1
00:00:00,000 --> 00:00:06,409


2
00:00:06,409 --> 00:00:08,874
[MUSIC PLAYING]

3
00:00:08,874 --> 00:00:20,033


4
00:00:20,033 --> 00:00:21,700
Hey Nathan, do you
know where Leslie is?

5
00:00:21,700 --> 00:00:23,560
Yeah, she's in EJ272.

6
00:00:23,560 --> 00:00:24,060
Thanks.

7
00:00:24,060 --> 00:00:25,896
I have a bagel for her.

8
00:00:25,896 --> 00:00:28,386
[MUSIC PLAYING]

9
00:00:28,386 --> 00:00:36,870


10
00:00:36,870 --> 00:00:38,196
Hey Leslie.

11
00:00:38,196 --> 00:00:39,152
A bagel.

12
00:00:39,152 --> 00:00:45,370


13
00:00:45,370 --> 00:00:48,760
Suppose Stan is
not a human being.

14
00:00:48,760 --> 00:00:53,860
Suppose Stan is a robot and has
to deliver a bagel to Leslie.

15
00:00:53,860 --> 00:00:56,740
The picture changes.

16
00:00:56,740 --> 00:00:59,800
What is a relatively
simple task for Stan

17
00:00:59,800 --> 00:01:04,420
is an extremely challenging
one for a robot.

18
00:01:04,420 --> 00:01:07,420
Computer scientists
at SRI International

19
00:01:07,420 --> 00:01:11,890
are developing a robot that can
perform complex tasks, tasks

20
00:01:11,890 --> 00:01:15,670
that require the robot to
have advanced perception,

21
00:01:15,670 --> 00:01:20,230
natural language, and
reasoning abilities.

22
00:01:20,230 --> 00:01:22,360
While researchers
have made advances

23
00:01:22,360 --> 00:01:24,580
in each of these
areas, up to now

24
00:01:24,580 --> 00:01:27,310
few have attempted to
combine these technologies

25
00:01:27,310 --> 00:01:29,740
into a single robot.

26
00:01:29,740 --> 00:01:33,325
That's where Flakey comes in.

27
00:01:33,325 --> 00:01:37,120
Flakey, the product of SRI's
Artificial Intelligence Center,

28
00:01:37,120 --> 00:01:40,630
is a robot consisting
of computers, sensors,

29
00:01:40,630 --> 00:01:43,690
and other hardware.

30
00:01:43,690 --> 00:01:50,230
Flakey's sensors include
vision sensors, sonar sensors,

31
00:01:50,230 --> 00:01:53,980
and contact sensors.

32
00:01:53,980 --> 00:01:57,880
Flakey's shaft encoders,
sensors connected to the wheels,

33
00:01:57,880 --> 00:02:02,860
can tell how far and how
fast the robot has traveled.

34
00:02:02,860 --> 00:02:06,340
A speech recognition system
allows those around Flakey

35
00:02:06,340 --> 00:02:10,580
to communicate using
voice commands.

36
00:02:10,580 --> 00:02:15,710
You have a bagel.

37
00:02:15,710 --> 00:02:17,150
Enter.

38
00:02:17,150 --> 00:02:21,530
Let's see how Flakey goes about
delivering a bagel to Leslie.

39
00:02:21,530 --> 00:02:23,240
Parsing sentence.

40
00:02:23,240 --> 00:02:26,560
You have a bagel.

41
00:02:26,560 --> 00:02:30,160
Finished successfully
parsing sentence.

42
00:02:30,160 --> 00:02:39,640
I want Leslie to have the bagel.

43
00:02:39,640 --> 00:02:41,630
Enter.

44
00:02:41,630 --> 00:02:45,750
I want Leslie to have the bagel.

45
00:02:45,750 --> 00:02:49,350
Finished successfully
parsing sentence.

46
00:02:49,350 --> 00:02:56,700
Nathan knows where Leslie is.

47
00:02:56,700 --> 00:02:59,030
Enter.

48
00:02:59,030 --> 00:03:00,730
Parsing sentence.

49
00:03:00,730 --> 00:03:03,970
Nathan knows where Leslie is.

50
00:03:03,970 --> 00:03:07,660
Finished successfully
parsing sentence.

51
00:03:07,660 --> 00:03:13,645
Nathan is in EJ227.

52
00:03:13,645 --> 00:03:14,145
Enter.

53
00:03:14,145 --> 00:03:17,750


54
00:03:17,750 --> 00:03:19,430
Parsing sentence.

55
00:03:19,430 --> 00:03:22,370
Nathan is in EJ227.

56
00:03:22,370 --> 00:03:24,485
Finished successfully
parsing sentence.

57
00:03:24,485 --> 00:03:35,228


58
00:03:35,228 --> 00:03:38,124
Moving to region 13.

59
00:03:38,124 --> 00:03:41,112
[MUSIC PLAYING]

60
00:03:41,112 --> 00:04:03,070


61
00:04:03,070 --> 00:04:04,540
OK.

62
00:04:04,540 --> 00:04:06,306
Well, I'll talk to you later.

63
00:04:06,306 --> 00:04:06,806
Bye bye.

64
00:04:06,806 --> 00:04:17,459


65
00:04:17,459 --> 00:04:20,070
Where is Leslie?

66
00:04:20,070 --> 00:04:26,080
Leslie is in EJ272.

67
00:04:26,080 --> 00:04:26,580
Enter.

68
00:04:26,580 --> 00:04:33,580


69
00:04:33,580 --> 00:04:35,410
Parsing sentence.

70
00:04:35,410 --> 00:04:40,260
Leslie is in EJ272.

71
00:04:40,260 --> 00:04:43,770
Finished successfully
parsing sentence.

72
00:04:43,770 --> 00:04:46,245
[MUSIC PLAYING]

73
00:04:46,245 --> 00:05:06,060


74
00:05:06,060 --> 00:05:09,756
Moving to region three.

75
00:05:09,756 --> 00:05:12,236
[MUSIC PLAYING]

76
00:05:12,236 --> 00:05:58,940


77
00:05:58,940 --> 00:06:00,155
Moving to region 18.

78
00:06:00,155 --> 00:06:02,905


79
00:06:02,905 --> 00:06:05,899
[MUSIC PLAYING]

80
00:06:05,899 --> 00:06:39,831


81
00:06:39,831 --> 00:06:42,927
Moving to region 15.

82
00:06:42,927 --> 00:06:45,422
[MUSIC PLAYING]

83
00:06:45,422 --> 00:07:39,850


84
00:07:39,850 --> 00:07:41,860
Take a bagel.

85
00:07:41,860 --> 00:07:42,610
Thank you, Flakey.

86
00:07:42,610 --> 00:07:48,630


87
00:07:48,630 --> 00:07:52,420
We've just seen Flakey
deliver a bagel to Leslie.

88
00:07:52,420 --> 00:07:56,460
Now let's take a closer
look at how Flakey operates.

89
00:07:56,460 --> 00:07:59,310
Flakey's sensors
constantly collect data

90
00:07:59,310 --> 00:08:01,270
from his environment.

91
00:08:01,270 --> 00:08:04,680
The data he gathers must
undergo a continual process

92
00:08:04,680 --> 00:08:07,080
of interpretation
in order for Flakey

93
00:08:07,080 --> 00:08:10,140
to select actions to perform.

94
00:08:10,140 --> 00:08:12,330
Here's how it works.

95
00:08:12,330 --> 00:08:17,380
Flakey's software is organized
into three layers, low, middle,

96
00:08:17,380 --> 00:08:19,710
and high level.

97
00:08:19,710 --> 00:08:22,260
Each layer can interpret
incoming information

98
00:08:22,260 --> 00:08:24,700
and carry out certain tasks.

99
00:08:24,700 --> 00:08:26,940
The higher the level,
the more abstract

100
00:08:26,940 --> 00:08:30,210
the information and tasks.

101
00:08:30,210 --> 00:08:32,640
Low level tracks
Flakey's location

102
00:08:32,640 --> 00:08:35,490
by finding walls and doors.

103
00:08:35,490 --> 00:08:39,240
This computer screen shows
a map of the corridors.

104
00:08:39,240 --> 00:08:42,990
It illustrates how the robot
uses motion feedback and sonar

105
00:08:42,990 --> 00:08:47,610
measurements to locate
and classify surfaces.

106
00:08:47,610 --> 00:08:50,880
By continually matching
descriptions of these surfaces

107
00:08:50,880 --> 00:08:53,520
against a stored
map of the corridor,

108
00:08:53,520 --> 00:08:56,115
Flakey reduces his uncertainty
about his position.

109
00:08:56,115 --> 00:08:58,740


110
00:08:58,740 --> 00:09:02,920
Low level also takes output from
the speech recognition system

111
00:09:02,920 --> 00:09:05,520
and collects word sequences
that can be interpreted

112
00:09:05,520 --> 00:09:08,940
by higher levels of software.

113
00:09:08,940 --> 00:09:12,060
Finally, low level is
responsible for generating

114
00:09:12,060 --> 00:09:14,580
Flakey's speech and
for steering him

115
00:09:14,580 --> 00:09:18,210
by controlling the
motion of his wheels.

116
00:09:18,210 --> 00:09:21,480
Going up the hierarchy,
we come to a middle level

117
00:09:21,480 --> 00:09:24,900
where linguistic
processing takes place.

118
00:09:24,900 --> 00:09:27,240
In order for Flakey to
understand sentences

119
00:09:27,240 --> 00:09:29,340
that are spoken
to him, mid-level

120
00:09:29,340 --> 00:09:33,570
must analyze them and construct
a symbolic representation.

121
00:09:33,570 --> 00:09:39,450
This is done using a
unification based chart parser.

122
00:09:39,450 --> 00:09:42,240
When Flakey decides
to speak, mid-level

123
00:09:42,240 --> 00:09:47,220
must convert symbolic forms
into English sentences.

124
00:09:47,220 --> 00:09:51,000
Mid-level uses the same grammar
to generate Flakey's sentences

125
00:09:51,000 --> 00:09:54,120
as it does to parse.

126
00:09:54,120 --> 00:09:56,160
Where is Leslie?

127
00:09:56,160 --> 00:09:58,890
Mid-level also
sequences and monitors

128
00:09:58,890 --> 00:10:00,495
the execution of robot actions.

129
00:10:00,495 --> 00:10:03,090


130
00:10:03,090 --> 00:10:06,480
At high level,
reasoning takes place.

131
00:10:06,480 --> 00:10:09,090
Reasoning is carried out
by a deductive system

132
00:10:09,090 --> 00:10:12,900
similar to Prolog but
incorporating full first order

133
00:10:12,900 --> 00:10:14,730
deduction.

134
00:10:14,730 --> 00:10:18,780
The deductive system operates
on a knowledge base of facts,

135
00:10:18,780 --> 00:10:21,810
including new facts
received from mid-level

136
00:10:21,810 --> 00:10:25,290
as well as previously
stored facts.

137
00:10:25,290 --> 00:10:28,500
This reasoning process
produces a plan of action

138
00:10:28,500 --> 00:10:33,120
that will cause Flakey to
achieve his overall goal.

139
00:10:33,120 --> 00:10:36,600
Let's look at some examples of
how these three levels interact

140
00:10:36,600 --> 00:10:40,470
to help Flakey reach his goal.

141
00:10:40,470 --> 00:10:44,070
In this instance, Flakey
receives a speech command.

142
00:10:44,070 --> 00:10:52,710
I want Leslie to have the bagel.

143
00:10:52,710 --> 00:10:54,690
Enter.

144
00:10:54,690 --> 00:10:58,470
The interpretive
process looks like this.

145
00:10:58,470 --> 00:11:02,400
Flakey receives speech
input from his sensors.

146
00:11:02,400 --> 00:11:06,340
Low level collects the
string of spoken words.

147
00:11:06,340 --> 00:11:10,710
The data is further refined
when it reaches mid-level.

148
00:11:10,710 --> 00:11:13,770
Here parsing produces
a symbolic formula

149
00:11:13,770 --> 00:11:17,010
that expresses the
meaning of the sentence.

150
00:11:17,010 --> 00:11:19,290
The data continues
up to high level,

151
00:11:19,290 --> 00:11:23,520
where reasoning takes place
and conclusions are drawn.

152
00:11:23,520 --> 00:11:27,180
High level forms plans by
projecting the consequences

153
00:11:27,180 --> 00:11:30,210
of alternative actions
until finally it selects

154
00:11:30,210 --> 00:11:34,200
the actions to be executed.

155
00:11:34,200 --> 00:11:36,960
In this instance, high
level concludes Flakey

156
00:11:36,960 --> 00:11:39,540
should move to region one.

157
00:11:39,540 --> 00:11:43,440
Mid-level initiates the
move command and low level

158
00:11:43,440 --> 00:11:46,740
steers the robot to region
one by controlling wheel

159
00:11:46,740 --> 00:11:49,590
velocities.

160
00:11:49,590 --> 00:11:52,830
Let's look at another example of
the interaction between levels

161
00:11:52,830 --> 00:11:55,620
of this hierarchy.

162
00:11:55,620 --> 00:11:59,280
Here Flakey encounters
and obstruction.

163
00:11:59,280 --> 00:12:04,320
His sensors identify a barrier
and immediately Flakey stops.

164
00:12:04,320 --> 00:12:08,040
The interpretive
process looks like this.

165
00:12:08,040 --> 00:12:12,000
Low level monitors data
gathered by the sonar sensors,

166
00:12:12,000 --> 00:12:15,330
detects a closed
door, and stops Flakey

167
00:12:15,330 --> 00:12:19,140
without waiting for
high level processing.

168
00:12:19,140 --> 00:12:23,100
The information about the closed
door travels up from low level

169
00:12:23,100 --> 00:12:26,010
through mid-level to high level.

170
00:12:26,010 --> 00:12:28,680
High level realizes
that the closed door

171
00:12:28,680 --> 00:12:32,250
invalidates a key step in the
original plan for achieving

172
00:12:32,250 --> 00:12:34,140
the goal.

173
00:12:34,140 --> 00:12:37,380
This leads high level to
formulate a new plan involving

174
00:12:37,380 --> 00:12:40,260
an alternate route.

175
00:12:40,260 --> 00:12:45,000
High level concludes that Flakey
should move to region two.

176
00:12:45,000 --> 00:12:47,370
As the command is
executed, Flakey

177
00:12:47,370 --> 00:12:51,150
turns away from the closed
door and begins his new route.

178
00:12:51,150 --> 00:12:55,710


179
00:12:55,710 --> 00:12:58,500
You have seen Flakey follow
a command given to him

180
00:12:58,500 --> 00:13:01,230
by Stan, a human being.

181
00:13:01,230 --> 00:13:04,750
Scientists at SRI
designed Flakey's software

182
00:13:04,750 --> 00:13:06,450
so he will be able
to communicate

183
00:13:06,450 --> 00:13:10,950
not only with human beings
but with other robots.

184
00:13:10,950 --> 00:13:14,070
The concept of robots
working together in teams

185
00:13:14,070 --> 00:13:17,910
suggests a wide range
of new applications.

186
00:13:17,910 --> 00:13:20,130
To make cooperating
robots a reality,

187
00:13:20,130 --> 00:13:22,650
however, will require
further research

188
00:13:22,650 --> 00:13:26,990
in the area of intelligent
communicating agents.

189
00:13:26,990 --> 00:13:30,380
Taking a step into the
future, suppose Stan

190
00:13:30,380 --> 00:13:32,060
is out of the picture.

191
00:13:32,060 --> 00:13:34,460
In his place is another robot.

192
00:13:34,460 --> 00:13:36,650
Let's call her Ramona.

193
00:13:36,650 --> 00:13:41,090
Together Flakey and
Ramona are a team.

194
00:13:41,090 --> 00:13:45,780
Their teamwork might
look something like this.

195
00:13:45,780 --> 00:13:48,800
Ramona, Leslie wants a bagel.

196
00:13:48,800 --> 00:13:51,110
Where is Leslie?

197
00:13:51,110 --> 00:13:54,020
Leslie is in EJ272.

198
00:13:54,020 --> 00:13:57,030
Please take a bagel to Leslie.

199
00:13:57,030 --> 00:13:59,870
OK, Flakey.

200
00:13:59,870 --> 00:14:03,520
[MUSIC PLAYING]

201
00:14:03,520 --> 00:14:36,000