distribution, <00:05:29.233>so <00:05:29.388>here <00:05:29.544>would <00:05:29.700>be

113
00:05:31.400 --> 00:05:31.900
<00:05:31.400>to <00:05:31.625>your <00:05:31.849>right.

114
00:05:32.700 --> 00:05:36.400
<00:05:32.700>This <00:05:32.900>is <00:05:33.099>where <00:05:33.300>we <00:05:33.500>have <00:05:33.700>these <00:05:33.900>unpredictable <00:05:34.099>rare,

115
00:05:35.400 --> 00:05:38.800
<00:05:35.400>but <00:05:35.633>often <00:05:35.866>safety <00:05:36.099>critical <00:05:36.333>scenarios <00:05:36.566>that <00:05:36.799>our

116
00:05:38.800 --> 00:05:41.300
<00:05:38.800>model <00:05:38.927>tends <00:05:39.054>to <00:05:39.181>struggle <00:05:39.309>with <00:05:39.436>because <00:05:39.563>we <00:05:39.690>don't <00:05:39.818>have <00:05:39.945>enough <00:05:40.072>training <00:05:40.200>data

117
00:05:41.300 --> 00:05:41.600
<00:05:41.300>on <00:05:41.570>those.

118
00:05:43.200 --> 00:05:46.200
<00:05:43.200>And <00:05:43.327>this <00:05:43.454>is <00:05:43.581>known <00:05:43.709>as <00:05:43.836>the <00:05:43.963>long <00:05:44.090>tail <00:05:44.218>problem <00:05:44.345>in <00:05:44.472>autonomous <00:05:44.599>driving.

119
00:05:46.900 --> 00:05:49.600
<00:05:46.900>And <00:05:47.075>here <00:05:47.250>are <00:05:47.424>some <00:05:47.599>concrete <00:05:47.775>examples <00:05:47.950>of <00:05:48.125>long <00:05:48.299>tail

120
00:05:49.600 --> 00:05:51.500
<00:05:49.600>scenarios. <00:05:49.880>For <00:05:50.160>example, <00:05:50.440>we <00:05:50.720>could <00:05:51.000>have

121
00:05:52.600 --> 00:05:55.500
<00:05:52.600>Noisy <00:05:52.800>sensor <00:05:53.000>data, <00:05:53.200>Maybe <00:05:53.400>due <00:05:53.600>to <00:05:53.800>adverse <00:05:54.000>where

122
00:05:55.500 --> 00:05:58.500
<00:05:55.500>the <00:05:55.733>conditions <00:05:55.966>we <00:05:56.200>could <00:05:56.433>have <00:05:56.666>a <00:05:56.900>distribution

123
00:05:58.500 --> 00:06:01.300
<00:05:58.500>shift. <00:05:58.655>For <00:05:58.811>example, <00:05:58.966>if <00:05:59.122>there's <00:05:59.277>an <00:05:59.433>unusual <00:05:59.588>event <00:05:59.744>going <00:05:59.900>on

124
00:06:01.300 --> 00:06:04.300
<00:06:01.300>on <00:06:01.455>the <00:06:01.611>road <00:06:01.766>that <00:06:01.922>causes <00:06:02.077>your <00:06:02.233>distribution <00:06:02.388>of <00:06:02.544>classes <00:06:02.700>to

125
00:06:04.300 --> 00:06:07.500
<00:06:04.300>be <00:06:04.427>different <00:06:04.554>from <00:06:04.681>your <00:06:04.809>training <00:06:04.936>set <00:06:05.063>or <00:06:05.190>it <00:06:05.318>could <00:06:05.445>be <00:06:05.572>that <00:06:05.700>you

126
00:06:07.500 --> 00:06:10.300
<00:06:07.500>train <00:06:07.600>a <00:06:07.700>model <00:06:07.800>in <00:06:07.900>one <00:06:08.000>country <00:06:08.100>and <00:06:08.200>then <00:06:08.300>deploy <00:06:08.400>it <00:06:08.500>on <00:06:08.600>the <00:06:08.700>road <00:06:08.800>in <00:06:08.900>a

127
00:06:10.300 --> 00:06:10.600
<00:06:10.300>different <00:06:10.570>country.

128
00:06:11.900 --> 00:06:15.000
<00:06:11.900>You <00:06:12.099>could <00:06:12.299>sometimes <00:06:12.500>have <00:06:12.700>rare <00:06:12.900>classes <00:06:13.099>unknown <00:06:13.299>objects

129
00:06:14.100 --> 00:06:17.800
<00:06:14.100>on <00:06:14.300>the <00:06:14.500>road <00:06:14.700>like <00:06:14.900>elephants. <00:06:15.100>For <00:06:15.300>example <00:06:15.500>that

130
00:06:17.800 --> 00:06:19.100
<00:06:17.800>a <00:06:17.995>car <00:06:18.190>doesn't <00:06:18.385>know <00:06:18.580>how <00:06:18.775>to <00:06:18.970>recognize.

131
00:06:20.200 --> 00:06:23.600
<00:06:20.200>And <00:06:20.433>then <00:06:20.666>you <00:06:20.900>also <00:06:21.133>have <00:06:21.366>familiar <00:06:21.599>objects,

132
00:06:23.600 --> 00:06:26.400
<00:06:23.600>but <00:06:23.833>occurring <00:06:24.066>in <00:06:24.300>unfamiliar <00:06:24.533>situations <00:06:24.766>that <00:06:25.000>might

133
00:06:26.400 --> 00:06:30.100
<00:06:26.400>also <00:06:26.555>confuse <00:06:26.711>the <00:06:26.866>car. <00:06:27.022>So <00:06:27.177>these <00:06:27.333>are <00:06:27.488>just <00:06:27.644>a <00:06:27.799>handful

134
00:06:29.100 --> 00:06:31.800
<00:06:29.100>of <00:06:29.450>different <00:06:29.800>scenarios <00:06:30.150>that <00:06:30.500>could

135
00:06:32.700 --> 00:06:35.200
<00:06:32.700>Be <00:06:32.875>unknown <00:06:33.050>unknowns <00:06:33.224>and <00:06:33.400>could <00:06:33.575>lead <00:06:33.750>to <00:06:33.925>confusion <00:06:34.099>in

136
00:06:35.200 --> 00:06:38.700
<00:06:35.200>the <00:06:35.340>model <00:06:35.479>closer <00:06:35.620>to <00:06:35.760>make <00:06:35.900>a <00:06:36.039>wrong <00:06:36.180>prediction <00:06:36.320>which <00:06:36.460>could <00:06:36.599>sometimes

137
00:06:38.700 --> 00:06:41.500
<00:06:38.700>have <00:06:38.979>disastrous <00:06:39.260>consequences <00:06:39.539>which <00:06:39.820>we <00:06:40.099>want

138
00:06:41.500 --> 00:06:41.800
<00:06:41.500>to <00:06:41.770>avoid.

139
00:06:42.900 --> 00:06:44.600
<00:06:42.900>So <00:06:43.133>how <00:06:43.366>do <00:06:43.599>we <00:06:43.833>recognize <00:06:44.066>these <00:06:44.299>situations?

140
00:06:45.800 --> 00:06:48.100
<00:06:45.800>There <00:06:45.975>are <00:06:46.150>some <00:06:46.325>methods <00:06:46.500>in <00:06:46.675>the <00:06:46.850>literature <00:06:47.025>that <00:06:47.200>are

141
00:06:48.100 --> 00:06:49.900
<00:06:48.100>pretty <00:06:48.255>well <00:06:48.411>known. <00:06:48.566>I'll <00:06:48.722>give <00:06:48.877>you <00:06:49.033>a <00:06:49.188>couple <00:06:49.344>examples <00:06:49.500>here.

142
00:06:51.100 --> 00:06:54.500
<00:06:51.100>We <00:06:51.275>have <00:06:51.450>the <00:06:51.625>uncertainty <00:06:51.800>based <00:06:51.975>methods <00:06:52.150>which <00:06:52.325>includes <00:06:52.500>the

143
00:06:54.500 --> 00:06:57.300
<00:06:54.500>well <00:06:54.780>known <00:06:55.060>Asian <00:06:55.340>inference <00:06:55.620>methods <00:06:55.900>such

144
00:06:57.300 --> 00:06:58.300
<00:06:57.300>as <00:06:57.600>empty <00:06:57.900>drop <00:06:58.200>belt.

145
00:06:59.300 --> 00:07:02.400
<00:06:59.300>For <00:06:59.455>example <00:06:59.611>also <00:06:59.766>have <00:06:59.922>some <00:07:00.077>temporal <00:07:00.233>methods <00:07:00.388>that <00:07:00.544>rely <00:07:00.700>on

146
00:07:02.400 --> 00:07:05.300
<00:07:02.400>checking <00:07:02.599>for <00:07:02.799>consistency <00:07:03.000>across <00:07:03.200>different <00:07:03.400>frames <00:07:03.599>over <00:07:03.799>time.

147
00:07:06.100 --> 00:07:09.000
<00:07:06.100>We <00:07:06.240>have <00:07:06.380>a <00:07:06.520>normally <00:07:06.660>detection <00:07:06.800>and <00:07:06.940>Novelty <00:07:07.080>detection <00:07:07.220>which <00:07:07.360>you <00:07:07.500>might

148
00:07:09.300 --> 00:07:12.600
<00:07:09.300>all <00:07:09.427>be <00:07:09.554>familiar <00:07:09.681>with <00:07:09.809>from <00:07:09.936>different <00:07:10.063>domains <00:07:10.190>already. <00:07:10.318>So <00:07:10.445>that <00:07:10.572>is <00:07:10.700>detecting

149
00:07:12.600 --> 00:07:15.500
<00:07:12.600>samples <00:07:12.800>that <00:07:13.000>are <00:07:13.200>out <00:07:13.400>of <00:07:13.600>distribution. <00:07:13.800>And <00:07:14.000>finally,

150
00:07:15.500 --> 00:07:18.800
<00:07:15.500>there <00:07:15.675>are <00:07:15.850>Ensemble <00:07:16.025>methods <00:07:16.200>which <00:07:16.375>rely <00:07:16.550>on <00:07:16.725>using <00:07:16.900>ensembles

151
00:07:18.800 --> 00:07:22.100
<00:07:18.800>of <00:07:19.033>neural <00:07:19.266>networks <00:07:19.500>instead <00:07:19.733>of <00:07:19.966>patient <00:07:20.200>inference

152
00:07:21.100 --> 00:07:23.300
<00:07:21.100>to <00:07:21.566>detect <00:07:22.033>on <00:07:22.500>certainty.

153
00:07:24.700 --> 00:07:27.600
<00:07:24.700>So <00:07:24.840>these <00:07:24.979>all <00:07:25.120>have <00:07:25.260>different <00:07:25.400>strength <00:07:25.539>and <00:07:25.680>weaknesses <00:07:25.820>and <00:07:25.960>at <00:07:26.099>cariad

154
00:07:27.600 --> 00:07:30.200
<00:07:27.600>we <00:07:27.775>experimenting <00:07:27.950>with <00:07:28.125>different <00:07:28.300>combinations <00:07:28.475>of <00:07:28.650>them <00:07:28.825>to <00:07:29.000>see

155
00:07:30.200 --> 00:07:31.700
<00:07:30.200>which <00:07:30.469>one <00:07:30.740>is <00:07:31.010>the <00:07:31.280>optimal <00:07:31.550>solution.

156
00:07:33.200 --> 00:07:36.800
<00:07:33.200>So <00:07:33.355>with <00:07:33.511>all <00:07:33.666>of <00:07:33.822>that <00:07:33.977>in <00:07:34.133>mind, <00:07:34.288>we <00:07:34.444>developed <00:07:34.599>a

157
00:07:36.800 --> 00:07:39.400
<00:07:36.800>piece <00:07:37.000>of <00:07:37.200>software <00:07:37.400>called <00:07:37.600>instinct <00:07:37.800>which <00:07:38.000>stands <00:07:38.200>for

158
00:07:39.400 --> 00:07:41.400
<00:07:39.400>institute <00:07:39.866>intelligent <00:07:40.333>data <00:07:40.799>collector.

159
00:07:42.100 --> 00:07:45.600
<00:07:42.100>So <00:07:42.227>there's <00:07:42.354>a <00:07:42.481>software <00:07:42.609>that's <00:07:42.736>based <00:07:42.863>on <00:07:42.990>deep <00:07:43.118>neural <00:07:43.245>networks <00:07:43.372>and <00:07:43.500>runs

160
00:07:45.600 --> 00:07:48.200
<00:07:45.600>inside <00:07:45.775>the <00:07:45.950>protected <00:07:46.125>area <00:07:46.300>blade <00:07:46.475>in <00:07:46.650>every <00:07:46.825>test <00:07:47.000>vehicle.

161
00:07:49.500 --> 00:07:52.300
<00:07:49.500>So <00:07:49.640>it <00:07:49.780>is <00:07:49.920>able <00:07:50.060>to <00:07:50.200>filter <00:07:50.340>out <00:07:50.480>the <00:07:50.620>junk <00:07:50.760>as <00:07:50.900>I

162
00:07:52.300 --> 00:07:55.400
<00:07:52.300>mentioned <00:07:52.455>earlier <00:07:52.611>and <00:07:52.766>only <00:07:52.922>upload <00:07:53.077>the <00:07:53.233>most <00:07:53.388>relevant <00:07:53.544>and <00:07:53.700>useful

163
00:07:55.400 --> 00:07:58.800
<00:07:55.400>information <00:07:55.575>to <00:07:55.750>the <00:07:55.924>cloud <00:07:56.099>for <00:07:56.275>further <00:07:56.450>analysis, <00:07:56.625>thereby <00:07:56.799>saving

164
00:07:58.800 --> 00:07:59.200
<00:07:58.800>on <00:07:59.159>costs.

165
00:08:00.500 --> 00:08:03.400
<00:08:00.500>So <00:08:00.616>we <00:08:00.733>are <00:08:00.850>already <00:08:00.966>in <00:08:01.083>the <00:08:01.200>process <00:08:01.316>of <00:08:01.433>scaling <00:08:01.550>up <00:08:01.666>our <00:08:01.783>test <00:08:01.900>lead

166
00:08:03.400 --> 00:08:06.100
<00:08:03.400>on <00:08:03.527>the <00:08:03.654>road <00:08:03.781>and <00:08:03.909>this <00:08:04.036>is <00:08:04.163>going <00:08:04.290>to <00:08:04.418>generate <00:08:04.545>a <00:08:04.672>very <00:08:04.799>large

167
00:08:06.100 --> 00:08:06.600
<00:08:06.100>amount <00:08:06.325>of <00:08:06.550>data.

168
00:08:07.700 --> 00:08:10.200
<00:08:07.700>For <00:08:07.855>every <00:08:08.011>single <00:08:08.166>car <00:08:08.322>just <00:08:08.477>the <00:08:08.633>cameras <00:08:08.788>alone <00:08:08.944>would <00:08:09.099>be

169
00:08:10.200 --> 00:08:14.300
<00:08:10.200>capturing <00:08:10.327>data <00:08:10.454>at <00:08:10.581>30 <00:08:10.709>frames <00:08:10.836>per <00:08:10.963>second. <00:08:11.090>So <00:08:11.218>that <00:08:11.345>means <00:08:11.472>108,000 <00:08:11.599>frames

170
00:08:14.300 --> 00:08:17.200
<00:08:14.300>per <00:08:14.427>hour <00:08:14.554>per <00:08:14.681>car, <00:08:14.809>which <00:08:14.936>is <00:08:15.063>a <00:08:15.190>lot <00:08:15.318>but <00:08:15.445>with <00:08:15.572>Instinct <00:08:15.700>we

171
00:08:17.200 --> 00:08:20.000
<00:08:17.200>aim <00:08:17.355>to <00:08:17.511>cut <00:08:17.666>that <00:08:17.822>just <00:08:17.977>10 <00:08:18.133>hours <00:08:18.288>10 <00:08:18.444>frames <00:08:18.599>for

172
00:08:20.500 --> 00:08:22.400
<00:08:20.500>our <00:08:20.700>car <00:08:20.900>without <00:08:21.100>losing <00:08:21.300>the <00:08:21.500>most <00:08:21.700>valuable <00:08:21.900>information.

173
00:08:24.500 --> 00:08:27.500
<00:08:24.500>Let <00:08:24.627>me <00:08:24.754>show <00:08:24.881>you <00:08:25.009>a <00:08:25.136>little <00:08:25.263>bit <00:08:25.390>more <00:08:25.518>about <00:08:25.645>one <00:08:25.772>particular <00:08:25.900>use

174
00:08:27.500 --> 00:08:30.500
<00:08:27.500>case <00:08:27.655>of <00:08:27.811>instinct. <00:08:27.966>Okay, <00:08:28.122>so <00:08:28.277>here <00:08:28.433>is <00:08:28.588>a <00:08:28.744>neural <00:08:28.900>network

175
00:08:30.500 --> 00:08:31.500
<00:08:30.500>that <00:08:30.800>detects <00:08:31.100>traffic <00:08:31.400>lights.

176
00:08:32.300 --> 00:08:35.900
<00:08:32.299>It <00:08:32.439>looks <00:08:32.579>at <00:08:32.719>the <00:08:32.859>input <00:08:33.000>image <00:08:33.140>and <00:08:33.280>then <00:08:33.419>it <00:08:33.559>generates <00:08:33.699>a

177
00:08:35.900 --> 00:08:38.300
<00:08:35.900>score <00:08:36.074>between <00:08:36.250>zero <00:08:36.424>and <00:08:36.600>one <00:08:36.775>for <00:08:36.949>every <00:08:37.125>pixel <00:08:37.299>all

178
00:08:38.300 --> 00:08:41.900
<00:08:38.299>this <00:08:38.474>image, <00:08:38.650>which <00:08:38.824>roughly <00:08:39.000>corresponds <00:08:39.174>the <00:08:39.349>probability <00:08:39.525>that <00:08:39.699>this

179
00:08:41.900 --> 00:08:44.500
<00:08:41.900>model <00:08:42.055>thinks <00:08:42.211>the <00:08:42.366>pixel <00:08:42.522>belongs <00:08:42.677>to <00:08:42.833>traffic <00:08:42.988>light. <00:08:43.144>So <00:08:43.299>this

180
00:08:44.500 --> 00:08:47.600
<00:08:44.500>is <00:08:44.700>so <00:08:44.900>far <00:08:45.100>what <00:08:45.300>any <00:08:45.500>vanilla <00:08:45.700>second <00:08:45.900>is

181
00:08:47.600 --> 00:08:48.700
<00:08:47.600>meant <00:08:47.798>to <00:08:47.996>segmentation <00:08:48.194>model <00:08:48.392>can <00:08:48.590>do

182
00:08:49.800 --> 00:08:51.600
<00:08:49.800>and <00:08:50.000>here's <00:08:50.199>where <00:08:50.400>the <00:08:50.599>Instant <00:08:50.800>Magic <00:08:51.000>comes <00:08:51.199>in.

183
00:08:52.500 --> 00:08:54.700
<00:08:52.500>So <00:08:52.675>Instinct <00:08:52.850>takes <00:08:53.025>the <00:08:53.200>output <00:08:53.375>of <00:08:53.550>the <00:08:53.725>neural <00:08:53.900>network.

184
00:08:55.400 --> 00:08:58.500
<00:08:55.400>And <00:08:55.555>then <00:08:55.711>calculate <00:08:55.866>an <00:08:56.022>uncertainty <00:08:56.177>score <00:08:56.333>for <00:08:56.488>every <00:08:56.644>single <00:08:56.800>component

185
00:08:58.500 --> 00:08:59.800
<00:08:58.500>detected <00:08:58.792>by <00:08:59.085>the <00:08:59.377>neural <00:08:59.670>network.

186
00:09:01.100 --> 00:09:04.500
<00:09:01.100>So <00:09:01.216>here <00:09:01.333>you <00:09:01.450>can <00:09:01.566>see <00:09:01.683>again <00:09:01.800>some <00:09:01.916>examples. <00:09:02.033>So <00:09:02.150>the <00:09:02.266>green <00:09:02.383>components <00:09:02.500>have

187
00:09:04.500 --> 00:09:06.200
<00:09:04.500>low <00:09:04.733>uncertainty. <00:09:04.966>So <00:09:05.200>these <00:09:05.433>tend <00:09:05.666>to <00:09:05.900>be

188
00:09:07.200 --> 00:09:10.100
<00:09:07.200>big <00:09:07.375>clear <00:09:07.550>traffic <00:09:07.725>lights. <00:09:07.900>They <00:09:08.075>are <00:09:08.250>unambiguous <00:09:08.425>and <00:09:08.600>the

189
00:09:10.100 --> 00:09:13.300
<00:09:10.100>model <00:09:10.240>that's <00:09:10.380>very <00:09:10.520>well <00:09:10.660>on <00:09:10.800>these <00:09:10.940>and <00:09:11.080>the <00:09:11.220>red <00:09:11.360>components <00:09:11.500>are

190
00:09:13.300 --> 00:09:17.100
<00:09:13.300>high <00:09:13.427>on <00:09:13.554>certainties. <00:09:13.681>So <00:09:13.809>these <00:09:13.936>tend <00:09:14.063>to <00:09:14.190>be <00:09:14.318>things <00:09:14.445>that <00:09:14.572>might <00:09:14.699>confuse

191
00:09:16.100 --> 00:09:19.200
<00:09:16.100>the <00:09:16.275>model <00:09:16.450>a <00:09:16.625>little <00:09:16.800>bit <00:09:16.975>such <00:09:17.150>as <00:09:17.325>tail <00:09:17.500>lights

192
00:09:19.200 --> 00:09:21.800
<00:09:19.200>mirrors <00:09:19.355>or <00:09:19.511>traffic <00:09:19.666>lights <00:09:19.822>that <00:09:19.977>are <00:09:20.133>small <00:09:20.288>how <00:09:20.444>to <00:09:20.600>detect

193
00:09:23.100 --> 00:09:26.200
<00:09:23.100>and <00:09:23.333>then <00:09:23.566>after <00:09:23.800>calculating <00:09:24.033>these <00:09:24.266>component-wise <00:09:24.500>scores

194
00:09:26.200 --> 00:09:29.000
<00:09:26.200>Instinct <00:09:26.433>generates <00:09:26.666>and <00:09:26.900>overall <00:09:27.133>score <00:09:27.366>for <00:09:27.600>the

195
00:09:29.000 --> 00:09:29.300
image.

196
00:09:30.100 --> 00:09:33.600
<00:09:30.100>So <00:09:30.216>if <00:09:30.333>an <00:09:30.450>image <00:09:30.566>has <00:09:30.683>a <00:09:30.800>lot <00:09:30.916>of <00:09:31.033>rate <00:09:31.150>components <00:09:31.266>and <00:09:31.383>it's <00:09:31.500>total

197
00:09:33.600 --> 00:09:36.500
<00:09:33.600>score <00:09:33.775>exceeds <00:09:33.950>some <00:09:34.125>threshold, <00:09:34.300>then <00:09:34.475>we <00:09:34.650>say <00:09:34.825>this <00:09:35.000>image

198
00:09:36.500 --> 00:09:39.500
<00:09:36.500>is <00:09:36.655>triggered <00:09:36.811>and <00:09:36.966>of <00:09:37.122>course <00:09:37.277>being <00:09:37.433>uploaded <00:09:37.588>for <00:09:37.744>further <00:09:37.900>analysis.

199
00:09:42.600 --> 00:09:45.900
<00:09:42.600>Here <00:09:42.755>I'm <00:09:42.911>showing <00:09:43.066>you <00:09:43.222>a <00:09:43.377>video <00:09:43.533>of <00:09:43.688>instinct <00:09:43.844>on <00:09:44.000>in

200
00:09:45.900 --> 00:09:48.900
<00:09:45.900>action <00:09:46.040>on <00:09:46.180>some <00:09:46.319>footage <00:09:46.459>that <00:09:46.600>we <00:09:46.740>took <00:09:46.880>in <00:09:47.020>English <00:09:47.160>that <00:09:47.300>so

201
00:09:48.900 --> 00:09:51.500
<00:09:48.900>the <00:09:49.100>middle <00:09:49.300>panel <00:09:49.500>you'll <00:09:49.699>see <00:09:49.900>the <00:09:50.100>raw <00:09:50.300>output

202
00:09:51.500 --> 00:09:53.600
<00:09:51.500>from <00:09:51.780>the <00:09:52.060>model <00:09:52.340>with <00:09:52.620>its <00:09:52.900>activations.

203
00:09:54.200 --> 00:09:57.200
<00:09:54.200>And <00:09:54.327>on <00:09:54.454>the <00:09:54.581>right <00:09:54.709>hand <00:09:54.836>side. <00:09:54.963>This <00:09:55.090>is <00:09:55.218>the <00:09:55.345>output <00:09:55.472>from <00:09:55.600>instinct.

204
00:09:58.300 --> 00:10:00.000
<00:09:58.300>Every <00:09:58.474>component <00:09:58.650>has <00:09:58.824>its <00:09:59.000>score <00:09:59.175>on <00:09:59.349>top <00:09:59.525>of <00:09:59.699>it.

205
00:10:00.800 --> 00:10:03.700
<00:10:00.800>And <00:10:01.079>also <00:10:01.359>the <00:10:01.640>color <00:10:01.920>indicates <00:10:02.199>this

206
00:10:03.700 --> 00:10:06.200
<00:10:03.700>uncertainty <00:10:03.875>level <00:10:04.050>with <00:10:04.225>red <00:10:04.400>being <00:10:04.575>high <00:10:04.750>on <00:10:04.925>certainty <00:10:05.100>and

207
00:10:06.200 --> 00:10:07.800
<00:10:06.200>moving <00:10:06.900>low <00:10:07.600>uncertainty.

208
00:10:08.500 --> 00:10:11.500
<00:10:08.500>And <00:10:08.627>you <00:10:08.754>might <00:10:08.881>notice <00:10:09.009>that <00:10:09.136>sometimes <00:10:09.263>in <00:10:09.390>the <00:10:09.518>header <00:10:09.645>the <00:10:09.772>header <00:10:09.900>turns

209
00:10:11.500 --> 00:10:12.000
<00:10:11.500>red <00:10:11.950>when?

210
00:10:13.200 --> 00:10:14.500
<00:10:13.200>Sorry, <00:10:13.367>the <00:10:13.534>video <00:10:13.701>is <00:10:13.868>actually <00:10:14.035>a <00:10:14.202>little <00:10:14.370>bit.

211
00:10:15.400 --> 00:10:18.100
<00:10:15.400>It's <00:10:15.574>not <00:10:15.750>playing <00:10:15.925>very <00:10:16.100>smoothly <00:10:16.275>but <00:10:16.449>you <00:10:16.625>might <00:10:16.800>have

212
00:10:18.100 --> 00:10:20.800
<00:10:18.100>noticed <00:10:18.227>the <00:10:18.354>head <00:10:18.481>is <00:10:18.609>sometimes <00:10:18.736>turns <00:10:18.863>red <00:10:18.990>if <00:10:19.118>the <00:10:19.245>image <00:10:19.372>is <00:10:19.500>triggered.

213
00:10:23.500 --> 00:10:26.500
<00:10:23.500>Okay, <00:10:23.607>and <00:10:23.715>now <00:10:23.823>let <00:10:23.930>me <00:10:24.038>show <00:10:24.146>you <00:10:24.253>big <00:10:24.361>loop <00:10:24.469>in <00:10:24.576>action <00:10:24.684>on <00:10:24.792>some <00:10:24.900>footage

214
00:10:26.500 --> 00:10:29.100
<00:10:26.500>that <00:10:26.607>we <00:10:26.715>took <00:10:26.823>in, <00:10:26.930>California. <00:10:27.038>So, <00:10:27.146>you <00:10:27.253>know, <00:10:27.361>I'm <00:10:27.469>not <00:10:27.576>just <00:10:27.684>talking <00:10:27.792>Theory <00:10:27.900>here.

215
00:10:29.900 --> 00:10:32.300
<00:10:29.900>And <00:10:30.074>this <00:10:30.250>is <00:10:30.425>the <00:10:30.600>same <00:10:30.775>process <00:10:30.949>that <00:10:31.125>will <00:10:31.300>take

216
00:10:32.300 --> 00:10:35.500
<00:10:32.300>place <00:10:32.407>in <00:10:32.515>all <00:10:32.623>of <00:10:32.730>our <00:10:32.838>test <00:10:32.946>Vehicles <00:10:33.053>when <00:10:33.161>they're <00:10:33.269>on <00:10:33.376>the <00:10:33.484>road. <00:10:33.592>So <00:10:33.699>first

217
00:10:35.500 --> 00:10:38.300
<00:10:35.500>the <00:10:35.733>vehicle <00:10:35.966>drives <00:10:36.200>collects <00:10:36.433>data, <00:10:36.666>the <00:10:36.900>data

218
00:10:38.300 --> 00:10:39.800
<00:10:38.300>gets <00:10:38.492>uploaded <00:10:38.685>to <00:10:38.878>the <00:10:39.071>cloud <00:10:39.264>some <00:10:39.457>of <00:10:39.650>it.

219
00:10:40.500 --> 00:10:43.200
<00:10:40.500>And <00:10:40.675>later <00:10:40.850>on <00:10:41.025>the <00:10:41.200>status <00:10:41.375>labeled <00:10:41.550>and <00:10:41.725>used <00:10:41.900>to

220
00:10:43.200 --> 00:10:46.500
<00:10:43.200>retrain <00:10:43.355>on <00:10:43.511>your <00:10:43.666>network <00:10:43.822>and <00:10:43.977>then <00:10:44.133>it's <00:10:44.288>deployed <00:10:44.444>again <00:10:44.600>into

221
00:10:46.500 --> 00:10:49.200
<00:10:46.500>the <00:10:46.616>car <00:10:46.733>and <00:10:46.850>the <00:10:46.966>cycle <00:10:47.083>starts <00:10:47.200>again, <00:10:47.316>and <00:10:47.433>this <00:10:47.550>is <00:10:47.666>actually <00:10:47.783>where <00:10:47.900>big

222
00:10:49.200 --> 00:10:50.000
<00:10:49.200>loop <00:10:49.440>gets <00:10:49.680>its <00:10:49.920>name.

223
00:10:51.200 --> 00:10:54.200
<00:10:51.200>So <00:10:51.300>let's <00:10:51.400>start <00:10:51.500>with <00:10:51.600>a <00:10:51.700>drive <00:10:51.800>and <00:10:51.900>upload <00:10:52.000>step <00:10:52.100>and <00:10:52.200>this <00:10:52.300>is <00:10:52.400>normally <00:10:52.500>where <00:10:52.600>I

224
00:10:54.200 --> 00:10:57.200
<00:10:54.200>would <00:10:54.327>actually <00:10:54.454>bring <00:10:54.581>in <00:10:54.709>a <00:10:54.836>live <00:10:54.963>demo <00:10:55.090>with <00:10:55.218>an <00:10:55.345>actual <00:10:55.472>car, <00:10:55.600>but

225
00:10:57.200 --> 00:11:00.200
<00:10:57.200>obviously <00:10:57.375>that's <00:10:57.550>gonna <00:10:57.725>be <00:10:57.900>difficult <00:10:58.075>to <00:10:58.250>do <00:10:58.425>here. <00:10:58.600>So

226
00:11:00.200 --> 00:11:03.600
<00:11:00.200>I'll <00:11:00.327>just <00:11:00.454>show <00:11:00.581>you <00:11:00.709>some <00:11:00.836>footage <00:11:00.963>that <00:11:01.090>I <00:11:01.218>we <00:11:01.345>did <00:11:01.472>in <00:11:01.600>a

227
00:11:03.600 --> 00:11:05.400
<00:11:03.600>live <00:11:03.833>session <00:11:04.066>with <00:11:04.300>our <00:11:04.533>colleagues <00:11:04.766>in, <00:11:05.000>California.

228
00:11:07.800 --> 00:11:08.600
<00:11:07.800>Okay, <00:11:08.160>so <00:11:08.520>here.

229
00:11:09.800 --> 00:11:12.700
<00:11:09.800>This <00:11:09.974>was <00:11:10.150>a <00:11:10.324>session <00:11:10.500>that <00:11:10.675>we <00:11:10.849>did <00:11:11.025>with <00:11:11.199>our

230
00:11:12.700 --> 00:11:15.300
<00:11:12.700>driver <00:11:12.855>was <00:11:13.011>driving <00:11:13.166>in <00:11:13.322>California <00:11:13.477>and <00:11:13.633>we <00:11:13.788>were <00:11:13.944>in <00:11:14.100>Germany

231
00:11:15.300 --> 00:11:18.400
<00:11:15.300>in <00:11:15.439>English <00:11:15.579>Dart <00:11:15.719>observing <00:11:15.859>these <00:11:16.000>images <00:11:16.140>in <00:11:16.280>real <00:11:16.420>time. <00:11:16.560>So <00:11:16.699>let

232
00:11:18.400 --> 00:11:18.900
<00:11:18.400>me <00:11:18.490>try <00:11:18.579>to <00:11:18.670>play <00:11:18.760>the <00:11:18.850>video.

233
00:11:20.400 --> 00:11:24.000
<00:11:20.400>So <00:11:20.516>you <00:11:20.633>can <00:11:20.750>see <00:11:20.866>that <00:11:20.983>we <00:11:21.100>were <00:11:21.216>connected <00:11:21.333>to <00:11:21.449>the <00:11:21.566>driver <00:11:21.683>via <00:11:21.800>a

234
00:11:23.100 --> 00:11:26.200
<00:11:23.100>video <00:11:23.275>call <00:11:23.450>in <00:11:23.625>the <00:11:23.800>bottom <00:11:23.975>right <00:11:24.150>corner. <00:11:24.325>And <00:11:24.500>this

235
00:11:26.200 --> 00:11:29.300
<00:11:26.200>is <00:11:26.300>more <00:11:26.400>or <00:11:26.500>less <00:11:26.600>in <00:11:26.700>sync <00:11:26.800>with <00:11:26.900>the <00:11:27.000>images <00:11:27.100>that <00:11:27.200>you <00:11:27.300>see <00:11:27.400>in <00:11:27.500>the <00:11:27.600>notebook.

236
00:11:29.300 --> 00:11:32.600
<00:11:29.300>Again. <00:11:29.500>The <00:11:29.699>middle <00:11:29.900>is <00:11:30.099>the <00:11:30.300>neural <00:11:30.500>networks <00:11:30.699>traditions

237
00:11:32.600 --> 00:11:35.900
<00:11:32.600>on <00:11:32.727>traffic <00:11:32.854>lights <00:11:32.981>and <00:11:33.109>the <00:11:33.236>right <00:11:33.363>hand <00:11:33.490>side. <00:11:33.618>You'll <00:11:33.745>see <00:11:33.872>instincts <00:11:34.000>predictions.

238
00:11:35.900 --> 00:11:38.300
<00:11:35.900>So <00:11:36.180>these <00:11:36.459>are <00:11:36.740>roughly <00:11:37.020>in <00:11:37.300>sync

239
00:11:38.300 --> 00:11:40.900
<00:11:38.300>which <00:11:38.455>means <00:11:38.611>instinct <00:11:38.766>is <00:11:38.922>doing <00:11:39.077>its <00:11:39.233>predictions <00:11:39.388>in <00:11:39.544>real <00:11:39.699>time.

240
00:11:41.900 --> 00:11:44.300
<00:11:41.900>And <00:11:42.100>you <00:11:42.300>can <00:11:42.500>also <00:11:42.699>see <00:11:42.900>that <00:11:43.100>this <00:11:43.300>model

241
00:11:44.300 --> 00:11:47.800
<00:11:44.300>was <00:11:44.474>never <00:11:44.650>training <00:11:44.824>California. <00:11:45.000>So <00:11:45.175>it's <00:11:45.349>the <00:11:45.525>distribution <00:11:45.699>shift

242
00:11:47.800 --> 00:11:50.700
<00:11:47.800>is <00:11:47.927>causing <00:11:48.054>the <00:11:48.181>model <00:11:48.309>to <00:11:48.436>struggle <00:11:48.563>a <00:11:48.690>little <00:11:48.818>bit <00:11:48.945>with <00:11:49.072>certain <00:11:49.199>components

243
00:11:50.700 --> 00:11:53.200
<00:11:50.700>that <00:11:50.855>is <00:11:51.011>finding <00:11:51.166>so <00:11:51.322>you <00:11:51.477>can <00:11:51.633>notice <00:11:51.788>some <00:11:51.944>of <00:11:52.100>these

244
00:11:53.200 --> 00:11:54.200
<00:11:53.200>flaws <00:11:53.425>if <00:11:53.650>you <00:11:53.875>look <00:11:54.100>carefully

245
00:11:55.400 --> 00:11:58.500
<00:11:55.400>An <00:11:55.540>instinct <00:11:55.680>is <00:11:55.819>actually <00:11:55.959>able <00:11:56.100>to <00:11:56.240>pick <00:11:56.380>up <00:11:56.520>these <00:11:56.660>model <00:11:56.800>failures.

246
00:12:02.300 --> 00:12:04.000
<00:12:02.300>And <00:12:02.579>now <00:12:02.859>after <00:12:03.140>this <00:12:03.420>is <00:12:03.699>done.

247
00:12:04.700 --> 00:12:07.400
<00:12:04.700>Some <00:12:04.875>images <00:12:05.050>are <00:12:05.225>uploaded <00:12:05.400>to <00:12:05.575>the <00:12:05.750>cloud <00:12:05.925>where <00:12:06.100>they

248
00:12:07.400 --> 00:12:10.200
<00:12:07.400>are <00:12:07.555>labeled <00:12:07.711>either <00:12:07.866>by <00:12:08.022>humans <00:12:08.177>or <00:12:08.333>by <00:12:08.488>some <00:12:08.644>auto-labing <00:12:08.800>service

249
00:12:10.200 --> 00:12:12.700
<00:12:10.200>that <00:12:10.355>we <00:12:10.511>have <00:12:10.666>and <00:12:10.822>they're <00:12:10.977>used <00:12:11.133>to <00:12:11.288>retrain <00:12:11.444>the <00:12:11.600>model.

250
00:12:13.700 --> 00:12:17.000
<00:12:13.700>So <00:12:13.855>here <00:12:14.011>you'll <00:12:14.166>see <00:12:14.322>a <00:12:14.477>side <00:12:14.633>to <00:12:14.788>side <00:12:14.944>comparison <00:12:15.100>of

251
00:12:16.200 --> 00:12:19.300
<00:12:16.200>the <00:12:16.400>model <00:12:16.600>before <00:12:16.800>and <00:12:17.000>after <00:12:17.200>retraining. <00:12:17.400>So <00:12:17.600>the

252
00:12:19.300 --> 00:12:22.800
<00:12:19.300>middle <00:12:19.427>is <00:12:19.554>before <00:12:19.681>and <00:12:19.809>the <00:12:19.936>right <00:12:20.063>hand <00:12:20.190>side <00:12:20.318>is <00:12:20.445>after <00:12:20.572>retraining <00:12:20.699>and

253
00:12:22.800 --> 00:12:25.300
<00:12:22.800>you'll <00:12:23.079>notice <00:12:23.359>and <00:12:23.640>improvement <00:12:23.920>from <00:12:24.199>left

254
00:12:25.300 --> 00:12:25.700
<00:12:25.300>to <00:12:25.660>right.

255
00:12:27.400 --> 00:12:30.400
<00:12:27.400>So <00:12:27.555>here <00:12:27.711>for <00:12:27.866>example <00:12:28.022>with <00:12:28.177>the <00:12:28.333>curses <00:12:28.488>pointing <00:12:28.644>traffic <00:12:28.800>lights

256
00:12:30.400 --> 00:12:33.500
<00:12:30.400>that <00:12:30.540>were <00:12:30.680>detected <00:12:30.819>not <00:12:30.959>very <00:12:31.100>well <00:12:31.240>before <00:12:31.380>our <00:12:31.520>detected <00:12:31.660>better <00:12:31.800>in

257
00:12:33.500 --> 00:12:36.400
<00:12:33.500>the <00:12:33.675>after <00:12:33.850>image <00:12:34.025>some <00:12:34.200>traffic <00:12:34.375>lights <00:12:34.550>that <00:12:34.725>were <00:12:34.900>missed

258
00:12:36.400 --> 00:12:40.200
<00:12:36.400>are <00:12:36.540>now <00:12:36.680>being <00:12:36.819>detected <00:12:36.959>by <00:12:37.100>the <00:12:37.240>model. <00:12:37.380>There <00:12:37.520>are <00:12:37.660>fewer <00:12:37.800>experience

259
00:12:39.200 --> 00:12:42.400
<00:12:39.200>detections <00:12:39.480>Etc. <00:12:39.760>And <00:12:40.040>this <00:12:40.320>we <00:12:40.600>were

260
00:12:42.400 --> 00:12:45.200
<00:12:42.400>able <00:12:42.516>to <00:12:42.633>accomplish <00:12:42.750>with <00:12:42.866>just <00:12:42.983>a <00:12:43.100>minimal <00:12:43.216>amount <00:12:43.333>of <00:12:43.449>new <00:12:43.566>data <00:12:43.683>compared <00:12:43.800>to

261
00:12:45.200 --> 00:12:48.100
<00:12:45.200>if <00:12:45.340>we <00:12:45.480>were <00:12:45.620>collecting <00:12:45.760>every <00:12:45.900>single <00:12:46.040>image <00:12:46.180>we <00:12:46.320>saw <00:12:46.460>on <00:12:46.600>the

262
00:12:48.100 --> 00:12:48.200
road.

263
00:12:50.800 --> 00:12:54.800
<00:12:50.800>And <00:12:50.955>after <00:12:51.111>this <00:12:51.266>step <00:12:51.422>the <00:12:51.577>retrain <00:12:51.733>model <00:12:51.888>is <00:12:52.044>then <00:12:52.199>uploaded

264
00:12:53.800 --> 00:12:56.400
<00:12:53.800>to <00:12:54.150>a <00:12:54.500>model <00:12:54.849>registry <00:12:55.199>through

265
00:12:56.400 --> 00:13:00.300
<00:12:56.400>our <00:12:56.555>release <00:12:56.711>Pipeline. <00:12:56.866>And <00:12:57.022>this <00:12:57.177>is <00:12:57.333>a <00:12:57.488>platform <00:12:57.644>independent <00:12:57.800>representation

266
00:12:59.300 --> 00:13:02.200
<00:12:59.300>of <00:12:59.533>a <00:12:59.766>model <00:13:00.000>which <00:13:00.233>then <00:13:00.466>can <00:13:00.699>be

267
00:13:02.200 --> 00:13:05.600
<00:13:02.200>directly <00:13:02.340>deployed <00:13:02.480>into <00:13:02.620>the <00:13:02.760>car <00:13:02.900>through <00:13:03.040>and <00:13:03.180>over <00:13:03.320>the <00:13:03.460>air <00:13:03.600>update.

268
00:13:06.300 --> 00:13:08.800
<00:13:06.300>And <00:13:06.439>this <00:13:06.579>update <00:13:06.719>can <00:13:06.859>happen <00:13:07.000>in <00:13:07.140>just <00:13:07.280>a <00:13:07.420>matter <00:13:07.560>of <00:13:07.699>minutes.

269
00:13:09.600 --> 00:13:12.500
<00:13:09.600>While <00:13:09.755>the <00:13:09.911>car <00:13:10.066>is <00:13:10.222>in <00:13:10.377>operation <00:13:10.533>because <00:13:10.688>the <00:13:10.844>protected <00:13:11.000>area

270
00:13:12.500 --> 00:13:15.500
<00:13:12.500>blade <00:13:12.600>is <00:13:12.700>isolated <00:13:12.800>from <00:13:12.900>the <00:13:13.000>rest <00:13:13.100>of <00:13:13.200>the <00:13:13.300>vehicle <00:13:13.400>so <00:13:13.500>we <00:13:13.600>can <00:13:13.700>do <00:13:13.800>the <00:13:13.900>safely.

271
00:13:17.700 --> 00:13:20.300
<00:13:17.700>So <00:13:17.875>that <00:13:18.050>was <00:13:18.225>the <00:13:18.400>big <00:13:18.575>loop <00:13:18.750>and <00:13:18.925>to <00:13:19.100>put

272
00:13:20.300 --> 00:13:23.400
<00:13:20.300>this <00:13:20.439>on <00:13:20.579>the <00:13:20.719>road <00:13:20.859>quickly. <00:13:21.000>Our <00:13:21.140>developers <00:13:21.280>have <00:13:21.420>created <00:13:21.560>a <00:13:21.699>system

273
00:13:23.400 --> 00:13:26.500
<00:13:23.400>that <00:13:23.600>can <00:13:23.800>be <00:13:24.000>retrofitted <00:13:24.199>to <00:13:24.400>existing <00:13:24.600>vehicles <00:13:24.800>that

274
00:13:26.500 --> 00:13:27.500
<00:13:26.500>are <00:13:26.680>already <00:13:26.860>out <00:13:27.040>on <00:13:27.220>the <00:13:27.400>road.

275
00:13:28.300 --> 00:13:31.500
<00:13:28.300>And <00:13:28.439>this <00:13:28.579>Fleet <00:13:28.719>of <00:13:28.859>vehicles <00:13:29.000>also <00:13:29.140>known <00:13:29.280>as <00:13:29.420>the <00:13:29.560>pioneering <00:13:29.699>Fleet.

276
00:13:32.500 --> 00:13:35.900
<00:13:32.500>Is <00:13:32.675>pioneering <00:13:32.850>a <00:13:33.025>big <00:13:33.200>loop <00:13:33.375>technology <00:13:33.550>today <00:13:33.725>before <00:13:33.900>it's

277
00:13:35.900 --> 00:13:39.300
<00:13:35.900>rolled <00:13:36.100>out <00:13:36.300>in <00:13:36.500>the <00:13:36.699>unified <00:13:36.900>and <00:13:37.100>scalable <00:13:37.300>carry

278
00:13:38.300 --> 00:13:39.700
<00:13:38.300>out <00:13:39.560>platform.

279
00:13:40.600 --> 00:13:43.700
<00:13:40.600>And <00:13:40.775>the <00:13:40.950>aim <00:13:41.125>is <00:13:41.300>that <00:13:41.475>by <00:13:41.650>2030 <00:13:41.825>40 <00:13:42.000>million

280
00:13:43.700 --> 00:13:46.400
<00:13:43.700>new <00:13:43.840>vehicles <00:13:43.980>per <00:13:44.120>year <00:13:44.260>will <00:13:44.400>be <00:13:44.540>running <00:13:44.680>on <00:13:44.820>the <00:13:44.960>folks <00:13:45.100>the

281
00:13:46.400 --> 00:13:47.900
<00:13:46.400>wagon <00:13:46.737>group's <00:13:47.074>own <00:13:47.412>operating <00:13:47.750>platform.

282
00:13:49.800 --> 00:13:52.500
<00:13:49.800>So <00:13:49.974>the <00:13:50.150>big <00:13:50.324>loop <00:13:50.500>together <00:13:50.675>was <00:13:50.849>this <00:13:51.025>pioneering <00:13:51.199>Fleet

283
00:13:52.500 --> 00:13:55.200
<00:13:52.500>will <00:13:52.780>allow <00:13:53.060>us <00:13:53.340>a <00:13:53.620>career <00:13:53.900>to

284
00:13:55.200 --> 00:13:58.700
<00:13:55.200>leverage <00:13:55.433>and <00:13:55.666>active <00:13:55.900>learning <00:13:56.133>and <00:13:56.366>continuous <00:13:56.600>data-driven

285
00:13:58.700 --> 00:14:01.400
<00:13:58.700>development <00:13:58.933>for <00:13:59.166>vehicle <00:13:59.400>software <00:13:59.633>and <00:13:59.866>automated <00:14:00.100>driving.

286
00:14:04.200 --> 00:14:07.100
<00:14:04.200>And <00:14:04.300>finally, <00:14:04.400>I'd <00:14:04.500>like <00:14:04.600>to <00:14:04.700>say <00:14:04.800>that <00:14:04.900>we <00:14:05.000>would <00:14:05.100>love <00:14:05.200>to <00:14:05.300>have <00:14:05.400>you <00:14:05.500>join <00:14:05.600>us

287
00:14:07.100 --> 00:14:07.800
<00:14:07.100>in <00:14:07.415>this <00:14:07.730>journey.

288
00:14:08.800 --> 00:14:11.200
<00:14:08.800>If <00:14:08.955>you <00:14:09.111>are <00:14:09.266>an <00:14:09.422>expert <00:14:09.577>in <00:14:09.733>any <00:14:09.888>of <00:14:10.044>the <00:14:10.199>following

289
00:14:11.200 --> 00:14:14.400
<00:14:11.200>areas, <00:14:11.355>you're <00:14:11.511>very <00:14:11.666>welcome <00:14:11.822>to <00:14:11.977>reach <00:14:12.133>out <00:14:12.288>to <00:14:12.444>us <00:14:12.600>or

290
00:14:14.400 --> 00:14:17.300
<00:14:14.400>visit <00:14:14.574>our <00:14:14.750>website <00:14:14.925>to <00:14:15.100>apply <00:14:15.275>to <00:14:15.449>any <00:14:15.625>of <00:14:15.800>the

291
00:14:17.300 --> 00:14:18.900
<00:14:17.300>hundreds <00:14:17.533>of <00:14:17.766>open <00:14:18.000>positions <00:14:18.233>that <00:14:18.466>we <00:14:18.699>have.

292
00:14:21.500 --> 00:14:21.900
<00:14:21.500>Thank <00:14:21.860>you.

293
00:14:28.300 --> 00:14:31.400
<00:14:28.300>Thank <00:14:28.439>you <00:14:28.579>so <00:14:28.719>much. <00:14:28.859>That <00:14:29.000>was <00:14:29.140>that <00:14:29.280>was <00:14:29.420>great. <00:14:29.560>Now, <00:14:29.699>please

294
00:14:31.400 --> 00:14:34.500
<00:14:31.400>send <00:14:31.555>in <00:14:31.711>your <00:14:31.866>questions, <00:14:32.022>especially <00:14:32.177>if <00:14:32.333>you <00:14:32.488>have <00:14:32.644>technical <00:14:32.800>ones,

295
00:14:34.500 --> 00:14:37.400
<00:14:34.500>please <00:14:34.733>remember <00:14:34.966>she <00:14:35.200>is <00:14:35.433>a <00:14:35.666>senior <00:14:35.900>data

296
00:14:37.400 --> 00:14:40.400
<00:14:37.400>scientist. <00:14:37.866>The <00:14:38.333>very <00:14:38.800>first

297
00:14:40.400 --> 00:14:43.500
<00:14:40.400>question <00:14:40.555>we <00:14:40.711>actually <00:14:40.866>got <00:14:41.022>was <00:14:41.177>more <00:14:41.333>of <00:14:41.488>a <00:14:41.644>philosophical <00:14:41.800>question

298
00:14:43.500 --> 00:14:47.800
<00:14:43.500>than <00:14:43.700>anything <00:14:43.900>else. <00:14:44.100>So <00:14:44.300>I'm <00:14:44.500>I'm <00:14:44.700>curious <00:14:44.900>what

299
00:14:46.800 --> 00:14:49.100
<00:14:46.800>your <00:14:47.150>opinion <00:14:47.500>is <00:14:47.849>on <00:14:48.199>this

300
00:14:49.100 --> 00:14:50.600
<00:14:49.100>if <00:14:49.437>you <00:14:49.775>even <00:14:50.112>have <00:14:50.450>one.

301
00:14:52.300 --> 00:14:56.200
<00:14:52.300>And <00:14:52.766>so <00:14:53.233>Christos <00:14:53.699>asks,

302
00:14:55.200 --> 00:14:58.700
<00:14:55.200>what <00:14:55.400>about <00:14:55.600>the <00:14:55.800>decision <00:14:56.000>of <00:14:56.200>saving <00:14:56.400>pedestrian's <00:14:56.600>life

303
00:14:58.700 --> 00:15:01.300
<00:14:58.700>or <00:14:58.933>the <00:14:59.166>driver's <00:14:59.400>one <00:14:59.633>during <00:14:59.866>a <00:15:00.100>possible

304
00:15:01.300 --> 00:15:04.300
<00:15:01.300>accident. <00:15:01.579>Now, <00:15:01.859>that's <00:15:02.140>a <00:15:02.420>one <00:15:02.699>sentence.

305
00:15:04.300 --> 00:15:07.400
<00:15:04.300>Can <00:15:04.455>you <00:15:04.611>give <00:15:04.766>an <00:15:04.922>answer <00:15:05.077>within <00:15:05.233>the <00:15:05.388>next <00:15:05.544>two <00:15:05.699>and

306
00:15:07.400 --> 00:15:10.100
<00:15:07.400>a <00:15:07.540>half <00:15:07.680>hours? <00:15:07.819>That <00:15:07.959>is <00:15:08.100>a <00:15:08.240>very <00:15:08.380>tough <00:15:08.520>question <00:15:08.660>to <00:15:08.800>answer.

307
00:15:10.100 --> 00:15:13.500
<00:15:10.100>Ideally. <00:15:10.240>We <00:15:10.380>want <00:15:10.520>to <00:15:10.660>save <00:15:10.800>both. <00:15:10.940>Okay, <00:15:11.080>so <00:15:11.220>I'll <00:15:11.360>approach <00:15:11.500>is

308
00:15:13.500 --> 00:15:14.200
<00:15:13.500>to <00:15:14.130>optimize.

309
00:15:15.500 --> 00:15:18.000
<00:15:15.500>The <00:15:15.675>solution <00:15:15.850>so <00:15:16.025>that <00:15:16.200>we <00:15:16.375>minimize <00:15:16.550>the <00:15:16.725>loss <00:15:16.900>of

310
00:15:18.200 --> 00:15:21.100
<00:15:18.200>life <00:15:18.340>no <00:15:18.480>matter <00:15:18.620>if <00:15:18.760>it's <00:15:18.900>a <00:15:19.040>pedestrian <00:15:19.180>or <00:15:19.320>driver, <00:15:19.460>but <00:15:19.600>I

311
00:15:21.100 --> 00:15:24.200
<00:15:21.100>think <00:15:21.187>in <00:15:21.275>real <00:15:21.362>life <00:15:21.450>what <00:15:21.537>will <00:15:21.625>happen <00:15:21.712>is <00:15:21.800>I <00:15:21.887>mean, <00:15:21.975>we <00:15:22.062>are <00:15:22.150>subject <00:15:22.237>to <00:15:22.325>the <00:15:22.412>limitations <00:15:22.500>of

312
00:15:24.200 --> 00:15:27.600
<00:15:24.200>the <00:15:24.327>vehicle <00:15:24.454>itself. <00:15:24.581>So <00:15:24.709>it <00:15:24.836>will <00:15:24.963>do <00:15:25.090>whatever <00:15:25.218>it <00:15:25.345>is <00:15:25.472>most <00:15:25.600>capable

313
00:15:27.600 --> 00:15:30.200
<00:15:27.600>of <00:15:27.775>doing <00:15:27.950>to <00:15:28.125>optimize <00:15:28.300>saving <00:15:28.475>off <00:15:28.650>lives <00:15:28.825>whether <00:15:29.000>it's

314
00:15:30.200 --> 00:15:31.100
<00:15:30.200>a <00:15:30.402>driver <00:15:30.605>or <00:15:30.807>The <00:15:31.010>Pedestrian

315
00:15:32.500 --> 00:15:35.900
<00:15:32.500>That's <00:15:32.675>that's <00:15:32.850>great. <00:15:33.025>Thank <00:15:33.200>you <00:15:33.375>so <00:15:33.550>much <00:15:33.725>now <00:15:33.900>let

316
00:15:35.900 --> 00:15:36.000
<00:15:35.900>me <00:15:35.990>see.

317
00:15:39.100 --> 00:15:40.800
so

318
00:15:42.300 --> 00:15:43.700
<00:15:42.300>Kareem <00:15:43.560>asks

319
00:15:44.500 --> 00:15:48.200
<00:15:44.500>what <00:15:44.700>the <00:15:44.900>decision <00:15:45.100>to <00:15:45.300>take <00:15:45.500>if <00:15:45.700>Cloud <00:15:45.900>timeouts

320
00:15:47.200 --> 00:15:50.800
<00:15:47.200>take <00:15:47.433>too <00:15:47.666>long <00:15:47.900>or <00:15:48.133>the <00:15:48.366>feedback <00:15:48.600>with

321
00:15:50.800 --> 00:15:54.500
<00:15:50.800>the <00:15:51.079>car <00:15:51.359>with <00:15:51.640>the <00:15:51.920>decision. <00:15:52.199>Sorry,

322
00:15:53.500 --> 00:15:56.600
<00:15:53.500>and <00:15:53.850>why <00:15:54.200>can't <00:15:54.550>the <00:15:54.900>model

323
00:15:56.600 --> 00:15:58.700
<00:15:56.600>be <00:15:56.880>locally <00:15:57.160>available <00:15:57.440>on <00:15:57.720>the <00:15:58.000>car?

324
00:16:00.200 --> 00:16:03.200
<00:16:00.200>So <00:16:00.307>the <00:16:00.415>model <00:16:00.523>if <00:16:00.630>you're <00:16:00.738>talking <00:16:00.846>about <00:16:00.953>the <00:16:01.061>neural <00:16:01.169>network, <00:16:01.276>it <00:16:01.384>actually <00:16:01.492>is <00:16:01.600>running

325
00:16:03.200 --> 00:16:06.100
<00:16:03.200>on <00:16:03.327>an <00:16:03.454>embedded <00:16:03.581>device <00:16:03.709>inside <00:16:03.836>the <00:16:03.963>car. <00:16:04.090>So <00:16:04.218>it <00:16:04.345>is <00:16:04.472>doing <00:16:04.600>that

326
00:16:06.100 --> 00:16:06.700
<00:16:06.100>in <00:16:06.370>real <00:16:06.640>time.

327
00:16:07.500 --> 00:16:08.300
<00:16:07.500>Yeah, <00:16:07.860>thank <00:16:08.219>you.

328
00:16:10.200 --> 00:16:13.500
<00:16:10.200>Oh <00:16:10.355>Jan <00:16:10.511>asked. <00:16:10.666>This <00:16:10.822>is <00:16:10.977>a <00:16:11.133>fun <00:16:11.288>one. <00:16:11.444>What's <00:16:11.600>the

329
00:16:13.500 --> 00:16:16.300
<00:16:13.500>best <00:16:13.655>part <00:16:13.811>to <00:16:13.966>be <00:16:14.122>about <00:16:14.277>being <00:16:14.433>a <00:16:14.588>data <00:16:14.744>scientist <00:16:14.900>at

330
00:16:16.300 --> 00:16:17.600
<00:16:16.300>carried <00:16:17.470>gosh.

331
00:16:18.500 --> 00:16:18.800
Thanks.

332
00:16:19.300 --> 00:16:21.500
<00:16:19.300>young <00:16:19.650>more <00:16:20.000>philosophical <00:16:20.349>questions <00:16:20.699>there

333
00:16:22.900 --> 00:16:26.000
<00:16:22.900>That <00:16:23.055>yarn <00:16:23.211>is <00:16:23.366>actually <00:16:23.522>a <00:16:23.677>colleague <00:16:23.833>of <00:16:23.988>mine. <00:16:24.144>So <00:16:24.300>working

334
00:16:25.200 --> 00:16:28.100
<00:16:25.200>with <00:16:25.400>yarn <00:16:25.600>is <00:16:25.800>probably <00:16:26.000>the <00:16:26.200>the <00:16:26.400>answer <00:16:26.600>he

335
00:16:28.100 --> 00:16:31.000
<00:16:28.100>would <00:16:28.275>be <00:16:28.450>looking <00:16:28.625>forward <00:16:28.800>to <00:16:28.975>I'll <00:16:29.150>say <00:16:29.325>it <00:16:29.500>is.

336
00:16:31.300 --> 00:16:34.300
<00:16:31.300>Great <00:16:31.416>with <00:16:31.533>young <00:16:31.650>but <00:16:31.766>I <00:16:31.883>think <00:16:32.000>at <00:16:32.116>Courier <00:16:32.233>it's <00:16:32.349>really <00:16:32.466>great <00:16:32.583>to <00:16:32.699>be

337
00:16:34.300 --> 00:16:37.600
<00:16:34.300>working <00:16:34.416>with <00:16:34.533>a <00:16:34.650>lot <00:16:34.766>of <00:16:34.883>very <00:16:35.000>smart <00:16:35.116>and <00:16:35.233>motivated <00:16:35.349>colleagues <00:16:35.466>and <00:16:35.583>what <00:16:35.699>I

338
00:16:37.600 --> 00:16:40.700
<00:16:37.600>really <00:16:37.716>like <00:16:37.833>about <00:16:37.950>data <00:16:38.066>science <00:16:38.183>is <00:16:38.300>that <00:16:38.416>this <00:16:38.533>is <00:16:38.650>such <00:16:38.766>a <00:16:38.883>new <00:16:39.000>and

339
00:16:40.700 --> 00:16:43.400
<00:16:40.700>dynamic <00:16:40.933>domain <00:16:41.166>that <00:16:41.400>I'm <00:16:41.633>always <00:16:41.866>learning <00:16:42.100>something

340
00:16:43.400 --> 00:16:43.900
<00:16:43.400>new <00:16:43.850>here.

341
00:16:45.500 --> 00:16:48.600
<00:16:45.500>Simon <00:16:45.700>asks, <00:16:45.900>how <00:16:46.100>do <00:16:46.300>you <00:16:46.500>select <00:16:46.700>an <00:16:46.900>appropriate

342
00:16:48.600 --> 00:16:50.100
<00:16:48.600>uncertainty <00:16:49.950>threshold?

343
00:16:56.600 --> 00:16:59.400
<00:16:56.600>And <00:16:56.800>there's <00:16:57.000>no <00:16:57.200>fixed <00:16:57.400>threshold <00:16:57.600>because <00:16:57.800>depends <00:16:58.000>on

344
00:16:59.400 --> 00:17:02.200
<00:16:59.400>the <00:16:59.516>context <00:16:59.633>you're <00:16:59.750>driving <00:16:59.866>in <00:16:59.983>right? <00:17:00.100>So <00:17:00.216>if <00:17:00.333>you're <00:17:00.449>driving <00:17:00.566>in <00:17:00.683>a <00:17:00.800>busy

345
00:17:02.200 --> 00:17:05.300
<00:17:02.200>City, <00:17:02.316>the <00:17:02.433>threshold <00:17:02.550>is <00:17:02.666>going <00:17:02.783>to <00:17:02.900>be <00:17:03.016>different <00:17:03.133>from <00:17:03.250>if <00:17:03.366>you're <00:17:03.483>driving <00:17:03.600>in

346
00:17:05.300 --> 00:17:08.200
<00:17:05.300>the <00:17:05.427>middle <00:17:05.554>of <00:17:05.681>nowhere <00:17:05.809>on <00:17:05.936>an <00:17:06.063>open <00:17:06.190>highway. <00:17:06.318>So <00:17:06.445>this <00:17:06.572>is <00:17:06.700>going

347
00:17:08.200 --> 00:17:11.400
<00:17:08.200>to <00:17:08.327>be <00:17:08.454>a <00:17:08.581>dynamic <00:17:08.709>threshold <00:17:08.836>that <00:17:08.963>we <00:17:09.090>pick <00:17:09.218>based <00:17:09.345>on <00:17:09.472>what <00:17:09.600>the

348
00:17:11.400 --> 00:17:12.200
<00:17:11.400>environment <00:17:11.760>is <00:17:12.120>like.

349
00:17:14.200 --> 00:17:17.600
<00:17:14.200>You <00:17:14.550>showed <00:17:14.900>a <00:17:15.250>car <00:17:15.600>in

350
00:17:17.600 --> 00:17:23.200
<00:17:17.599>California <00:17:17.800>that <00:17:18.000>was <00:17:18.199>not <00:17:18.399>trained <00:17:18.599>in <00:17:18.800>California. <00:17:19.000>Felix

351
00:17:20.200 --> 00:17:23.600
us

352
00:17:23.600 --> 00:17:26.500
<00:17:23.599>do <00:17:23.833>cars <00:17:24.066>in <00:17:24.300>different <00:17:24.533>regions <00:17:24.766>receive <00:17:25.000>different

353
00:17:26.500 --> 00:17:29.500
<00:17:26.500>data <00:17:26.700>and <00:17:26.900>are <00:17:27.099>they <00:17:27.300>trained <00:17:27.500>with <00:17:27.700>with <00:17:27.900>different

354
00:17:29.500 --> 00:17:29.700
data?

355
00:17:30.800 --> 00:17:33.300
<00:17:30.800>he <00:17:30.975>also <00:17:31.149>our <00:17:31.325>hope <00:17:31.500>is <00:17:31.675>that <00:17:31.850>we <00:17:32.024>don't <00:17:32.200>have

356
00:17:33.300 --> 00:17:36.500
<00:17:33.300>to <00:17:33.407>specialize <00:17:33.515>too <00:17:33.623>much <00:17:33.730>for <00:17:33.838>the <00:17:33.946>region <00:17:34.053>because <00:17:34.161>it <00:17:34.269>will <00:17:34.376>be <00:17:34.484>really <00:17:34.592>costly <00:17:34.700>to

357
00:17:36.500 --> 00:17:38.100
<00:17:36.500>develop <00:17:36.850>different <00:17:37.200>model <00:17:37.550>for <00:17:37.900>every

358
00:17:38.900 --> 00:17:41.300
<00:17:38.900>City <00:17:39.133>yeah <00:17:39.366>in <00:17:39.600>the <00:17:39.833>world, <00:17:40.066>so <00:17:40.300>we

359
00:17:41.300 --> 00:17:44.200
<00:17:41.300>try <00:17:41.440>to <00:17:41.580>just <00:17:41.720>have <00:17:41.860>a <00:17:42.000>very <00:17:42.139>vast <00:17:42.280>and <00:17:42.419>diverse <00:17:42.560>data <00:17:42.700>set

360
00:17:44.200 --> 00:17:48.300
<00:17:44.200>to <00:17:44.293>train <00:17:44.386>the <00:17:44.480>model <00:17:44.573>on <00:17:44.666>so <00:17:44.760>that <00:17:44.853>we <00:17:44.946>don't <00:17:45.040>have <00:17:45.133>to <00:17:45.226>do <00:17:45.320>too <00:17:45.413>much <00:17:45.506>specializing <00:17:45.600>to

361
00:17:47.300 --> 00:17:48.500
<00:17:47.300>begin <00:17:48.380>with.

362
00:17:49.500 --> 00:17:53.000
<00:17:49.500>I <00:17:49.655>I <00:17:49.811>own <00:17:49.966>a <00:17:50.122>motorcycle <00:17:50.277>and <00:17:50.433>have <00:17:50.588>a <00:17:50.744>TomTom <00:17:50.900>which

363
00:17:52.300 --> 00:17:55.200
<00:17:52.300>gives <00:17:52.533>me <00:17:52.766>navigation <00:17:53.000>and <00:17:53.233>whenever <00:17:53.466>I <00:17:53.700>go

364
00:17:55.200 --> 00:17:58.900
<00:17:55.200>on <00:17:55.316>holiday, <00:17:55.433>it's <00:17:55.550>like <00:17:55.666>okay, <00:17:55.783>let <00:17:55.900>me <00:17:56.016>upload <00:17:56.133>the <00:17:56.250>maps <00:17:56.366>and <00:17:56.483>that <00:17:56.600>would

365
00:17:58.900 --> 00:18:01.300
<00:17:58.900>really <00:17:59.075>suck <00:17:59.250>if <00:17:59.425>you <00:17:59.600>have <00:17:59.775>to <00:17:59.950>wait. <00:18:00.125>I <00:18:00.300>don't

366
00:18:01.300 --> 00:18:05.000
<00:18:01.300>know <00:18:01.399>a <00:18:01.500>couple <00:18:01.600>days <00:18:01.700>before <00:18:01.800>you <00:18:01.899>can <00:18:02.000>go <00:18:02.100>on <00:18:02.200>holiday <00:18:02.300>with <00:18:02.399>your <00:18:02.500>car. <00:18:02.600>So <00:18:02.700>that's

367
00:18:04.400 --> 00:18:08.500
<00:18:04.400>great. <00:18:04.680>Let <00:18:04.960>me <00:18:05.240>see. <00:18:05.520>Oh, <00:18:05.800>yes.

368
00:18:08.500 --> 00:18:11.600
<00:18:08.500>PR <00:18:08.850>us <00:18:09.200>a <00:18:09.550>philosophical <00:18:09.900>question

369
00:18:11.600 --> 00:18:14.800
<00:18:11.600>again <00:18:12.300>is <00:18:13.000>full

370
00:18:14.800 --> 00:18:17.800
<00:18:14.800>self-driving <00:18:15.080>a <00:18:15.360>goal <00:18:15.639>and <00:18:15.919>this <00:18:16.200>is

371
00:18:19.600 --> 00:18:22.500
<00:18:19.600>Even <00:18:19.949>tougher <00:18:20.300>question. <00:18:20.649>What's <00:18:21.000>its

372
00:18:22.600 --> 00:18:22.700
ETA?

373
00:18:27.700 --> 00:18:29.300
<00:18:27.700>the <00:18:28.166>correct <00:18:28.633>one <00:18:29.100>it

374
00:18:31.600 --> 00:18:33.500
<00:18:31.600>is <00:18:31.740>leave <00:18:31.879>here, <00:18:32.020>but <00:18:32.159>I <00:18:32.300>cannot <00:18:32.439>really <00:18:32.580>give <00:18:32.719>you <00:18:32.860>an <00:18:33.000>ETA.

375
00:18:34.500 --> 00:18:37.500
<00:18:34.500>I <00:18:34.627>don't <00:18:34.754>think <00:18:34.881>anyone <00:18:35.009>knows. <00:18:35.136>No, <00:18:35.263>I <00:18:35.390>think <00:18:35.518>everybody <00:18:35.645>that's <00:18:35.772>tried <00:18:35.900>has

376
00:18:37.500 --> 00:18:40.800
<00:18:37.500>been <00:18:37.780>mistaken <00:18:38.060>so <00:18:38.340>far. <00:18:38.620>Let <00:18:38.900>me

377
00:18:40.800 --> 00:18:43.500
<00:18:40.800>see <00:18:40.927>how <00:18:41.054>much <00:18:41.181>time <00:18:41.309>do <00:18:41.436>we <00:18:41.563>have <00:18:41.690>left? <00:18:41.818>I <00:18:41.945>think <00:18:42.072>time <00:18:42.200>for

378
00:18:43.500 --> 00:18:44.500
<00:18:43.500>one <00:18:43.950>more <00:18:44.400>question.

379
00:18:48.200 --> 00:18:51.300
<00:18:48.200>Actually, <00:18:48.480>oh <00:18:48.760>one <00:18:49.040>I'm <00:18:49.320>curious <00:18:49.600>what

380
00:18:51.300 --> 00:18:54.100
<00:18:51.300>you <00:18:51.427>would <00:18:51.554>like <00:18:51.681>to <00:18:51.809>to <00:18:51.936>give <00:18:52.063>the <00:18:52.190>crowd. <00:18:52.318>Is <00:18:52.445>there <00:18:52.572>anything <00:18:52.700>that

381
00:18:54.100 --> 00:18:57.100
<00:18:54.100>you <00:18:54.216>cut <00:18:54.333>from <00:18:54.449>your <00:18:54.566>presentation <00:18:54.683>where <00:18:54.800>you <00:18:54.916>like? <00:18:55.033>I <00:18:55.149>really <00:18:55.266>wanted <00:18:55.383>to <00:18:55.500>share

382
00:18:57.100 --> 00:18:57.700
<00:18:57.100>that <00:18:57.280>with <00:18:57.460>the <00:18:57.640>audience.

383
00:18:59.300 --> 00:19:02.600
<00:18:59.300>Well, <00:18:59.407>I <00:18:59.515>just <00:18:59.623>want <00:18:59.730>to <00:18:59.838>say <00:18:59.946>that <00:19:00.053>this <00:19:00.161>is <00:19:00.269>a <00:19:00.376>really <00:19:00.484>exciting <00:19:00.592>adventure <00:19:00.700>and

384
00:19:02.600 --> 00:19:04.500
<00:19:02.600>we <00:19:02.774>would <00:19:02.949>love <00:19:03.125>to <00:19:03.300>have <00:19:03.475>more <00:19:03.649>people <00:19:03.824>join <00:19:04.000>us.

385
00:19:05.400 --> 00:19:08.400
<00:19:05.400>Excellent. <00:19:05.750>So <00:19:06.100>apply <00:19:06.450>for <00:19:06.800>a

386
00:19:08.400 --> 00:19:11.900
<00:19:08.400>job <00:19:08.540>at <00:19:08.680>carried <00:19:08.820>and <00:19:08.960>become <00:19:09.100>an <00:19:09.240>excellent <00:19:09.380>colleague <00:19:09.520>of <00:19:09.660>young <00:19:09.800>and

387
00:19:11.900 --> 00:19:15.500
<00:19:11.900>of <00:19:12.366>course <00:19:12.833>yourself <00:19:13.300>big

388
00:19:14.500 --> 00:19:16.700
<00:19:14.500>big <00:19:14.966>round <00:19:15.433>of <00:19:15.900>applause.

389
00:19:17.400 --> 00:19:17.700
<00:19:17.400>Thank <00:19:17.670>you.