1
00:00:00,000 --> 00:00:03,320
Audrow Nash: Hey everyone.
Welcome to the Audrow Nash

2
00:00:03,320 --> 00:00:06,954
Podcast. Today we're diving into
the world of robotics and

3
00:00:06,954 --> 00:00:10,838
manufacturing with Saman Farid,
the founder and CEO of Formic.

4
00:00:10,838 --> 00:00:14,033
Ever wonder how robots could
revolutionize American

5
00:00:14,033 --> 00:00:17,541
factories? Saman and his team
are tackling this, head on

6
00:00:17,541 --> 00:00:21,050
making robotic automation
accessible to small and medium

7
00:00:21,050 --> 00:00:24,934
sized manufacturers across the
United States. They're not just

8
00:00:24,934 --> 00:00:27,878
selling robots. They're
providing fully managed

9
00:00:27,878 --> 00:00:31,762
solutions that are transforming
how factories operate. In this

10
00:00:31,762 --> 00:00:35,396
episode, we'll explore how
Formic is using cutting edge AI

11
00:00:35,396 --> 00:00:39,030
to streamline robot deployment,
why focusing on uptime and

12
00:00:39,030 --> 00:00:42,663
utilization is critical for
success, the challenges facing

13
00:00:42,663 --> 00:00:46,422
American manufacturing and how
robotics can help. If you are

14
00:00:46,422 --> 00:00:49,742
interested in robotics, ai, or
the future of American

15
00:00:49,742 --> 00:00:53,000
manufacturing, you're in for a
treat. Let's jump in.

16
00:00:53,000 --> 00:00:55,000
Audrow Nash: Hey Saman, would
you introduce yourself?

17
00:00:55,000 --> 00:00:59,042
Saman Farid: Sure. My name is
Saman Farid. I'm the CEO of

18
00:00:59,042 --> 00:01:03,579
Formic. prior to Formic, I built
and sold two software companies

19
00:01:03,579 --> 00:01:08,117
and then was a, VC for about 10
years where I was investing in a

20
00:01:08,117 --> 00:01:12,726
lot of robotics and ai. and, saw
a lot of challenges with the way

21
00:01:12,726 --> 00:01:16,768
that people are building
robotics technologies and trying

22
00:01:16,768 --> 00:01:20,738
to get them into market. So it
started formic to, really

23
00:01:20,738 --> 00:01:24,000
accelerate the, event of the
robot revolution.

24
00:01:24,000 --> 00:01:28,846
Audrow Nash: Hell yeah. Love it.
So tell me a bit about Formic,

25
00:01:28,846 --> 00:01:33,692
the, like general details of it.
How many people, when were you

26
00:01:33,692 --> 00:01:38,462
founded, round of funding that
you guys are at? These kinds of

27
00:01:38,462 --> 00:01:39,000
things?

28
00:01:39,000 --> 00:01:44,027
Saman Farid: Yeah, so we started
Formic, about three and a half

29
00:01:44,027 --> 00:01:48,895
years ago. we are, we've raised
total about $60 million over,

30
00:01:48,895 --> 00:01:53,923
three rounds. and we have about
60 employees right now. We are,

31
00:01:53,923 --> 00:01:58,072
deploying robots in, a variety
of different types of

32
00:01:58,072 --> 00:02:02,701
environments. We are currently
already deployed in about a

33
00:02:02,701 --> 00:02:07,249
hundred manufacturing facilities
across the us, and we're

34
00:02:07,249 --> 00:02:11,000
continuing to scale that
business very quickly.

35
00:02:11,000 --> 00:02:14,389
Audrow Nash: That's awesome.
Yeah. It's so exciting. You guys

36
00:02:14,389 --> 00:02:17,778
are in the, scaling phase right
now. so tell me about Formic.

37
00:02:17,778 --> 00:02:21,056
Tell me what are you guys do,
what problems all you are you

38
00:02:21,056 --> 00:02:23,000
solving? how does your system
work?

39
00:02:23,000 --> 00:02:28,146
Saman Farid: Yeah. what we do at
Formic is we try really hard to

40
00:02:28,146 --> 00:02:32,649
improve the accessibility of
robotics and automation for

41
00:02:32,649 --> 00:02:37,715
manufacturing facilities. what
that means is, we drive down all

42
00:02:37,715 --> 00:02:42,138
of the peripheral costs related
to deploying a robot by

43
00:02:42,138 --> 00:02:46,802
automating them with ai. So that
means automating the site

44
00:02:46,802 --> 00:02:51,868
evaluation process, automating
the configuration and deployment

45
00:02:51,868 --> 00:02:56,854
process, operating all of the
kind of post deployment, service

46
00:02:56,854 --> 00:03:01,276
and maintenance processes, so
that we can provide, full

47
00:03:01,276 --> 00:03:06,342
end-to-end, drop-in robots, that
customers pay for by the hour.

48
00:03:06,342 --> 00:03:11,167
and what that allows them to do
is, not have to become robot

49
00:03:11,167 --> 00:03:16,394
experts themselves. a lot of our
customers, can show up and just,

50
00:03:16,394 --> 00:03:20,897
quickly access, the best,
possible robots for as long as

51
00:03:20,897 --> 00:03:25,722
they need, whether that's one
month or 10 years. and, we can

52
00:03:25,722 --> 00:03:30,225
solve a bunch of their problems,
very quickly. in a more

53
00:03:30,225 --> 00:03:35,452
practical sense, we. are able to
use technology to drive down the

54
00:03:35,452 --> 00:03:40,357
cost and improve the speed of
deploying robots by about seven

55
00:03:40,357 --> 00:03:41,000
to 10 x.

56
00:03:41,000 --> 00:03:42,000
Audrow Nash: Wow.

57
00:03:42,000 --> 00:03:46,319
Saman Farid: we're able to, turn
that into kind of instant

58
00:03:46,319 --> 00:03:49,000
operational deployment. Speed.
Yeah.

59
00:03:49,000 --> 00:03:52,142
Audrow Nash: how do you,
benchmark that? Like how, do you

60
00:03:52,142 --> 00:03:55,118
get seven to 10 x and that's
bonkers. Big. So 700 to a

61
00:03:55,118 --> 00:03:56,000
thousand percent

62
00:03:56,000 --> 00:04:01,000
Saman Farid: That's right. Yeah.
Yeah.

63
00:04:01,000 --> 00:04:04,825
Saman Farid: there's a lot of
stats around it, but for,

64
00:04:04,825 --> 00:04:09,207
currently in the United States
for a typical, robot deployment,

65
00:04:09,207 --> 00:04:13,589
The, timeline from decide to do
a, robot deployment to actually

66
00:04:13,589 --> 00:04:17,275
see the system installed is
typically on the order of

67
00:04:17,275 --> 00:04:21,379
magnitude of four to six months.
all of the steps, involved

68
00:04:21,379 --> 00:04:24,717
related to that kind of site
preparation, system

69
00:04:24,717 --> 00:04:28,403
installation, procuring all the
parts and components,

70
00:04:28,403 --> 00:04:32,785
programming, things like that.
so we're, we are typically doing

71
00:04:32,785 --> 00:04:36,680
that in about two weeks. so
that's a massive savings, on

72
00:04:36,680 --> 00:04:40,853
kind of deployment speed. in
terms of costs, it's, there's a

73
00:04:40,853 --> 00:04:45,096
lot more kind of fluctuation
depending on the type of system.

74
00:04:45,096 --> 00:04:49,408
the benchmark that we use is,
similar systems. and, we're able

75
00:04:49,408 --> 00:04:53,651
to. as, you're, sure familiar
for a typical, robot deployment

76
00:04:53,651 --> 00:04:57,406
in the us, 60 to 70% of the cost
involved is in custom

77
00:04:57,406 --> 00:05:01,023
engineering services. that
custom engineering can be

78
00:05:01,023 --> 00:05:05,057
anything, right? It can be
figuring out what the thickness

79
00:05:05,057 --> 00:05:09,578
of the concrete is that needs to
go, beneath the system. It could

80
00:05:09,578 --> 00:05:13,543
be configuring end of arm
tooling. it could be, designing

81
00:05:13,543 --> 00:05:17,368
all the safety configurations
and interlocks and safety

82
00:05:17,368 --> 00:05:21,541
fencing and where the safety
scanners need to be. And all of

83
00:05:21,541 --> 00:05:25,853
them are site specific. So these
are not things that you could

84
00:05:25,853 --> 00:05:30,026
say, oh, I have a AI powered
robot that's much smarter, so I

85
00:05:30,026 --> 00:05:34,338
don't need that anymore. Like,
all of those things regardless,

86
00:05:34,338 --> 00:05:38,651
need to happen. and we are able
to reduce that portion, of the

87
00:05:38,651 --> 00:05:43,102
deployment to near zero, because
we've automated most of it, not

88
00:05:43,102 --> 00:05:46,997
by making the robot more
intelligent, but by focusing on

89
00:05:46,997 --> 00:05:50,892
those specific processes where
there's a lot of time and

90
00:05:50,892 --> 00:05:55,204
energy, that needs to go into,
robot deployment. those are the

91
00:05:55,204 --> 00:05:59,099
kind of the two axes that we
think about. the other, the

92
00:05:59,099 --> 00:06:02,924
third one is total cost of
ownership of a robot. So the

93
00:06:02,924 --> 00:06:06,611
kind of initial upfront
deployment cost of a robot is

94
00:06:06,611 --> 00:06:10,923
typically, only one portion of,
your, if you're, a factory and

95
00:06:10,923 --> 00:06:15,165
you wanna deploy robots, the
upfront cost is one portion. the

96
00:06:15,165 --> 00:06:19,408
really big cost is actually the
ongoing, service and support,

97
00:06:19,408 --> 00:06:23,790
right? your needs change, your
programming needs to change your

98
00:06:23,790 --> 00:06:27,546
configuration, hardware
configuration needs to change,

99
00:06:27,546 --> 00:06:31,997
your safety configuration needs
to change. And then robots, wear

100
00:06:31,997 --> 00:06:36,448
and tear is natural, right? And
as those, all those things start

101
00:06:36,448 --> 00:06:40,830
to happen, there's a huge cost
that's typically incurred on the

102
00:06:40,830 --> 00:06:45,073
order of magnitude of 10 to 15%
of the cost of the system per

103
00:06:45,073 --> 00:06:49,524
year, to keep that system up and
running in a kind of perform in

104
00:06:49,524 --> 00:06:53,836
a performative state. one of the
areas where, all the software

105
00:06:53,836 --> 00:06:57,383
and tooling and kind of
operational excellence that

106
00:06:57,383 --> 00:07:01,000
we've built, drives that down
significantly as well.

107
00:07:01,000 --> 00:07:04,788
Audrow Nash: That's really cool.
How do you expect, so when you

108
00:07:04,788 --> 00:07:08,576
were saying, like safety related
things might change, what does

109
00:07:08,576 --> 00:07:12,123
that mean? Like, why would
safety things change, during the

110
00:07:12,123 --> 00:07:15,671
course of operating? Is it just
like new standards, they're

111
00:07:15,671 --> 00:07:19,098
doing new processes, or how, why
are some of these things

112
00:07:19,098 --> 00:07:20,000
changing? I sus

113
00:07:20,000 --> 00:07:23,763
Saman Farid: Yeah, so like the,
I'll give you a very simple

114
00:07:23,763 --> 00:07:27,908
example. if you are just picking
up boxes, if you're picking up a

115
00:07:27,908 --> 00:07:31,734
10 pound box, over and over
again, there's a certain kind of

116
00:07:31,734 --> 00:07:35,624
safety regime associated with
that. if you're picking up a 40

117
00:07:35,624 --> 00:07:39,642
pound box or a 60 pound box or
80 pound box, just that, even if

118
00:07:39,642 --> 00:07:43,341
it's the exact same operation
and the exact same movement,

119
00:07:43,341 --> 00:07:46,720
the, safety constraints
associated with that suddenly

120
00:07:46,720 --> 00:07:50,483
change a lot. because if that
box drops when, let's say the

121
00:07:50,483 --> 00:07:54,118
kind of, compressed air, stops
coming to that system. The

122
00:07:54,118 --> 00:07:57,881
system loses a vacuum and that
box drops, the damage that's

123
00:07:57,881 --> 00:08:01,707
caused by 40 pound bucks or 10
pound box are very different.

124
00:08:01,707 --> 00:08:05,533
that's a very trivial and small
example, but, the production

125
00:08:05,533 --> 00:08:09,104
needs in a manufacturing
facility are not static, right?

126
00:08:09,104 --> 00:08:13,186
There's always things that come
up. And, as those configurations

127
00:08:13,186 --> 00:08:16,374
need to develop, you need to
readjust, you need to

128
00:08:16,374 --> 00:08:20,265
recalculate all of your, safe
distance calculations. You need

129
00:08:20,265 --> 00:08:23,836
to, figure out what happens. if
somebody enters the cell

130
00:08:23,836 --> 00:08:27,853
prematurely, does the, robot go
into emergency stop? Or does it

131
00:08:27,853 --> 00:08:31,552
slow down? How does it slow
down? what are the things that

132
00:08:31,552 --> 00:08:35,315
it's interlocked to? What are
the safe, what are the safety

133
00:08:35,315 --> 00:08:39,205
scan scanners that you need to
have in order to support that?

134
00:08:39,205 --> 00:08:42,840
there's all these things that
need to, be addressed. And,

135
00:08:42,840 --> 00:08:46,539
there, yeah, like the maximum
speed that robot can operate

136
00:08:46,539 --> 00:08:50,237
will be, different. if you're
stack, let's just say you're

137
00:08:50,237 --> 00:08:54,255
stacking that box, into a pallet
configuration, the height, the

138
00:08:54,255 --> 00:08:58,209
maximum height that you're able
to reach, will have to change.

139
00:08:58,209 --> 00:09:01,653
So there's all these things
that, are just fundamental

140
00:09:01,653 --> 00:09:05,543
physics related things that you
have to address. And, I think

141
00:09:05,543 --> 00:09:09,433
that there's a lot of optimism
around AI powered smart robots

142
00:09:09,433 --> 00:09:13,450
that will solve all of that for
you. Unfortunately, like that's

143
00:09:13,450 --> 00:09:17,341
not really the case, right? I
think, the AI on the robot will

144
00:09:17,341 --> 00:09:20,976
make it easier and easier to
program the robot to do that

145
00:09:20,976 --> 00:09:25,121
task. but at the end of the day,
there's still all of these other

146
00:09:25,121 --> 00:09:29,075
considerations that go beyond,
can I make robot move from A to

147
00:09:29,075 --> 00:09:32,773
B? From A to B? there's all
these other con considerations

148
00:09:32,773 --> 00:09:36,281
that need to be taken into
account. what we've found is

149
00:09:36,281 --> 00:09:39,916
that there's not a lot of,
emphasis being placed on those

150
00:09:39,916 --> 00:09:41,000
things right now.

151
00:09:41,000 --> 00:09:46,000
Audrow Nash: Yeah. And so that
sounds really cool.

152
00:09:46,000 --> 00:09:49,669
Audrow Nash: before we get into
all of the really cool things

153
00:09:49,669 --> 00:09:53,278
you guys are doing regarding
configuration, just to make the

154
00:09:53,278 --> 00:09:57,068
whole thing a bit more tangible,
do you wanna tell me some case

155
00:09:57,068 --> 00:10:00,737
studies, that show the diverse
sets of problems that you guys

156
00:10:00,737 --> 00:10:02,000
are helping to solve?

157
00:10:02,000 --> 00:10:06,213
Saman Farid: Yeah, so we're,
we're deployed right now in,

158
00:10:06,213 --> 00:10:10,648
like I mentioned earlier, about
a hundred plus manufacturing

159
00:10:10,648 --> 00:10:15,157
facilities. one of the things
I'm very proud of is, about 70%

160
00:10:15,157 --> 00:10:19,000
of those customers, never had
robots, before Formic.

161
00:10:19,000 --> 00:10:22,000
Audrow Nash: That's sick. Hell
yeah.

162
00:10:22,000 --> 00:10:26,000
Saman Farid: yeah, I think we're
really trying to, increase

163
00:10:26,000 --> 00:10:28,000
Audrow Nash: by that 30% too.

164
00:10:28,000 --> 00:10:30,000
Saman Farid: Yeah, there, there
are definitely some that

165
00:10:30,000 --> 00:10:31,000
Audrow Nash: nice.

166
00:10:31,000 --> 00:10:34,949
Saman Farid: it's, the, story of
that 30% is actually a little

167
00:10:34,949 --> 00:10:38,643
bit more sad than the 70%,
because typically the ones that

168
00:10:38,643 --> 00:10:41,000
have had robots in the past,
actually

169
00:10:41,000 --> 00:10:42,000
Audrow Nash: been a bummer.

170
00:10:41,000 --> 00:10:45,386
Saman Farid: yeah, they have a
robot graveyard, in the corner

171
00:10:45,386 --> 00:10:49,700
somewhere. That's a source of
embarrassment. because they've

172
00:10:49,700 --> 00:10:54,015
deployed some robots, some,
halfway through they didn't work

173
00:10:54,015 --> 00:10:58,545
or they spent a bunch of money
on 'em and they ended up getting

174
00:10:58,545 --> 00:11:03,075
stuck with something that didn't
work. there were very few that

175
00:11:03,075 --> 00:11:07,173
have high functioning robots,
when we arrived. but that's

176
00:11:07,173 --> 00:11:11,487
representative of kind of all of
us manufacturing and 90% of

177
00:11:11,487 --> 00:11:15,802
American factories don't have,
even a single robot. We have.

178
00:11:15,802 --> 00:11:20,116
Yeah, it's, unbelievable. And I
think the main thing is also

179
00:11:20,116 --> 00:11:23,999
it's, untenable, right? I think
for America to have, a

180
00:11:23,999 --> 00:11:28,672
manufacturing base that sustains
through, the 21st century and is

181
00:11:28,672 --> 00:11:32,915
competitive, we are drastically
under tooled, right? These,

182
00:11:32,915 --> 00:11:35,000
manufacturers are stuck with,

183
00:11:35,000 --> 00:11:36,000
Audrow Nash: percent.

184
00:11:35,000 --> 00:11:39,430
Saman Farid: really, high labor
costs, really low overall

185
00:11:39,430 --> 00:11:44,171
utilization of these facilities,
and actually just completely

186
00:11:44,171 --> 00:11:49,145
labor in, unavailability. these
fa these facilities are, really,

187
00:11:49,145 --> 00:11:50,000
struggling.

188
00:11:50,000 --> 00:11:54,226
Audrow Nash: Yeah, for sure. And
I'm, we'll talk about that for

189
00:11:54,226 --> 00:11:57,915
sure. 'cause yeah, labor
shortages are such a pervasive

190
00:11:57,915 --> 00:12:01,671
thing. but that I think are
driving a lot of interest in

191
00:12:01,671 --> 00:12:05,963
robotics, but, okay, so you have
a hundred, roughly a hundred of

192
00:12:05,963 --> 00:12:09,585
these. Manufacturing like
companies or what, a hundred

193
00:12:09,585 --> 00:12:12,000
sites or how are we referring to
it?

194
00:12:12,000 --> 00:12:14,000
Saman Farid: Yeah. Facilities,
yeah. Sites.

195
00:12:14,000 --> 00:12:16,895
Audrow Nash: A hundred
facilities. and so what are some

196
00:12:16,895 --> 00:12:20,000
of the examples of what you're
actually doing? What are you

197
00:12:20,000 --> 00:12:22,000
helping them with in those
facilities?

198
00:12:22,000 --> 00:12:26,709
Saman Farid: so our biggest
segment is, producing consumer

199
00:12:26,709 --> 00:12:31,662
packaged goods. so we're helping
robots that we're putting in

200
00:12:31,662 --> 00:12:36,859
robots that help produce things
like, chocolate chip cookies and

201
00:12:36,859 --> 00:12:41,731
tortillas and matcha powder for
Starbucks and, dog food. our

202
00:12:41,731 --> 00:12:46,603
robots make windshield wiper
fluid, all kinds of things that

203
00:12:46,603 --> 00:12:51,312
are, very commonly found in,
target or CVS or Walgreens or

204
00:12:51,312 --> 00:12:56,021
Walmart. so those types of
products, have, upfront kind of

205
00:12:56,021 --> 00:13:01,137
food production part of it, and
then a lot of packaging related

206
00:13:01,137 --> 00:13:05,603
work that needs to happen. So
putting items into boxes,

207
00:13:05,603 --> 00:13:10,474
putting boxes onto pallets,
moving box, moving pallets, into

208
00:13:10,474 --> 00:13:15,427
trucks or into, shelving units.
those are all. Parts that are

209
00:13:15,427 --> 00:13:20,462
very highly labor intensive. so
we do a lot of tho those kinds

210
00:13:20,462 --> 00:13:24,603
of things with robots. big
category for us is metal

211
00:13:24,603 --> 00:13:29,718
fabrication. and we have a lot
of facilities that we serve that

212
00:13:29,718 --> 00:13:34,996
make, components for automotive,
aerospace, and defense. you know

213
00:13:34,996 --> 00:13:40,111
what a lot of people don't often
realize is that, for every one

214
00:13:40,111 --> 00:13:44,739
kind of GM or Ford plant, there
are typically, 200 to 500

215
00:13:44,739 --> 00:13:49,368
suppliers, that have to have
facilities that make all the

216
00:13:49,368 --> 00:13:54,239
components that go to the GM and
or Ford plant for the final

217
00:13:54,239 --> 00:13:57,000
assembly. and there's this kind
of

218
00:13:57,000 --> 00:13:59,444
Audrow Nash: Don't those
suppliers also have suppliers?

219
00:13:59,444 --> 00:14:01,000
Like it just keeps going for a
long

220
00:14:01,000 --> 00:14:04,698
Saman Farid: without a doubt.
Without a doubt. and in

221
00:14:04,698 --> 00:14:08,884
aerospace it's even more, right?
So for every single, Boeing

222
00:14:08,884 --> 00:14:12,651
plant, for example, you would
have, about two to 3000,

223
00:14:12,651 --> 00:14:16,000
immediate suppliers, that are
making components.

224
00:14:16,000 --> 00:14:19,532
Audrow Nash: Our world is
unbelievably globalized. It's

225
00:14:19,532 --> 00:14:23,642
just the craziest thing. with so
many, I dunno, factories making

226
00:14:23,642 --> 00:14:27,688
so many specialized things and
the networks and efficiencies of

227
00:14:27,688 --> 00:14:30,000
scale, it's absolutely
unbelievable.

228
00:14:30,000 --> 00:14:36,284
Saman Farid: yeah. I love that.
I don't know if you've, have you

229
00:14:36,284 --> 00:14:42,568
ever heard of this, this example
that like, nobody, in the world

230
00:14:42,568 --> 00:14:48,066
today knows how to make, a
pencil? I. and, let, alone, a

231
00:14:48,066 --> 00:14:54,252
more complicated product, like a
car or a, or airplane. I think

232
00:14:54,252 --> 00:14:59,751
we are in a world where we're
inevitably interdependent,

233
00:14:59,751 --> 00:15:05,642
right? I think anybody who spent
time in manufacturing knows

234
00:15:05,642 --> 00:15:11,337
that, and, you can't just,
improve one part of the, of the

235
00:15:11,337 --> 00:15:17,130
supply chain and say, oh, yeah,
we, solved that bottleneck.

236
00:15:17,130 --> 00:15:22,825
that's not how it works, right?
you have to put, provide a

237
00:15:22,825 --> 00:15:28,421
solution that all those, few
hundred suppliers and all of

238
00:15:28,421 --> 00:15:34,215
their suppliers can use if you
really wanna see sustainable

239
00:15:34,215 --> 00:15:35,000
benefit.

240
00:15:35,000 --> 00:15:38,416
Audrow Nash: yep. Okay. So it's,
so when you would do metal

241
00:15:38,416 --> 00:15:41,891
fabrication, what? Parts are you
involved with? are you just

242
00:15:41,891 --> 00:15:45,423
moving the metal from here and
there? Are you putting it into

243
00:15:45,423 --> 00:15:48,782
bending machines? Are you
cutting it? Are you like, where,

244
00:15:48,782 --> 00:15:52,372
what are the specific tasks that
your robots are helping with?

245
00:15:52,372 --> 00:15:55,846
Because it sounds like there's
quite a bit of like logistics

246
00:15:55,846 --> 00:15:59,379
related things that it's you're
helping with, but I wonder in

247
00:15:59,379 --> 00:16:01,000
some of these other domains.

248
00:16:01,000 --> 00:16:05,885
Saman Farid: So in CPG, a lot of
it, a lot of it is packaging and

249
00:16:05,885 --> 00:16:10,544
logistics. in metal fabrication,
the most common use cases are

250
00:16:10,544 --> 00:16:15,128
machine tending. and so that
could be, tending a CNC lathe, a

251
00:16:15,128 --> 00:16:19,487
CNC milling machine, a press,
which is doing, the bending,

252
00:16:19,487 --> 00:16:23,395
motions, welding, and
inspection. these are all kind

253
00:16:23,395 --> 00:16:28,054
of very common use cases. for
example, one of our deployments,

254
00:16:28,054 --> 00:16:32,337
they're making, metal parts that
end up in, in a Chrysler

255
00:16:32,337 --> 00:16:36,546
vehicle. and, they have to take
thousands a day of these

256
00:16:36,546 --> 00:16:41,130
individual components. Cut them
to the right length, put them

257
00:16:41,130 --> 00:16:45,939
into a CNC machine, let the CNC
machine run a certain operation,

258
00:16:45,939 --> 00:16:50,599
then they take out the part and
then put it into a lathe, have

259
00:16:50,599 --> 00:16:55,108
it do a secondary operation, and
then take it out. debur it,

260
00:16:55,108 --> 00:16:59,917
inspect it and then put it into,
a box for kind of the next step

261
00:16:59,917 --> 00:17:04,351
of the manufacturing process
that historically used to be a

262
00:17:04,351 --> 00:17:08,409
person that would stand there
all day and kind of pick

263
00:17:08,409 --> 00:17:12,843
something up, put it over there,
take it out, put it there.

264
00:17:12,843 --> 00:17:17,202
there's this kind of very
repetitive process that was also

265
00:17:17,202 --> 00:17:21,560
fraught with a lot of error.
and, and it was also very low

266
00:17:21,560 --> 00:17:26,370
utilization because imagine that
process. You actually have, six

267
00:17:26,370 --> 00:17:29,000
or seven machines that are tied
up,

268
00:17:29,000 --> 00:17:31,000
Audrow Nash: Yep. They're not
being used in parallel.

269
00:17:31,000 --> 00:17:35,326
Saman Farid: Yeah. And, yeah,
they're not being used in

270
00:17:35,326 --> 00:17:40,438
parallel and you're very limited
by the schedule of that, of the,

271
00:17:40,438 --> 00:17:44,528
of when you have labor
available. and in, in typical

272
00:17:44,528 --> 00:17:48,618
American manufacturing
facilities, especially in the

273
00:17:48,618 --> 00:17:53,101
metal fabrication world, they
run one shift a day. So one

274
00:17:53,101 --> 00:17:57,191
shift a day means they're
running about 175 hours, a

275
00:17:57,191 --> 00:18:01,674
month, which is, about 30% of
the or, depending on how it

276
00:18:01,674 --> 00:18:06,000
calculated 20 to 30% of the
available hours, in a given

277
00:18:06,000 --> 00:18:09,539
Audrow Nash: you're saying for
an individual, like the math you

278
00:18:09,539 --> 00:18:13,022
did was 40 times four and a half
to get one seven or something

279
00:18:13,022 --> 00:18:16,000
like that, and that's one person
would work 175 hours

280
00:18:16,000 --> 00:18:17,000
Saman Farid: One, one

281
00:18:17,000 --> 00:18:18,000
Audrow Nash: kind of thing, but

282
00:18:18,000 --> 00:18:19,000
Saman Farid: because a lot of
these

283
00:18:19,000 --> 00:18:21,000
Audrow Nash: shifts or something

284
00:18:21,000 --> 00:18:22,000
Saman Farid: but yes,

285
00:18:22,000 --> 00:18:25,000
Audrow Nash: it was labor, if
labor could sustain it.

286
00:18:25,000 --> 00:18:28,571
Saman Farid: Exactly right.
There's not, enough labor

287
00:18:28,571 --> 00:18:32,411
available. and so most American
manufacturing facilities,

288
00:18:32,411 --> 00:18:36,048
especially most metal
fabrication facilities in the US

289
00:18:36,048 --> 00:18:40,360
do run one shift a day. so what
that means is, like there's just

290
00:18:40,360 --> 00:18:44,335
massive under utilization.
They're se 75% of the time these

291
00:18:44,335 --> 00:18:48,175
facilities are sitting idle. All
of those machines that I

292
00:18:48,175 --> 00:18:52,554
mentioned, the CNC lay, then the
mill, the inspection process and

293
00:18:52,554 --> 00:18:56,798
the de debriefing machine, all
of those are sitting idle 75% of

294
00:18:56,798 --> 00:18:57,000
the

295
00:18:57,000 --> 00:19:00,000
Audrow Nash: Most of the time.
Yeah. That's wild.

296
00:19:00,000 --> 00:19:03,861
Saman Farid: when we put that
robot in there. What happened

297
00:19:03,861 --> 00:19:07,656
was, step one, we helped fill an
empty headcount that they

298
00:19:07,656 --> 00:19:11,452
couldn't hire somebody for. So
win already. But, step two,

299
00:19:11,452 --> 00:19:15,509
which was, even better, was then
they realized, actually I can

300
00:19:15,509 --> 00:19:19,305
run this robot 24 7. So they
would, they would concentrate

301
00:19:19,305 --> 00:19:22,969
their labor on all the upstream
processes to prepare the

302
00:19:22,969 --> 00:19:26,961
materials. They would load up
the robot, before they went to,

303
00:19:26,961 --> 00:19:30,888
to, to bed, or left, for the
day. They would load up a bunch

304
00:19:30,888 --> 00:19:34,683
of blanks. They would go home,
and they'd come back in the

305
00:19:34,683 --> 00:19:38,282
morning and the robot had run
another 10 to 15 hours of

306
00:19:38,282 --> 00:19:42,470
additional work. So now all the
machines got higher utilization.

307
00:19:42,470 --> 00:19:46,397
the entire facility, ends up
being much more productive with

308
00:19:46,397 --> 00:19:50,192
the same amount of labor. and
their part quality improved.

309
00:19:50,192 --> 00:19:53,661
They're, they're able to win
more business with their

310
00:19:53,661 --> 00:19:57,391
customers because now they
actually have more throughput.

311
00:19:57,391 --> 00:20:01,448
So there's all of these kind of
secondary and tertiary effects

312
00:20:01,448 --> 00:20:04,000
that come from. being able to
automate,

313
00:20:04,000 --> 00:20:07,971
Audrow Nash: very much. Do you?
So the way I'm imagining it is,

314
00:20:07,971 --> 00:20:11,880
okay, we're understaffed because
we don't have many people. So

315
00:20:11,880 --> 00:20:15,347
then what we can do, like
there's not enough people for

316
00:20:15,347 --> 00:20:18,940
all the labor, all the, jobs
that we need. So now in this

317
00:20:18,940 --> 00:20:22,659
machine tending example, you
have a person who is moving it

318
00:20:22,659 --> 00:20:26,567
from one machine to the next and
probably, I wonder if they're

319
00:20:26,567 --> 00:20:29,845
doing like. So you start on
machine one, it does the

320
00:20:29,845 --> 00:20:33,312
process. You move the part to
machine two. Do you start

321
00:20:33,312 --> 00:20:37,284
another pro, part on machine one
and you just keep them moving?

322
00:20:37,284 --> 00:20:40,940
So every machine is as fully
utilized as possible? Or does

323
00:20:40,940 --> 00:20:44,911
the person go and complete the
part than start over? 'cause I'm

324
00:20:44,911 --> 00:20:48,000
imagining how robots could be
more efficient, but

325
00:20:48,000 --> 00:20:51,945
Saman Farid: so you're supposed
to do what you just described,

326
00:20:51,945 --> 00:20:55,000
which is load the second
machine. Load the first

327
00:20:55,000 --> 00:20:56,000
Audrow Nash: As soon as the
first one.

328
00:20:56,000 --> 00:20:57,000
Saman Farid: Yeah. And have them

329
00:20:57,000 --> 00:20:59,762
Audrow Nash: So everything is
working as much as possible. So

330
00:20:59,762 --> 00:21:02,615
you have one in operation, then
two, then three, then four, and

331
00:21:02,615 --> 00:21:05,332
then you have the max number of
machines. hand. Say you have

332
00:21:05,332 --> 00:21:07,958
seven machines, you have seven
parts now in circulation as

333
00:21:07,958 --> 00:21:09,000
they're moving through.

334
00:21:09,000 --> 00:21:13,872
Saman Farid: But the in
practice, the typical American

335
00:21:13,872 --> 00:21:19,105
manufacturing facility has a
200% per year, labor turnover

336
00:21:19,105 --> 00:21:21,000
rate, meaning people,

337
00:21:21,000 --> 00:21:23,818
Audrow Nash: as extreme as other
industries, but that's pretty

338
00:21:23,818 --> 00:21:24,000
bad.

339
00:21:24,000 --> 00:21:28,177
Saman Farid: Yeah, for every
headcount they're hiring twice a

340
00:21:28,177 --> 00:21:32,148
year. which means that the
person who's standing, in front

341
00:21:32,148 --> 00:21:36,393
of that machine, probably was
only working for the last two or

342
00:21:36,393 --> 00:21:40,639
three months, in that, at the
most. and most likely they're at

343
00:21:40,639 --> 00:21:44,884
the beginning of their journey.
So what that means is to train

344
00:21:44,884 --> 00:21:48,924
somebody to do what you're
describing, which is oh, let me,

345
00:21:48,924 --> 00:21:52,895
make sure that every system is
at max. Utilization is very

346
00:21:52,895 --> 00:21:57,003
hard, and most of the time they
end up being sequential. and

347
00:21:57,003 --> 00:22:01,111
there's just, so many things,
about that process that can be

348
00:22:01,111 --> 00:22:05,288
done much better when, a robot
can actually consistently meet

349
00:22:05,288 --> 00:22:07,000
that performance through.

350
00:22:07,000 --> 00:22:10,129
Audrow Nash: you, so I'm
imagining, okay. You have a

351
00:22:10,129 --> 00:22:13,979
robot, that robot knows roughly
how long every machine that it's

352
00:22:13,979 --> 00:22:17,710
tending takes, and it has a few
of them that it needs to do in

353
00:22:17,710 --> 00:22:21,199
series. You can probably do
pretty intelligent scheduling.

354
00:22:21,199 --> 00:22:24,990
Is this something that you guys
are doing? So you go basically,

355
00:22:24,990 --> 00:22:28,660
okay. I just started this one. I
just started that one before

356
00:22:28,660 --> 00:22:32,089
that. I know when those are
gonna be done. I'm gonna just

357
00:22:32,089 --> 00:22:36,000
you'd have one arm moving around
doing one step at a time. pretty

358
00:22:36,000 --> 00:22:39,704
Saman Farid: all of that
ultimately boils down to what is

359
00:22:39,704 --> 00:22:43,148
your average cycle time, on
every part. and so that's

360
00:22:43,148 --> 00:22:46,852
something we track religiously,
and then is automatically

361
00:22:46,852 --> 00:22:50,296
improved. So we have, we
actually surface all of that

362
00:22:50,296 --> 00:22:54,000
kind realtime production data to
our customer. that shows

363
00:22:54,000 --> 00:22:55,000
Audrow Nash: they love

364
00:22:54,000 --> 00:22:59,602
Saman Farid: exactly what your
kind of part, rate was on every

365
00:22:59,602 --> 00:23:04,663
different component, in your
process. and, being able to

366
00:23:04,663 --> 00:23:09,000
optimize it, is. Like one of
the, one of the key

367
00:23:09,000 --> 00:23:10,000
Audrow Nash: Oh, killer.

368
00:23:10,000 --> 00:23:13,229
Saman Farid: what step do you do
when, do you debu before you

369
00:23:13,229 --> 00:23:16,457
load or unload, right? there's
all these things that you need

370
00:23:16,457 --> 00:23:19,527
to figure out, in terms of what
the sequence of operations

371
00:23:19,527 --> 00:23:22,333
should be. and, yeah,
absolutely. Like you said, with

372
00:23:22,333 --> 00:23:25,614
a robot, like you can create a
recipe, right? And then you can

373
00:23:25,614 --> 00:23:28,790
make sure that you hit it every
single time. The other thing

374
00:23:28,790 --> 00:23:32,019
that, that benefits from this,
right? is part quality, right?

375
00:23:32,019 --> 00:23:35,089
Part quality goes up very
significantly because, it's very

376
00:23:35,089 --> 00:23:38,000
common for operators who are
standing in front of those

377
00:23:38,000 --> 00:23:41,229
machines to load a part in, an
imperfect way. and they either

378
00:23:41,229 --> 00:23:44,299
tighten it too much or too
little, or, they didn't realize

379
00:23:44,299 --> 00:23:47,369
that the tool needs to be
changed or whatever it might be.

380
00:23:47,369 --> 00:23:50,227
There. there's a lot of
different things that you need

381
00:23:50,227 --> 00:23:53,350
to pay attention to. and if
those don't all happen, you end

382
00:23:53,350 --> 00:23:56,684
up with subpar parts. And worst
of all, you often don't realize

383
00:23:56,684 --> 00:23:59,860
you're ending up with subpar
parts until, it's way too late,

384
00:23:59,860 --> 00:24:02,877
right? You've, created a 2000,
parts from today, somebody

385
00:24:02,877 --> 00:24:06,000
checks them and realized, oh,
actually we missed something.

386
00:24:06,000 --> 00:24:09,596
Audrow Nash: yeah. 'cause you're
just doing it the same way over

387
00:24:09,596 --> 00:24:13,079
and over again. with an error in
your process for this kind of

388
00:24:13,079 --> 00:24:16,674
thing. But yeah, in a robot you
could just tell it the exact way

389
00:24:16,674 --> 00:24:20,270
to do it and then it would just
keep repeating that, which would

390
00:24:20,270 --> 00:24:21,000
be very nice.

391
00:24:21,000 --> 00:24:22,000
Saman Farid: That's right.

392
00:24:22,000 --> 00:24:26,218
Audrow Nash: so with the cycle
time, do you have, so it's

393
00:24:26,218 --> 00:24:31,027
really wonderful that the robots
can work 24 7 as opposed to just

394
00:24:31,027 --> 00:24:35,245
one shift, a day. And it's crazy
because people aren't as

395
00:24:35,245 --> 00:24:39,981
efficient 'cause they're getting
trained. 'cause the turnover is

396
00:24:39,981 --> 00:24:44,347
so intense. Do you have any idea
how the cycle time of your

397
00:24:44,347 --> 00:24:48,786
systems, and I'm sure this is
case by case, like application

398
00:24:48,786 --> 00:24:53,448
by application, but do you, have
a range of how your cycle time

399
00:24:53,448 --> 00:24:57,000
compares to, person cycle time
on various tasks?

400
00:24:57,000 --> 00:25:01,911
Saman Farid: Yeah. what we're
seeing today is on average, 30%.

401
00:25:01,911 --> 00:25:05,000
th 30 to 40% improvement in
cycle time.

402
00:25:05,000 --> 00:25:06,000
Audrow Nash: Wow.

403
00:25:06,000 --> 00:25:09,886
Saman Farid: but, that
improvement is generally the

404
00:25:09,886 --> 00:25:14,457
smaller one of the two. The
main, the, bigger improvement is

405
00:25:14,457 --> 00:25:19,257
the increase utilization, right?
the fact that you can just run

406
00:25:19,257 --> 00:25:23,829
it for more hours in the day,
because, like you, what you're

407
00:25:23,829 --> 00:25:28,324
doing there is, you're getting
kind of 200% benefit or 300%

408
00:25:28,324 --> 00:25:30,000
benefit if you go from

409
00:25:30,000 --> 00:25:31,000
Audrow Nash: Oh yeah, totally.

410
00:25:31,000 --> 00:25:33,000
Saman Farid: shifts or three
shifts.

411
00:25:33,000 --> 00:25:38,088
Audrow Nash: Yeah, I'm imagining
130% of, the products are being

412
00:25:38,088 --> 00:25:43,017
produced and then you're also
doing it for three times as much

413
00:25:43,017 --> 00:25:47,548
time. So that's 400, I don't
know, 400% roughly, which is

414
00:25:47,548 --> 00:25:52,000
bonkers. That's really awesome.
As a, as an improvement.

415
00:25:52,000 --> 00:25:57,176
Saman Farid: Yeah. And,
typically right, the constraints

416
00:25:57,176 --> 00:26:03,000
once you do that are, upstream
and downstream. Because you, you

417
00:26:03,000 --> 00:26:05,884
Audrow Nash: You hit some other
bottleneck. 'cause this part's

418
00:26:05,884 --> 00:26:07,000
highly optimized. I see.

419
00:26:07,000 --> 00:26:08,000
Saman Farid: yeah, Exactly.

420
00:26:08,000 --> 00:26:12,037
Audrow Nash: But then you can
probably, grow out to those if

421
00:26:12,037 --> 00:26:15,872
the types of tasks are well
suited for what you're doing.

422
00:26:15,872 --> 00:26:19,774
'cause it sounds like, I'm
imagining the path of you guys,

423
00:26:19,774 --> 00:26:23,474
maybe you started in logistics
and then grew into these

424
00:26:23,474 --> 00:26:27,376
manufacturing tasks. 'cause a
lot of it's just, assembling

425
00:26:27,376 --> 00:26:30,000
things in a sense. like
assembling your

426
00:26:30,000 --> 00:26:33,439
Saman Farid: We, actually were
very focused on manufacturing

427
00:26:33,439 --> 00:26:36,994
from the beginning. didn't, we
tried not to do too much on the

428
00:26:36,994 --> 00:26:39,000
logistics world, right? But I
think

429
00:26:39,000 --> 00:26:41,000
Audrow Nash: because there are
many companies doing it.

430
00:26:41,000 --> 00:26:44,992
Saman Farid: There's already a
lot of automation there. And

431
00:26:44,992 --> 00:26:49,389
also, in the logistics industry,
the adoption rate of robotics is

432
00:26:49,389 --> 00:26:53,245
generally also quite a lot
higher. and we felt if we were

433
00:26:53,245 --> 00:26:57,440
gonna have a bigger impact, we
wanted to focus on the industry

434
00:26:57,440 --> 00:27:01,567
where, like just robots make
such a bigger difference, right?

435
00:27:01,567 --> 00:27:05,220
the labor availability in a
warehouse is actually much

436
00:27:05,220 --> 00:27:08,805
higher than the labor
availability in a manufacturing

437
00:27:08,805 --> 00:27:12,121
facility because these are
typically much harsher

438
00:27:12,121 --> 00:27:13,000
environments.

439
00:27:13,000 --> 00:27:14,000
Audrow Nash: Oh,

440
00:27:13,000 --> 00:27:15,000
Saman Farid: yeah. Yeah.

441
00:27:15,000 --> 00:27:19,547
Audrow Nash: very interesting.
But, so the first set of things

442
00:27:19,547 --> 00:27:23,800
in the consumer packaged goods
sounded more like logistics

443
00:27:23,800 --> 00:27:26,000
tasks. Did you grow into that?

444
00:27:26,000 --> 00:27:27,000
Saman Farid: yeah,

445
00:27:27,000 --> 00:27:28,000
Audrow Nash: you, end up doing
those tasks?

446
00:27:28,000 --> 00:27:30,667
Saman Farid: yeah. So it's not
even, those tasks are not

447
00:27:30,667 --> 00:27:33,000
strictly logistics in the sense
that we're doing,

448
00:27:33,000 --> 00:27:37,000
Audrow Nash: All right. I'm just
thinking palletizing relating.

449
00:27:37,000 --> 00:27:38,000
Saman Farid: So palletizing

450
00:27:38,000 --> 00:27:41,000
Audrow Nash: think of that very
much as logistics

451
00:27:41,000 --> 00:27:44,578
Saman Farid: Yeah. Yeah. we,
yeah, I think by that

452
00:27:44,578 --> 00:27:48,872
definition, that would be a
logistics task, although, like a

453
00:27:48,872 --> 00:27:52,736
lot of the other processes that
we do in manufacturing

454
00:27:52,736 --> 00:27:56,386
facilities, there's, case
packing, and palletizing.

455
00:27:56,386 --> 00:28:00,537
there's, also a different
inspection applications. There's

456
00:28:00,537 --> 00:28:04,831
depalletizing for raw materials
coming in. but you're right,

457
00:28:04,831 --> 00:28:09,053
generally what we, call those
are material handling, tasks.

458
00:28:09,053 --> 00:28:12,488
because when I think of
logistics, generally I'm

459
00:28:12,488 --> 00:28:16,925
thinking about a ASRS systems
or, typically what they're doing

460
00:28:16,925 --> 00:28:19,000
is moving pallets around from

461
00:28:19,000 --> 00:28:21,000
Audrow Nash: I don't know what
the, ASRS.

462
00:28:21,000 --> 00:28:24,486
Saman Farid: automated storage
and retrieval system. So those

463
00:28:24,486 --> 00:28:27,743
are like the kind of giant
shelving units where, you say,

464
00:28:27,743 --> 00:28:31,171
Hey, I want, a toothbrush, and
then, the robot will go up to

465
00:28:31,171 --> 00:28:33,000
the right shelf and pull it out.

466
00:28:33,000 --> 00:28:35,000
Audrow Nash: Oh yeah. Okay.

467
00:28:35,000 --> 00:28:39,318
Audrow Nash: So what are the,
core technologies that you,

468
00:28:39,318 --> 00:28:43,333
before we get to the
configuration side, what are the

469
00:28:43,333 --> 00:28:48,030
core technologies in your robot
operation, that you guys have,

470
00:28:48,030 --> 00:28:52,727
like core capabilities that you
end up using in various places

471
00:28:52,727 --> 00:28:55,000
across all of your facilities?

472
00:28:55,000 --> 00:28:59,249
Saman Farid: Yeah. I'll break it
down a little bit by phase. so

473
00:28:59,249 --> 00:29:03,565
in the upfront phase, a lot of,
the software that we put to work

474
00:29:03,565 --> 00:29:07,881
is in automating the, gathering
of mission parameters, right? So

475
00:29:07,881 --> 00:29:11,590
if I walk into a factory
tomorrow and, they say, Hey, I

476
00:29:11,590 --> 00:29:15,771
wanna automate this task. the
immediate next thing that has to

477
00:29:15,771 --> 00:29:19,547
happen is I need to ask them
about 300 to 500 questions,

478
00:29:19,547 --> 00:29:23,931
right. It's what's the thickness
of the concrete here? What's the

479
00:29:23,931 --> 00:29:28,247
ceiling height? what's the power
that you have available? Do you

480
00:29:28,247 --> 00:29:30,000
have compressed air? what,

481
00:29:30,000 --> 00:29:33,000
Audrow Nash: Do they always know
those?

482
00:29:33,000 --> 00:29:36,000
Saman Farid: and no. So that's
the other, that's the other

483
00:29:36,000 --> 00:29:37,000
Audrow Nash: thickness,

484
00:29:36,000 --> 00:29:38,000
Saman Farid: Yeah, you,

485
00:29:38,000 --> 00:29:40,990
Audrow Nash: they're renting the
space. I, don't imagine they

486
00:29:40,990 --> 00:29:43,000
know that part of the building,
but Yeah.

487
00:29:43,000 --> 00:29:46,326
Saman Farid: yeah, exactly.
there's the, reality is like, of

488
00:29:46,326 --> 00:29:49,818
the 300 questions that I need to
ask, probably 70% of them they

489
00:29:49,818 --> 00:29:53,254
don't know. and there's, it's
like even, simple things, right?

490
00:29:53,254 --> 00:29:56,247
what are your box sizes and
weights and, what are your

491
00:29:56,247 --> 00:29:59,739
typical rates that you need to
hit? What is the cycle time that

492
00:29:59,739 --> 00:30:03,286
your human is currently hitting
and how do I need, how much do I

493
00:30:03,286 --> 00:30:06,612
need to improve that in order
for the robot to be able to do

494
00:30:06,612 --> 00:30:07,000
better?

495
00:30:07,000 --> 00:30:08,000
Audrow Nash: Oh my gosh.

496
00:30:08,000 --> 00:30:12,164
Saman Farid: just, all of those,
are, just very, difficult to

497
00:30:12,164 --> 00:30:15,781
know because the, your
production is just changing so

498
00:30:15,781 --> 00:30:19,808
fast. So you have a production
plan somewhere on somebody's

499
00:30:19,808 --> 00:30:24,040
laptop, right? But in reality,
by the time that you show up on

500
00:30:24,040 --> 00:30:27,589
that site, like all that
information is outdated and

501
00:30:27,589 --> 00:30:31,889
it's, a, a act of God to go and
gather all of that information.

502
00:30:31,889 --> 00:30:35,917
we have found that, I. for,
robot automation in general, it

503
00:30:35,917 --> 00:30:39,875
doesn't matter what kind of
robot you're trying to deploy.

504
00:30:39,875 --> 00:30:43,902
the first thing is you need to
know what the robot needs to

505
00:30:43,902 --> 00:30:47,793
accomplish. and so gathering,
gathering all that data, is

506
00:30:47,793 --> 00:30:52,025
something that technology can
assist very significantly. So we

507
00:30:52,025 --> 00:30:55,847
use a combination of, like lidar
scanners to get all the

508
00:30:55,847 --> 00:30:59,738
dimensional data about the
facility, within, two minutes.

509
00:30:59,738 --> 00:31:03,697
we get, we use, 2D cameras to
observe some of the existing

510
00:31:03,697 --> 00:31:07,587
processes and automatically
figure out what are all those

511
00:31:07,587 --> 00:31:11,000
manual steps that need to happen
on a typical day.

512
00:31:11,000 --> 00:31:13,938
Audrow Nash: when you say 2D
camera, you just mean like a

513
00:31:13,938 --> 00:31:16,000
normal camera, not a depth
camera. Okay.

514
00:31:16,000 --> 00:31:19,093
Saman Farid: so we put in, a
couple of cameras. We observe

515
00:31:19,093 --> 00:31:22,133
the process, over a certain
period of time. and we gather

516
00:31:22,133 --> 00:31:25,280
most of the application data
that we need around, like what

517
00:31:25,280 --> 00:31:28,320
is the actual operation? What
are, the steps that need to

518
00:31:28,320 --> 00:31:31,680
happen? what are the rates and
timing and things like that. and

519
00:31:31,680 --> 00:31:34,720
then, we've created a bunch of
different tooling for data

520
00:31:34,720 --> 00:31:37,813
ingestion where we can quickly
connect to their ERP or CRM

521
00:31:37,813 --> 00:31:40,960
systems, or whether it's an
Excel file or other kind of UNR

522
00:31:40,960 --> 00:31:43,947
scribbles that are on a piece of
paper. Whatever kind of

523
00:31:43,947 --> 00:31:47,147
structured or unstructured data
that may or may not have. We

524
00:31:47,147 --> 00:31:50,187
feed that into our, engine,
which basically ingest all of

525
00:31:50,187 --> 00:31:53,227
that information and tur and,
builds kinda like a mission

526
00:31:53,227 --> 00:31:56,320
plan, right? what does this
robot need to do? what are the

527
00:31:56,320 --> 00:31:59,680
constraints? and how are we able
to accomplish that? so that's,

528
00:31:59,680 --> 00:32:02,987
there's a lot of, there's a lot
of energy and effort that went

529
00:32:02,987 --> 00:32:04,000
into building that.

530
00:32:04,000 --> 00:32:08,629
Audrow Nash: Yeah. if you had to
guess for this part of it, for

531
00:32:08,629 --> 00:32:13,332
the gathering the data part, is
that like of, say your man hours

532
00:32:13,332 --> 00:32:17,815
or, person hours that have gone
in, would you say it's like a

533
00:32:17,815 --> 00:32:22,371
fairly substantial part? is it
like 50% of the whole effort of

534
00:32:22,371 --> 00:32:27,000
what you guys have done or how
would you, what would you think?

535
00:32:27,000 --> 00:32:30,733
Saman Farid: from r and d
perspective, I would say it's

536
00:32:30,733 --> 00:32:34,738
probably about 30%, of the
effort that our r and d team has

537
00:32:34,738 --> 00:32:39,014
expended, has been on building
this, this set of tooling. yeah,

538
00:32:39,014 --> 00:32:43,425
the, it's not insignificant, but
the, hard part about that, about

539
00:32:43,425 --> 00:32:46,955
building this kind of data
ingestion pipeline is two

540
00:32:46,955 --> 00:32:51,095
things. one is actually building
it right, which is hard, but

541
00:32:51,095 --> 00:32:55,100
really the really, hard thing is
knowing what are the right

542
00:32:55,100 --> 00:32:57,000
things that you need to ask.

543
00:32:57,000 --> 00:32:58,000
Audrow Nash: percent.

544
00:32:58,000 --> 00:33:01,610
Saman Farid: what are the
qualification points for every

545
00:33:01,610 --> 00:33:05,671
type of task? and that requires
a lot of end-to-end experience,

546
00:33:05,671 --> 00:33:09,732
right? You can't design those in
a vacuum. You need to deploy a

547
00:33:09,732 --> 00:33:13,471
bunch of systems and then
realize I should have asked this

548
00:33:13,471 --> 00:33:17,403
question, right early on. and
that would've really saved me a

549
00:33:17,403 --> 00:33:21,271
bunch of time or design effort
or whatever it might be. just

550
00:33:21,271 --> 00:33:24,945
like a, simple thing, right?
what is your ceiling height,

551
00:33:24,945 --> 00:33:28,813
right? if you don't ask that
question, you can very much end

552
00:33:28,813 --> 00:33:32,745
up stuck. the, effort that is
expanded by the kind of RD team

553
00:33:32,745 --> 00:33:36,226
to build it is one part, but
really the effort that is

554
00:33:36,226 --> 00:33:37,000
expanded to,

555
00:33:37,000 --> 00:33:38,000
Audrow Nash: The iteration.

556
00:33:38,000 --> 00:33:41,567
Saman Farid: yeah, take that
knowledge that like a seasoned

557
00:33:41,567 --> 00:33:45,135
engineer would have in their
head and turn it into, turn it

558
00:33:45,135 --> 00:33:48,944
into kind of software that does
it automatically. that's really

559
00:33:48,944 --> 00:33:51,000
where the majority of the work
is.

560
00:33:51,000 --> 00:33:55,621
Audrow Nash: Yep. And so with
that part included, so the r and

561
00:33:55,621 --> 00:34:00,391
d for like how it works, that's
been 30% for the iteration. And,

562
00:34:00,391 --> 00:34:04,491
having those really powerful
heuristics of what kind of

563
00:34:04,491 --> 00:34:09,186
information and what things we
should even ask, that's probably

564
00:34:09,186 --> 00:34:13,360
an ongoing thing I would
imagine. But I'd imagine that's

565
00:34:13,360 --> 00:34:15,000
been a big effort too.

566
00:34:15,000 --> 00:34:17,247
Saman Farid: It is, yeah. it's a
continuous effort. It's very

567
00:34:17,247 --> 00:34:19,532
hard for me to estimate kinda
like what percentage of our time

568
00:34:19,532 --> 00:34:21,447
and energy went into that,
because it permeates into

569
00:34:21,447 --> 00:34:22,000
everything that

570
00:34:22,000 --> 00:34:23,000
Audrow Nash: It's

571
00:34:23,000 --> 00:34:26,250
Saman Farid: Yeah. as every time
we deploy a system, we look back

572
00:34:26,250 --> 00:34:29,500
at did we get the things that we
needed? What are the, challenges

573
00:34:29,500 --> 00:34:32,300
that we encountered in the
deployment, and how do we use

574
00:34:32,300 --> 00:34:35,000
that to improve our kind of
upfront screening process?

575
00:34:35,000 --> 00:34:38,642
Audrow Nash: Yeah. So you have
the lidar, which gives you a

576
00:34:38,642 --> 00:34:42,654
good, like you, you were getting
a 3D map of the environment, and

577
00:34:42,654 --> 00:34:45,617
then you have this data
ingestion part. is there

578
00:34:45,617 --> 00:34:49,444
anything else that you have in
the initial setup for this kind

579
00:34:49,444 --> 00:34:50,000
of thing?

580
00:34:50,000 --> 00:34:54,370
Saman Farid: yeah. And so, a lot
of it's just pure software based

581
00:34:54,370 --> 00:34:58,000
right? ingesting text and PDFs
and whatever files that

582
00:34:58,000 --> 00:35:02,296
Audrow Nash: Yeah. And the
injester has a lot of things in

583
00:35:02,296 --> 00:35:06,296
it, I imagine, like it's a
complex, big combination of

584
00:35:06,296 --> 00:35:08,000
bunch of software, but,

585
00:35:08,000 --> 00:35:11,277
Saman Farid: So then, like all
of those kinda mission

586
00:35:11,277 --> 00:35:15,049
parameters get fed to system
number two, right? System number

587
00:35:15,049 --> 00:35:18,264
two is really where, I think
there's much more, like

588
00:35:18,264 --> 00:35:21,789
difficulty, right? So system one
is like technically less

589
00:35:21,789 --> 00:35:25,437
difficult, but, operationally
very complex. section two, on

590
00:35:25,437 --> 00:35:29,023
our software tools, is what we
call fast Formic automation

591
00:35:29,023 --> 00:35:32,238
software tools, which is
basically what allows us to

592
00:35:32,238 --> 00:35:36,072
rapidly configure a robot work
cell. And what that means is we

593
00:35:36,072 --> 00:35:40,029
take all that information and we
say, okay, what is the solution

594
00:35:40,029 --> 00:35:43,739
here? for this set of mission
parameters, what type of robot

595
00:35:43,739 --> 00:35:47,634
do I need? Do I need a big robot
arm or a small robot arm, or a

596
00:35:47,634 --> 00:35:51,592
collaborative robot or a gantry
robot? Do I need, safety fencing

597
00:35:51,592 --> 00:35:55,425
or do I need area scanners? Do I
need this kind of conveyor or

598
00:35:55,425 --> 00:35:59,259
that kind of conveyor? What is
the height? What is the, depth,

599
00:35:59,259 --> 00:36:02,845
what is the cycle rate? What
kind of end of arm tools do I

600
00:36:02,845 --> 00:36:06,740
need? That used to be a process
that was highly manual, where a

601
00:36:06,740 --> 00:36:10,265
person would basically spend
months, going through all of

602
00:36:10,265 --> 00:36:13,851
those iterations, and simulating
all of them to figure out

603
00:36:13,851 --> 00:36:17,561
whether or not you meet the
demand. and what we've built, is

604
00:36:17,561 --> 00:36:21,024
really basically an automated
system configuration tool,

605
00:36:21,024 --> 00:36:24,795
right? So we take those mission
parameters, we feed them into

606
00:36:24,795 --> 00:36:28,505
our simulator. We iteratively
cycle through different, robot

607
00:36:28,505 --> 00:36:29,000
designs,

608
00:36:29,000 --> 00:36:31,000
Audrow Nash: Oh, wow.

609
00:36:31,000 --> 00:36:34,909
Saman Farid: choose the right
system for that task, as, as

610
00:36:34,909 --> 00:36:38,885
well as all the peripherals
associated with it. and then we

611
00:36:38,885 --> 00:36:43,131
run that robot in simulation to
validate what the cycle time is

612
00:36:43,131 --> 00:36:47,175
gonna be, right? Because, the
cycle time is often a function

613
00:36:47,175 --> 00:36:51,421
of capability. So let's say, I
only need to hit four cycles per

614
00:36:51,421 --> 00:36:55,532
minute. I might be fine using a
collaborative robot. and that

615
00:36:55,532 --> 00:36:59,374
collaborative robot, means that
I don't need to have area

616
00:36:59,374 --> 00:37:03,350
scanners, I don't need to have
safety fencing. and a lot of

617
00:37:03,350 --> 00:37:07,192
things get much simpler. but
that if that cycle rate goes

618
00:37:07,192 --> 00:37:11,573
from four parts per minute to 12
parts per minute or 20 parts per

619
00:37:11,573 --> 00:37:15,684
minute, Your physical footprint
may stay the same, right? But

620
00:37:15,684 --> 00:37:18,987
all of a sudden you like,
there's these cascading

621
00:37:18,987 --> 00:37:22,626
implications into all the
different, all the different

622
00:37:22,626 --> 00:37:26,737
components that go into that
work cell. and as that starts to

623
00:37:26,737 --> 00:37:30,848
happen, you, encounter a trade
off, right? You say, okay, the

624
00:37:30,848 --> 00:37:34,960
customer told me they only need
to hit five parts per minute,

625
00:37:34,960 --> 00:37:39,003
but should we design it for
seven or should we design it for

626
00:37:39,003 --> 00:37:42,980
four? and like we have to gather
all this information. and,

627
00:37:42,980 --> 00:37:47,158
that's, quote, like quoting and
pricing and lead times are all

628
00:37:47,158 --> 00:37:51,000
highly interdependent on the,
all those design decisions.

629
00:37:51,000 --> 00:37:52,000
Audrow Nash: Oh, a hundred
percent.

630
00:37:52,000 --> 00:37:55,776
Saman Farid: and at the end of
the day, like it's not a matter

631
00:37:55,776 --> 00:37:59,613
of. yeah, you can't make those
choices in a vacuum. And the way

632
00:37:59,613 --> 00:38:03,450
that it happens today typically
is, every robot vendor has some

633
00:38:03,450 --> 00:38:07,226
system, right? They're like, I
only make these systems, right?

634
00:38:07,226 --> 00:38:10,454
And, so if you were a
manufacturer and you go to them

635
00:38:10,454 --> 00:38:14,230
and you say, Hey, I want a robot
that, loads and unloads a CNC

636
00:38:14,230 --> 00:38:17,946
machine, let's say, I have this
right? And that may or may be

637
00:38:17,946 --> 00:38:21,783
the right thing. it may or may
not be the right thing, you have

638
00:38:21,783 --> 00:38:25,193
no idea of knowing. and the
supplier also is unlikely to

639
00:38:25,193 --> 00:38:28,726
know, because they're just
trying to sell you the piece of

640
00:38:28,726 --> 00:38:32,441
equipment that they have. and
this is where the kind of power

641
00:38:32,441 --> 00:38:35,974
of our business model comes in,
right? It's because we are

642
00:38:35,974 --> 00:38:39,811
committed to the performance of
that system for the entirety of

643
00:38:39,811 --> 00:38:43,587
its life. when we deploy a robot
and we say we're gonna hit 10

644
00:38:43,587 --> 00:38:46,876
parts per minute, we are
committed to hitting 10 parts

645
00:38:46,876 --> 00:38:50,652
per minute for from the day that
it arrives until the day that

646
00:38:50,652 --> 00:38:54,002
you say, yeah. until the day
that you're done with your

647
00:38:54,002 --> 00:38:57,534
contract. and so that requires,
a whole different level of

648
00:38:57,534 --> 00:39:00,823
design and, we are then
incentivized to make the right

649
00:39:00,823 --> 00:39:04,416
design choices, right? what is
the thing that's gonna, meet

650
00:39:04,416 --> 00:39:06,000
that, target consistently.

651
00:39:06,000 --> 00:39:09,796
Audrow Nash: Yeah. It's, we
talked about this in our first

652
00:39:09,796 --> 00:39:13,984
call, like the Alex Hormozi idea
of, they'd have to be stupid to

653
00:39:13,984 --> 00:39:17,911
not do it kind of thing. So
you're like, we're gonna make it

654
00:39:17,911 --> 00:39:21,641
so that we are hitting your
targets from day one for this

655
00:39:21,641 --> 00:39:25,634
kind of thing, and we're gonna
guarantee that, and fill these

656
00:39:25,634 --> 00:39:29,757
labor shortage gaps and make it
so you have higher utilization.

657
00:39:29,757 --> 00:39:31,000
This is all awesome

658
00:39:31,000 --> 00:39:34,713
Saman Farid: And it, requires
putting a lot more on the line

659
00:39:34,713 --> 00:39:38,611
right. Than, the typical robot
supplier who says, I'll sell you

660
00:39:38,611 --> 00:39:42,324
the thing. You figure out how to
live with it. we are set up

661
00:39:42,324 --> 00:39:46,099
where we take a lot of risk on,
If we deploy a robot and it's

662
00:39:46,099 --> 00:39:50,121
not getting used, that's bad for
us, right? And we do a lot, post

663
00:39:50,121 --> 00:39:53,958
deployment to make sure that the
customer is continuing to get

664
00:39:53,958 --> 00:39:56,000
the most value outta that
system.

665
00:39:56,000 --> 00:39:59,106
Audrow Nash: Yeah. I also
imagine, because you are

666
00:39:59,106 --> 00:40:02,087
providing metrics and
everything, like if you're

667
00:40:02,087 --> 00:40:05,006
coming in on a customer and
their supply, their

668
00:40:05,006 --> 00:40:08,174
manufacturing process is
constantly changing due to

669
00:40:08,174 --> 00:40:12,087
changes in demand or whatever it
is, seasonality, anything. and

670
00:40:12,087 --> 00:40:15,752
they're not tracking metrics
very well, like you are coming

671
00:40:15,752 --> 00:40:19,478
in and you're adding a lot of
value there. as you supply the

672
00:40:19,478 --> 00:40:23,081
metrics, help them figure out
where they have bottlenecks,

673
00:40:23,081 --> 00:40:26,000
help 'em optimize other parts of
their process.

674
00:40:26,000 --> 00:40:27,000
Saman Farid: Yeah.

675
00:40:27,000 --> 00:40:30,000
Audrow Nash: So that, oh, go
ahead.

676
00:40:30,000 --> 00:40:33,528
Saman Farid: I was gonna say, so
you know, this, kind of

677
00:40:33,528 --> 00:40:37,246
automated robot configuration
process, it, the thing that's

678
00:40:37,246 --> 00:40:40,522
related to that is also
automatically generating the

679
00:40:40,522 --> 00:40:44,554
robot program, right? to get the
to, because to run the robot in

680
00:40:44,554 --> 00:40:48,523
simulation, you need to have a
fully functioning robot program,

681
00:40:48,523 --> 00:40:52,052
that does that task. and so
that's another area where we

682
00:40:52,052 --> 00:40:55,391
spend a lot of energy is
automatically generating the

683
00:40:55,391 --> 00:40:59,234
program for the robot so that we
don't have to spend a lot of

684
00:40:59,234 --> 00:41:03,141
manual time kind of configuring
to do the task. that, that's a

685
00:41:03,141 --> 00:41:06,921
kind of where, generative AI
comes in, to really help, speed

686
00:41:06,921 --> 00:41:09,000
up the robot programming
process.

687
00:41:09,000 --> 00:41:13,053
Audrow Nash: What? So I'm
imagining with generative ai, I

688
00:41:13,053 --> 00:41:17,320
don't, my, current model of it
is that it's not that good at

689
00:41:17,320 --> 00:41:21,942
very fine details for things. So
if I wanted it to write the full

690
00:41:21,942 --> 00:41:25,000
program in c plus, that'd
probably be hard.

691
00:41:25,000 --> 00:41:26,000
Saman Farid: Yeah.

692
00:41:26,000 --> 00:41:30,466
Audrow Nash: but it can probably
pick action blocks very well for

693
00:41:30,466 --> 00:41:34,382
assembling your program, to
accomplish some goal. how are

694
00:41:34,382 --> 00:41:38,779
you guys using generative ai? Is
it like higher level behavioral

695
00:41:38,779 --> 00:41:42,626
things or is it pretty low
level, or what kind of things

696
00:41:42,626 --> 00:41:44,000
are you doing there?

697
00:41:44,000 --> 00:41:47,398
Saman Farid: Yeah. so it's a mix
of the two. So there's some

698
00:41:47,398 --> 00:41:50,967
selection of action blocks, but
there's also, some modification

699
00:41:50,967 --> 00:41:54,592
of those action blocks based on,
the components that you select,

700
00:41:54,592 --> 00:41:58,047
right? So if you use a vacuum
tool versus a clamp tool, or if

701
00:41:58,047 --> 00:42:01,672
you use, a vacuum tool that has
one zone versus three zones like

702
00:42:01,672 --> 00:42:05,240
that's a design choice that you
need to make. and then all of a

703
00:42:05,240 --> 00:42:08,299
sudden that impacts your
programming because if you're

704
00:42:08,299 --> 00:42:11,697
doing, if you're, let's just,
sorry, I'm, going into detail,

705
00:42:11,697 --> 00:42:13,000
right? But to give some

706
00:42:13,000 --> 00:42:14,000
Audrow Nash: no, I wanna hear
it.

707
00:42:14,000 --> 00:42:17,953
Saman Farid: right? if you have
a three zone, if you, in order

708
00:42:17,953 --> 00:42:22,097
to hit, let's say the nine parts
per minute rate, you might. Have

709
00:42:22,097 --> 00:42:25,540
a robot cycle time of three
cycles per minute. but you

710
00:42:25,540 --> 00:42:29,366
triple pick, right? If you pick
three at a time, you can hit

711
00:42:29,366 --> 00:42:33,000
nine parts per minute. if you're
picking three at a time,

712
00:42:33,000 --> 00:42:35,000
Audrow Nash: Just do three.
Yeah.

713
00:42:35,000 --> 00:42:37,952
Saman Farid: And what that means
is, one, you have to validate

714
00:42:37,952 --> 00:42:40,429
that the robot can hit the
weight and all the things

715
00:42:40,429 --> 00:42:43,095
associated with, picking up
three at a time. You need to

716
00:42:43,095 --> 00:42:45,857
have, end of arm tool that
supports that. You need to have

717
00:42:45,857 --> 00:42:48,714
indexing system upstream that
supplies three parts at a time

718
00:42:48,714 --> 00:42:51,762
for the robot to be able to pick
three parts at three parts at a

719
00:42:51,762 --> 00:42:54,571
time, at a fast enough rate. and
then you need to have your

720
00:42:54,571 --> 00:42:57,429
programming configure that,
right? Because if you're picking

721
00:42:57,429 --> 00:43:00,190
up three parts at a time, you
may not wanna drop all those

722
00:43:00,190 --> 00:43:03,000
three parts in the same way,
right? You might pick up three

723
00:43:03,000 --> 00:43:04,000
Audrow Nash: Oh,

724
00:43:04,000 --> 00:43:06,864
Saman Farid: and you need to
drop them separately. so then

725
00:43:06,864 --> 00:43:09,877
you need a vacuum tool that has
three zones anyway. Like I'm,

726
00:43:09,877 --> 00:43:12,000
maybe I'm going too far into the
weeds, but

727
00:43:12,000 --> 00:43:13,000
Audrow Nash: No, This is great.

728
00:43:13,000 --> 00:43:16,297
Saman Farid: I'm trying to
demonstrate here is that, it's

729
00:43:16,297 --> 00:43:19,710
not just a matter of drag and
drop the action block, right?

730
00:43:19,710 --> 00:43:22,892
Pick up apart, drop apart,
right? It's not as simple as

731
00:43:22,892 --> 00:43:25,842
that. The reality is, once
you've made those design

732
00:43:25,842 --> 00:43:29,371
decisions, you have to modify
all those action blocks and you

733
00:43:29,371 --> 00:43:33,073
need to see what that cycle time
looks like. In, the real world,

734
00:43:33,073 --> 00:43:36,486
and then you may need to come
back and change some of those

735
00:43:36,486 --> 00:43:40,130
design decisions again. because
if you pick up three at a time,

736
00:43:40,130 --> 00:43:43,774
but your indexing system is so
slow, that you end up, having, a

737
00:43:43,774 --> 00:43:47,129
low cycle time, you may be
better off picking two parts at

738
00:43:47,129 --> 00:43:50,774
a time or one part at a time. So
it's, it's, these things don't

739
00:43:50,774 --> 00:43:54,129
scale linear linearly. And I
think this is where, what, we

740
00:43:54,129 --> 00:43:57,773
found, and this is going back a
little bit into my career as a,

741
00:43:57,773 --> 00:44:01,360
vc, what I found is that a lot
of robotics companies were very

742
00:44:01,360 --> 00:44:04,946
focused on, how do I make the
robot pick the thing up and drop

743
00:44:04,946 --> 00:44:08,533
the thing off as efficiently as
possible? How can I make it so

744
00:44:08,533 --> 00:44:12,235
smart that I don't need to tell
it anything, I just need to pick

745
00:44:12,235 --> 00:44:15,763
up the thing and put it down.
but the reality is that doesn't

746
00:44:15,763 --> 00:44:19,292
actually solve for all these
things that we're talking about.

747
00:44:19,292 --> 00:44:22,821
how do you index the part? Where
do you drop them? How do you

748
00:44:22,821 --> 00:44:26,002
drop them? what is your kind of
upstream and downstream

749
00:44:26,002 --> 00:44:29,531
constraints? What robot do you
select to do that task? All of

750
00:44:29,531 --> 00:44:32,944
those are design decisions that
need to happen. and what we

751
00:44:32,944 --> 00:44:36,473
found is like a lot of robotic
startups, feel like that's the

752
00:44:36,473 --> 00:44:40,117
dirty work, right? They're like,
I don't wanna get involved in,

753
00:44:40,117 --> 00:44:43,183
that. I just wanna provide a
software, API and I want

754
00:44:43,183 --> 00:44:46,653
everybody else to figure out all
those things. and, like the

755
00:44:46,653 --> 00:44:50,356
reality is like that means that
nobody's gonna be able to adopt,

756
00:44:50,356 --> 00:44:54,000
right? Because, that's, really
where the rubber meets the road.

757
00:44:54,000 --> 00:44:58,778
Audrow Nash: Yeah, very much.
Yeah. I've seen a whole bunch of

758
00:44:58,778 --> 00:45:03,324
robotics companies that make a
product and then drop it off

759
00:45:03,324 --> 00:45:07,562
and, that, or they have to do
all of the custom work of

760
00:45:07,562 --> 00:45:12,340
fitting it to the warehouse or
fitting it to the manufacturing

761
00:45:12,340 --> 00:45:16,963
facility or whatever. Every time
with every customer, and it

762
00:45:16,963 --> 00:45:21,741
makes it, it's a severe limit on
their scaling ability. so it,

763
00:45:21,741 --> 00:45:26,364
like what you guys have worked
on, and I guess this probably

764
00:45:26,364 --> 00:45:31,219
goes back to your VC experience,
is it seems like you have gone

765
00:45:31,219 --> 00:45:36,228
right to the hardest part, which
is the configuration so that you

766
00:45:36,228 --> 00:45:38,000
can scale a lot easier.

767
00:45:38,000 --> 00:45:41,559
Saman Farid: Yeah. Yeah. using
reinforcement learning to do the

768
00:45:41,559 --> 00:45:44,609
pick and place part is table
stakes today. I've gotten

769
00:45:44,609 --> 00:45:47,829
pitched on that, probably a
hundred times in the last two

770
00:45:47,829 --> 00:45:51,162
years, right? and I continue to
get new people coming to me

771
00:45:51,162 --> 00:45:54,608
every day and saying, oh, I'm
like, a really, smart, Stanford

772
00:45:54,608 --> 00:45:58,166
graduate and I have this really
great way to pick up things and

773
00:45:58,166 --> 00:46:01,556
put them down with the robot
and, I'm gonna save you a whole

774
00:46:01,556 --> 00:46:04,606
bunch of money on your
programming costs. And then the

775
00:46:04,606 --> 00:46:08,052
reality is that's not, true at
all, right? 'cause programming

776
00:46:08,052 --> 00:46:11,498
the robot to pick the thing up
and put it down is maybe 2% of

777
00:46:11,498 --> 00:46:14,718
my deployment, frustration,
right? 98% of it is all these

778
00:46:14,718 --> 00:46:15,000
other

779
00:46:15,000 --> 00:46:18,937
Audrow Nash: Ha. Yeah, for sure.
And especially as you want to

780
00:46:18,937 --> 00:46:22,556
grow, like the point of a
startup is growth, I think. And

781
00:46:22,556 --> 00:46:26,048
if you are just, you're
effectively, I've been thinking

782
00:46:26,048 --> 00:46:29,667
of a lot of companies as like
making a distinction, where

783
00:46:29,667 --> 00:46:33,476
it's, you have startups and
those are focused on growth, and

784
00:46:33,476 --> 00:46:37,413
you have contractors that are
focused on, they may get revenue

785
00:46:37,413 --> 00:46:41,349
and they may gradually be able
to get more efficient, but they

786
00:46:41,349 --> 00:46:45,222
are very often, doing things
that don't scale very well for a

787
00:46:45,222 --> 00:46:48,968
long time. And they may call
themselves a startup, but they

788
00:46:48,968 --> 00:46:52,841
very often are effectively just
consulting around the problem

789
00:46:52,841 --> 00:46:56,587
they're solving. and so you
guys, it seems you're very much

790
00:46:56,587 --> 00:46:59,000
growth focused for this kind of
thing,

791
00:46:59,000 --> 00:47:00,000
Saman Farid: yeah.

792
00:47:00,000 --> 00:47:03,000
Audrow Nash: the, oh, go ahead.

793
00:47:03,000 --> 00:47:06,231
Saman Farid: Oh, I, was gonna, I
was gonna move on to the next,

794
00:47:06,231 --> 00:47:09,000
next block of software that we
built, but, if you have

795
00:47:09,000 --> 00:47:13,102
Audrow Nash: I still have more
questions for this. yeah. so you

796
00:47:13,102 --> 00:47:17,139
mentioned reinforcement learning
as a way of doing this. I was

797
00:47:17,139 --> 00:47:20,589
wondering how are you actually
going back to just the

798
00:47:20,589 --> 00:47:23,910
generative right before the
reinforcement learning,

799
00:47:23,910 --> 00:47:28,012
generative ai. So the way that
I'm imagining it, and maybe this

800
00:47:28,012 --> 00:47:31,983
isn't quite correct, so largely
you're selecting actions, but

801
00:47:31,983 --> 00:47:35,825
you're also modifying those
actions. That's really cool. is

802
00:47:35,825 --> 00:47:39,992
it like you have small parts of
it that you let it so there like

803
00:47:39,992 --> 00:47:43,833
this is a larger action and
we're saying, okay, I'm scoping

804
00:47:43,833 --> 00:47:48,000
you, you can do this one thing.
You're writing this input output

805
00:47:48,000 --> 00:47:50,915
Saman Farid: that's exactly
right. Yeah. We basically give

806
00:47:50,915 --> 00:47:54,131
it a template of what we want it
to build. And then we say, here

807
00:47:54,131 --> 00:47:56,945
are the inputs that are
different, from kinda that, that

808
00:47:56,945 --> 00:47:59,859
phase one, right? as we gather
all this data, we generated

809
00:47:59,859 --> 00:48:03,126
these kind of requirements. Here
are the requirements. Here's the

810
00:48:03,126 --> 00:48:06,191
template I want you to fill.
Here are the things that we need

811
00:48:06,191 --> 00:48:08,000
to figure out. and it fills 'em
out.

812
00:48:08,000 --> 00:48:12,085
Audrow Nash: That makes a lot
more sense to me. 'cause the

813
00:48:12,085 --> 00:48:15,817
other one would be pretty
finicky I would imagine, in

814
00:48:15,817 --> 00:48:20,183
terms of it works sometimes, but
it wouldn't work a lot of the

815
00:48:20,183 --> 00:48:24,620
time. but confining it would be
much better. What are you guys,

816
00:48:24,620 --> 00:48:28,775
are you using, are you using
like Facebook's LLM or Meta's,

817
00:48:28,775 --> 00:48:33,282
LLM or how, I'd be curious about
your specific, generative ai, I

818
00:48:33,282 --> 00:48:37,507
don't know, program or service
you're using for this kind of

819
00:48:37,507 --> 00:48:42,085
thing? For this. I would imagine
that it would probably be Meta's

820
00:48:42,085 --> 00:48:43,000
Lambda model.

821
00:48:43,000 --> 00:48:49,000
Saman Farid: Llama. Yeah, we use
La Llama with some, so

822
00:48:49,000 --> 00:48:51,000
Audrow Nash: Modifications for

823
00:48:51,000 --> 00:48:54,000
Saman Farid: and, a bunch of
tooling around it.

824
00:48:54,000 --> 00:48:57,398
Audrow Nash: hell. Yeah. It's
cool. It's a neat thing. I,

825
00:48:57,398 --> 00:49:00,737
yeah, I was imagining as a
company you don't want to use

826
00:49:00,737 --> 00:49:04,016
like. Claude or something
directly or chat GPT directly

827
00:49:04,016 --> 00:49:07,534
for this kind of thing, having
your own control of it would

828
00:49:07,534 --> 00:49:11,051
probably be good. not having
them like chat. GPT releases a

829
00:49:11,051 --> 00:49:14,867
new version. It breaks all your
stuff. Yeah. Or just some crappy

830
00:49:14,867 --> 00:49:16,000
update or something

831
00:49:16,000 --> 00:49:19,685
Saman Farid: Yeah, But the, the
good thing is right, the,

832
00:49:19,685 --> 00:49:23,241
quality of these, multimodal
models is, getting so much

833
00:49:23,241 --> 00:49:27,250
better, so fast that like we can
start to ingest more and more

834
00:49:27,250 --> 00:49:31,000
types of things and skip more
and more of the steps. Yeah.

835
00:49:31,000 --> 00:49:35,628
Audrow Nash: And then, so just
so I understand your fast,

836
00:49:35,628 --> 00:49:40,825
automation tool a bit more, you
are, the way I'm imagining it is

837
00:49:40,825 --> 00:49:45,859
you. Have your ingested data,
which gives you a structure in a

838
00:49:45,859 --> 00:49:50,650
sense, like you're inferring
some sort of requirements from

839
00:49:50,650 --> 00:49:55,197
your ingested data. And then
from that you are going to,

840
00:49:55,197 --> 00:49:59,906
given constraints of things
you've, observed, you're going

841
00:49:59,906 --> 00:50:05,021
to try to figure out what's the
best way to meet the customer's

842
00:50:05,021 --> 00:50:09,000
goals at the lowest price for
this kind of thing.

843
00:50:09,000 --> 00:50:10,000
Saman Farid: Yep.

844
00:50:09,000 --> 00:50:13,538
Audrow Nash: And then to do
that, what you do is you, maybe

845
00:50:13,538 --> 00:50:18,231
you use LLMs again and they're
selecting like base components

846
00:50:18,231 --> 00:50:22,769
or something like this, and
these are like this robot here,

847
00:50:22,769 --> 00:50:27,615
this kind of thing. and then you
are running that in simulation

848
00:50:27,615 --> 00:50:32,538
with a good way of establishing
metrics for it and seeing how it

849
00:50:32,538 --> 00:50:35,000
does or how, does all this work?

850
00:50:35,000 --> 00:50:38,670
Saman Farid: so you could think
of it a little bit as, like

851
00:50:38,670 --> 00:50:42,278
instead of human in the loop,
like we have AI in the loop,

852
00:50:42,278 --> 00:50:46,134
right? So like the, core of what
we're doing here lives in the

853
00:50:46,134 --> 00:50:49,867
simulator. so our simulation
tool takes those configuration,

854
00:50:49,867 --> 00:50:53,412
takes that configuration
information, and then generates,

855
00:50:53,412 --> 00:50:57,393
or, and runs, the full work cell
based on what it has available.

856
00:50:57,393 --> 00:51:01,126
and then the ai, what it does is
it looks at the performance

857
00:51:01,126 --> 00:51:04,671
data, right? It says, look,
cycle time is too low. Change

858
00:51:04,671 --> 00:51:08,403
this, or, space constraints are
being exceeded. change that,

859
00:51:08,403 --> 00:51:12,260
right? and, and it goes through
this process where it iterates

860
00:51:12,260 --> 00:51:15,930
through a couple of different
variants, and usually what it

861
00:51:15,930 --> 00:51:19,476
spits out is three or four
candidate configuration files,

862
00:51:19,476 --> 00:51:22,959
right? and, that's where the
human gets involved, right?

863
00:51:22,959 --> 00:51:26,380
Which it says, Hey, we have
kinda option A, which is, a

864
00:51:26,380 --> 00:51:29,802
hundred thousand dollars of
cost, using a collaborative

865
00:51:29,802 --> 00:51:33,223
robot, and we're able to hit
four parts per minute. and

866
00:51:33,223 --> 00:51:36,893
option B, it's, $150,000 of
cost. It's an industrial robot.

867
00:51:36,893 --> 00:51:40,501
It takes a bigger footprint and
it can hit eight parts per

868
00:51:40,501 --> 00:51:44,482
minute. option C is against your
robot and it can do blah, blah,

869
00:51:44,482 --> 00:51:48,214
blah, and it can hit. 12 parts
per minute, whatever it might

870
00:51:48,214 --> 00:51:52,070
be. And we will, it'll generate
those options for us. And then

871
00:51:52,070 --> 00:51:56,052
our, then we do have humans in,
the loop, right? At, that point,

872
00:51:56,052 --> 00:51:59,659
who look at those configuration
files and make the kind of

873
00:51:59,659 --> 00:52:03,454
business trade-offs related to
it, right? It's okay, what are

874
00:52:03,454 --> 00:52:07,497
the, if I want to hit the hourly
rate that, that hits, that saves

875
00:52:07,497 --> 00:52:11,478
money for this customer. And, we
want a system that's relatively

876
00:52:11,478 --> 00:52:15,397
redeployable. if that, product
makes changes in that customer's

877
00:52:15,397 --> 00:52:19,254
site, which of these, option A,
B, or C is gonna be the better

878
00:52:19,254 --> 00:52:20,000
fit, Because

879
00:52:20,000 --> 00:52:23,000
Audrow Nash: like that you do
that. You give them options

880
00:52:23,000 --> 00:52:24,000
Saman Farid: Yeah. because

881
00:52:24,000 --> 00:52:25,000
Audrow Nash: let them fine tune

882
00:52:25,000 --> 00:52:28,679
Saman Farid: yeah. The, it's not
static, right? the, product

883
00:52:28,679 --> 00:52:32,236
requirements are not static.
What you might have today, is

884
00:52:32,236 --> 00:52:36,222
something that only needs to hit
four parts per minute. but three

885
00:52:36,222 --> 00:52:40,085
months from now, you may win a
new order and you might have to,

886
00:52:40,085 --> 00:52:43,948
increase that, to 12 parts per
minute. And you don't wanna have

887
00:52:43,948 --> 00:52:47,934
to rip out that whole system and
put in a new one. So you need to

888
00:52:47,934 --> 00:52:51,000
kind of design for a couple of,
potential changes.

889
00:52:51,000 --> 00:52:54,000
Audrow Nash: That's super cool.

890
00:52:54,000 --> 00:52:57,879
Audrow Nash: Okay, so how do
you, can you tell me a bit about

891
00:52:57,879 --> 00:53:01,630
your software stack? are you
using ROS, are you using Isaac

892
00:53:01,630 --> 00:53:05,000
Sim? Are you using, do you make
your own stuff? What?

893
00:53:05,000 --> 00:53:10,300
Saman Farid: not using ROS. we
are using, yeah, Isaac

894
00:53:10,300 --> 00:53:15,600
simulation for our, like
hardware and loop simulation

895
00:53:15,600 --> 00:53:22,000
processes. we also use, another
tool called Robo dk, which yeah,

896
00:53:22,000 --> 00:53:27,000
robo DK is a, robot simulator.
that, there's a lot

897
00:53:27,000 --> 00:53:30,100
Audrow Nash: give you a bunch of
off the shelf robots for this

898
00:53:30,100 --> 00:53:31,000
kind of thing? Was

899
00:53:31,000 --> 00:53:35,604
Saman Farid: have the, one of
the big benefits is they have a

900
00:53:35,604 --> 00:53:40,283
huge, template library, that we
could use. so all of the, they

901
00:53:40,283 --> 00:53:45,189
have all the configuration files
for kind of, most of the off the

902
00:53:45,189 --> 00:53:49,566
shelf robot arms that we might
use. and they, like the, to

903
00:53:49,566 --> 00:53:51,000
'cause a simulation

904
00:53:51,000 --> 00:53:54,692
Audrow Nash: Oh, is dk is that
developer kit? it's like SDK,

905
00:53:54,692 --> 00:53:55,000
but a

906
00:53:55,000 --> 00:53:57,746
Saman Farid: what it stands for.
I don't know what it stands for.

907
00:53:57,746 --> 00:53:58,000
Maybe,

908
00:53:58,000 --> 00:54:00,000
Audrow Nash: ' cause Robot
developer kit.

909
00:54:00,000 --> 00:54:01,000
Saman Farid: Could

910
00:54:01,000 --> 00:54:02,000
Audrow Nash: okay. Just guessing

911
00:54:02,000 --> 00:54:04,000
Saman Farid: Yeah, it could be,

912
00:54:04,000 --> 00:54:06,805
Audrow Nash: based on your
description. That's what it

913
00:54:06,805 --> 00:54:08,000
sounds like, but, okay,

914
00:54:08,000 --> 00:54:09,000
Saman Farid: yeah,

915
00:54:09,000 --> 00:54:10,677
Audrow Nash: so you use that,
you have Isaacs in for

916
00:54:10,677 --> 00:54:11,000
simulation

917
00:54:11,000 --> 00:54:12,000
Saman Farid: built a bunch of

918
00:54:12,000 --> 00:54:13,000
Audrow Nash: for robot
configuration.

919
00:54:13,000 --> 00:54:16,529
Saman Farid: Yeah. We built a
bunch of plugins that go on top

920
00:54:16,529 --> 00:54:19,826
of Robot DK that kind of
automate a lot of the steps, for

921
00:54:19,826 --> 00:54:20,000
it.

922
00:54:20,000 --> 00:54:24,609
Audrow Nash: Okay. Hell yeah.
And then, okay, so that's the

923
00:54:24,609 --> 00:54:29,141
fast part of this, that your
automation tool, what goes on

924
00:54:29,141 --> 00:54:30,000
top of that

925
00:54:30,000 --> 00:54:31,000
Audrow Nash: ?

926
00:54:30,000 --> 00:54:34,160
Saman Farid: the, so the next
phase is, build validation,

927
00:54:34,160 --> 00:54:38,465
right? it's one thing to have
the perfect robot designed in

928
00:54:38,465 --> 00:54:42,698
simulation. then, physically
building it often, means that

929
00:54:42,698 --> 00:54:47,004
you, end up with a bunch of
differences, right? we, some of

930
00:54:47,004 --> 00:54:51,091
'em we built in-house, but most
of the time we use third

931
00:54:51,091 --> 00:54:54,886
parties, who build, the hardware
for us based on our

932
00:54:54,886 --> 00:54:59,483
configuration, right? So we say
we need this robot arm and this

933
00:54:59,483 --> 00:55:04,081
conveyor and these sensors. and
what, as that system is getting

934
00:55:04,081 --> 00:55:07,000
built, you need to validate.
Everything.

935
00:55:07,000 --> 00:55:08,000
Audrow Nash: Yep.

936
00:55:08,000 --> 00:55:12,023
Saman Farid: one of the, if even
one of the sets of wires are,

937
00:55:12,023 --> 00:55:15,592
are, installed, backwards,
right? Which we, which would

938
00:55:15,592 --> 00:55:19,031
have, which happens all the
time, all of a sudden the

939
00:55:19,031 --> 00:55:22,729
performance of that system is
works wrong, right? if this

940
00:55:22,729 --> 00:55:26,557
sensor is connected to that IO
and, and it's the wrong one,

941
00:55:26,557 --> 00:55:30,710
right? it expects that the other
sensors connect to the io, like

942
00:55:30,710 --> 00:55:34,538
your performance all just goes
out the window. and so we've

943
00:55:34,538 --> 00:55:38,496
built a bunch of tools that's
that. we, send an industrial PC

944
00:55:38,496 --> 00:55:42,000
to every system builder. And as
they build the system,

945
00:55:42,000 --> 00:55:45,000
Audrow Nash: What's that?

946
00:55:45,000 --> 00:55:48,647
Saman Farid: that industrial PC
gets a industrial pc, which is

947
00:55:48,647 --> 00:55:50,000
an industrial computer.

948
00:55:50,000 --> 00:55:51,000
Audrow Nash: Oh, I see. Yeah.

949
00:55:51,000 --> 00:55:54,862
Saman Farid: the industrial pc,
sits in the control box,

950
00:55:54,862 --> 00:55:58,862
connects to the PLC, the robot
programming and, sorry, the

951
00:55:58,862 --> 00:56:02,931
robot controller, and all of the
other peripherals. And, it

952
00:56:02,931 --> 00:56:06,724
validates that everything was
done correctly. it has ba

953
00:56:06,724 --> 00:56:10,586
essentially it's unit tests,
right? And software is very

954
00:56:10,586 --> 00:56:14,379
common to do unit tests. We
basically do unit tests for

955
00:56:14,379 --> 00:56:18,793
hardware as well. So we'll have
the robot, do a certain movement

956
00:56:18,793 --> 00:56:22,862
block, a certain sensor, and
then we'll say, did we get the

957
00:56:22,862 --> 00:56:27,138
reading that we expected from
that sensor or not? or, we will,

958
00:56:27,138 --> 00:56:31,000
have the robot automatically
tripped some of the safety,

959
00:56:31,000 --> 00:56:35,000
Audrow Nash: so cool.

960
00:56:35,000 --> 00:56:38,375
Saman Farid: see if that
triggers the response that we

961
00:56:38,375 --> 00:56:42,250
were expecting or not. and so
we're able to validate that, the

962
00:56:42,250 --> 00:56:43,000
built system

963
00:56:43,000 --> 00:56:44,000
Audrow Nash: is what you

964
00:56:44,000 --> 00:56:46,261
Saman Farid: performs in the
same way exactly as the

965
00:56:46,261 --> 00:56:47,000
simulated system.

966
00:56:47,000 --> 00:56:51,058
Audrow Nash: How do you, so I'm
imagining this is all again, so

967
00:56:51,058 --> 00:56:55,115
not fully, I mean it's probably
somewhat generated, but largely

968
00:56:55,115 --> 00:56:59,237
templated for this kind of thing
where you say, okay, we added a

969
00:56:59,237 --> 00:57:03,166
camera. Let's make a movement
that blocks it for this kind of

970
00:57:03,166 --> 00:57:06,000
thing. That's super cool. I
really like that

971
00:57:06,000 --> 00:57:09,095
Saman Farid: our deployment
project manager will have a kind

972
00:57:09,095 --> 00:57:12,139
of, a validation checklist for
every system, right? They'll

973
00:57:12,139 --> 00:57:15,389
say, Hey, here are the 23 things
that we would check if we were

974
00:57:15,389 --> 00:57:18,278
on site. what percentage of
those can be done? Can we do

975
00:57:18,278 --> 00:57:19,000
automatically?

976
00:57:19,000 --> 00:57:22,715
Audrow Nash: That's awesome. I
really like, because these are

977
00:57:22,715 --> 00:57:26,612
things that are. Time consuming
for most companies, and I really

978
00:57:26,612 --> 00:57:30,266
like that you've spent the time
to automate them. And I also

979
00:57:30,266 --> 00:57:34,042
feel like, I'm curious because
when did you, you said you guys

980
00:57:34,042 --> 00:57:37,574
have been running for three
years or something, right? did

981
00:57:37,574 --> 00:57:38,000
you say

982
00:57:38,000 --> 00:57:40,400
Saman Farid: and a half. Yeah.
20 end of 2020 is when we

983
00:57:40,400 --> 00:57:41,000
started. Yeah.

984
00:57:41,000 --> 00:57:45,568
Audrow Nash: End of 2020. How,
much, did you pivot quite a bit

985
00:57:45,568 --> 00:57:50,063
with generative ai? 'cause open
AI came out, I don't remember

986
00:57:50,063 --> 00:57:54,411
when that was. Like 2022 or
what, like end of 20, 22, or am

987
00:57:54,411 --> 00:57:55,000
I wrong?

988
00:57:55,000 --> 00:57:58,013
Saman Farid: I don't remember
exactly when opening AI came

989
00:57:58,013 --> 00:57:59,000
out, but we didn't,

990
00:57:59,000 --> 00:58:02,720
Audrow Nash: Did, you pivot hard
or was it just it assisted in

991
00:58:02,720 --> 00:58:05,000
what you were trying to do
already or?

992
00:58:05,000 --> 00:58:08,629
Saman Farid: These were all
things that were on the roadmap

993
00:58:08,629 --> 00:58:12,443
from the beginning. they just
became much more capable, right?

994
00:58:12,443 --> 00:58:15,765
As LLM started to get much
better. now all of a sudden

995
00:58:15,765 --> 00:58:19,640
these tools, started to become
more capable. Although, a lot of

996
00:58:19,640 --> 00:58:23,331
these things that we're doing
don't have anything to do with

997
00:58:23,331 --> 00:58:27,022
LLMs, right? Like the kind of
simulation piece is not really

998
00:58:27,022 --> 00:58:30,651
related to LLMs. The, robot,
performance validation set, is

999
00:58:30,651 --> 00:58:34,404
not really related to LLMs. I
think AI has a part to play for

1000
00:58:34,404 --> 00:58:37,972
sure. but. A lot of these
things, have to draw from the, a

1001
00:58:37,972 --> 00:58:41,293
foundation of a bunch of
different, things that people

1002
00:58:41,293 --> 00:58:44,000
have developed over the last 20
years. yeah,

1003
00:58:44,000 --> 00:58:47,599
Audrow Nash: Yeah, very much.
Yeah, it is quite cool. And I

1004
00:58:47,599 --> 00:58:51,504
guess you could have just always
had a human in the loop and you

1005
00:58:51,504 --> 00:58:55,286
in, some specific parts and it
wouldn't have been that much of

1006
00:58:55,286 --> 00:58:58,947
a slowdown, most likely. but you
get better efficiency using

1007
00:58:58,947 --> 00:59:02,729
these kinds of tools. And even
if a person then curates it, or

1008
00:59:02,729 --> 00:59:06,329
says, ah, these are results
suck, run it again, or whatever

1009
00:59:06,329 --> 00:59:07,000
it might be

1010
00:59:07,000 --> 00:59:10,742
Audrow Nash: . How does your,
data ingestion part work? So if

1011
00:59:10,742 --> 00:59:14,546
they have scraps of paper and
things like this, or how do you,

1012
00:59:14,546 --> 00:59:18,104
getting that initial data of
their setup, I understand the

1013
00:59:18,104 --> 00:59:21,847
LIDAR scan, but understanding a
lot of the other things about

1014
00:59:21,847 --> 00:59:25,466
them. How does that all work?
That sounds like a thing that

1015
00:59:25,466 --> 00:59:27,000
LLMs would be pretty good

1016
00:59:27,000 --> 00:59:30,609
Saman Farid: yeah. That, is
something that helps a lot. So

1017
00:59:30,609 --> 00:59:34,219
usually what happens is our
salespeople who are, who go on

1018
00:59:34,219 --> 00:59:38,077
site, Know all the things that
they're supposed to gather. and

1019
00:59:38,077 --> 00:59:41,624
they just take pictures and
videos of all the things that

1020
00:59:41,624 --> 00:59:45,482
they can, right? So if the guy
who's working on the site will,

1021
00:59:45,482 --> 00:59:48,656
is doing a certain set of
motion, take a picture of

1022
00:59:48,656 --> 00:59:52,265
whatever they've written down,
if you have boxes that have

1023
00:59:52,265 --> 00:59:56,186
serial numbers and, weights and
sizes that are written on them,

1024
00:59:56,186 --> 00:59:59,857
will take a picture of that.
that whole set of things that,

1025
00:59:59,857 --> 01:00:03,716
gets loaded into, our, our kind
of, configuration tool, right?

1026
01:00:03,716 --> 01:00:07,387
and that's what generates the
mission parameters. So I. It,

1027
01:00:07,387 --> 01:00:10,810
can come in a variety of
different formats. It can also

1028
01:00:10,810 --> 01:00:14,730
even come in the form of audio
recordings. like our salespeople

1029
01:00:14,730 --> 01:00:18,464
use chorus, which is the tool
that kind of records meetings.

1030
01:00:18,464 --> 01:00:22,447
Usually in those meetings, a lot
of the information that we need

1031
01:00:22,447 --> 01:00:26,243
to gather, can be like, is asked
and answered, right? So that

1032
01:00:26,243 --> 01:00:30,039
also gets fed into the same,
model for that project. So every

1033
01:00:30,039 --> 01:00:33,897
project, basically has a set of,
context data that it's given.

1034
01:00:33,897 --> 01:00:37,693
and then the LLM is responsible
for filling out this kind of,

1035
01:00:37,693 --> 01:00:39,000
filling out this form

1036
01:00:39,000 --> 01:00:41,815
Audrow Nash: you have a data
structure effectively filling it

1037
01:00:41,815 --> 01:00:42,000
out.

1038
01:00:42,000 --> 01:00:44,783
Saman Farid: And then surfacing
the things that it doesn't have.

1039
01:00:44,783 --> 01:00:47,000
It says, Hey, nobody asked,
about this thing that I

1040
01:00:47,000 --> 01:00:49,000
Audrow Nash: Oh, ooh.

1041
01:00:49,000 --> 01:00:51,000
Saman Farid: can somebody go
back to the customer and ask it?

1042
01:00:51,000 --> 01:00:54,042
Audrow Nash: how do you do that?
Because that's a very

1043
01:00:54,042 --> 01:00:57,366
interesting thing. 'cause LLMs
are not that good at knowing

1044
01:00:57,366 --> 01:01:00,859
what they don't know. and having
it be like, I don't have this

1045
01:01:00,859 --> 01:01:03,000
information is very interesting
to me.

1046
01:01:03,000 --> 01:01:05,353
Saman Farid: a lot of that just
comes with the right kind of

1047
01:01:05,353 --> 01:01:07,000
prompting, right? It's do you
know this or

1048
01:01:07,000 --> 01:01:08,000
Audrow Nash: Ah ha

1049
01:01:08,000 --> 01:01:10,000
Saman Farid: did you, can you
find

1050
01:01:10,000 --> 01:01:14,348
Audrow Nash: did you see this or
no? You, you, say yes or no. I

1051
01:01:14,348 --> 01:01:18,627
see. Yeah, I, since chat GPTI
have been looking for how to use

1052
01:01:18,627 --> 01:01:22,699
LLMs in some way to make my
podcast posting more efficient.

1053
01:01:22,699 --> 01:01:26,702
And, it's very funny 'cause I
feel like in some sense it's

1054
01:01:26,702 --> 01:01:30,982
like I'm trying all the things
right as they're available. And

1055
01:01:30,982 --> 01:01:35,261
so it's a very good excuse to
become quite familiar with them,

1056
01:01:35,261 --> 01:01:38,781
but it's very funny how, tough
it is to reduce some

1057
01:01:38,781 --> 01:01:42,784
hallucinations. it's, but I
guess just asking, did you see

1058
01:01:42,784 --> 01:01:46,856
this? And if so, what is it?
Kind of thing. It's, no, I did

1059
01:01:46,856 --> 01:01:50,928
not see it. It can probably
answer that pretty reliably and

1060
01:01:50,928 --> 01:01:55,000
then you can ask it to check
itself for this kind of thing.

1061
01:01:55,000 --> 01:01:56,000
Saman Farid: Yeah. Yeah.

1062
01:01:56,000 --> 01:01:59,727
Audrow Nash: the new llama
model, I can access it through

1063
01:01:59,727 --> 01:02:03,781
perplexity. I haven't actually
run it locally. And, the Claude

1064
01:02:03,781 --> 01:02:07,965
3.5 sonnet are really good. Like
they're genuinely useful, which

1065
01:02:07,965 --> 01:02:12,215
is just super, super cool 'cause
nothing was useful until, Claude

1066
01:02:12,215 --> 01:02:13,000
Opus for me.

1067
01:02:13,000 --> 01:02:16,941
Saman Farid: yeah. I think the
thing I, my, my perspective on a

1068
01:02:16,941 --> 01:02:20,820
lot of these LLMs is that, The
large language model themselves

1069
01:02:20,820 --> 01:02:24,198
are never gonna be perfect,
right? Like the thing that

1070
01:02:24,198 --> 01:02:27,764
matters is if you have a good
filtering mechanism for the

1071
01:02:27,764 --> 01:02:31,392
output that comes out of it,
right? Like you need to have,

1072
01:02:31,392 --> 01:02:35,146
rule-based things a lot of the
time that check the output of

1073
01:02:35,146 --> 01:02:38,399
the, AI model. because
oftentimes you know what your

1074
01:02:38,399 --> 01:02:42,090
expected output should be, it's
one of these few options or

1075
01:02:42,090 --> 01:02:46,031
whatever, right? And so the LLM
is not very good at sticking to

1076
01:02:46,031 --> 01:02:49,972
the right out, answer. but you
can check that with a rule-based

1077
01:02:49,972 --> 01:02:53,725
system. And so you need to, I
feel like you need to sandwich

1078
01:02:53,725 --> 01:02:57,542
the LLM between upstream. You
need to. give it information in

1079
01:02:57,542 --> 01:03:01,483
a very specific way. that's the
prompt, right? So you constrain

1080
01:03:01,483 --> 01:03:05,049
the input, and then on the
output you need to sandwich it

1081
01:03:05,049 --> 01:03:08,427
with another rule-based
filtering mechanism, right? So

1082
01:03:08,427 --> 01:03:12,180
it, it does whatever it's gonna
do in its black box, but you

1083
01:03:12,180 --> 01:03:16,121
have a screen for what actually
comes out of the other side. so

1084
01:03:16,121 --> 01:03:19,562
that it, you constrain the
problem, a little bit better

1085
01:03:19,562 --> 01:03:20,000
because

1086
01:03:20,000 --> 01:03:21,000
Audrow Nash: both ends.

1087
01:03:21,000 --> 01:03:24,714
Saman Farid: Yeah, most of the
time, you're not gonna ask, I'm

1088
01:03:24,714 --> 01:03:28,429
not gonna ask an LLM to write a
poem for me, right? in, in the

1089
01:03:28,429 --> 01:03:31,903
context of Formic. so I know
that I can limit the kinds of

1090
01:03:31,903 --> 01:03:34,000
outputs that are gonna come,
right?

1091
01:03:34,000 --> 01:03:39,676
Audrow Nash: Yep. Yeah, very,
true. It's an interesting time

1092
01:03:39,676 --> 01:03:41,000
with all this.

1093
01:03:41,000 --> 01:03:42,000
Saman Farid: Yeah.

1094
01:03:41,000 --> 01:03:43,000
Audrow Nash: Okay

1095
01:03:43,000 --> 01:03:47,519
Audrow Nash: . so you have your
robot dk, and the simulation,

1096
01:03:47,519 --> 01:03:51,000
and you are performing
validation on the actual

1097
01:03:51,000 --> 01:03:55,593
hardware. To make sure that it
is working as you would expect.

1098
01:03:55,593 --> 01:04:00,185
What do you do from there? do
you install it in the customer's

1099
01:04:00,185 --> 01:04:03,000
site or how, do we proceed from
there?

1100
01:04:03,000 --> 01:04:06,880
Saman Farid: Yeah. the next step
is usually we do, a full

1101
01:04:06,880 --> 01:04:10,557
acceptance test, which I
described parts of, we do the

1102
01:04:10,557 --> 01:04:14,437
acceptance testing, at the
system, whoever's building the

1103
01:04:14,437 --> 01:04:18,454
system for us. we run a full
acceptance test. if it passes,

1104
01:04:18,454 --> 01:04:22,879
we ship it to the customer site,
we install it, and then we run a

1105
01:04:22,879 --> 01:04:27,100
secondary acceptance test that
is in their environment, right?

1106
01:04:27,100 --> 01:04:30,708
Because there's usually
differences, for a variety of

1107
01:04:30,708 --> 01:04:34,657
different reasons. and we do
that acceptance testing. That

1108
01:04:34,657 --> 01:04:39,014
usually happens in about half a
day. and then we also generate a

1109
01:04:39,014 --> 01:04:42,962
bunch of, Like peripheral
information. So it like, usually

1110
01:04:42,962 --> 01:04:47,319
that looks like kind of training
handbooks for the operators who

1111
01:04:47,319 --> 01:04:48,000
are there.

1112
01:04:48,000 --> 01:04:49,000
Audrow Nash: Cool.

1113
01:04:49,000 --> 01:04:51,871
Saman Farid: because you have a
bunch of people who are

1114
01:04:51,871 --> 01:04:55,003
interacting with that machine.
and so you need, really clear

1115
01:04:55,003 --> 01:04:58,343
instructions for how to interact
with the robot. If you've never

1116
01:04:58,343 --> 01:05:01,527
been around a robot, how do you
start it? How do you stop it?

1117
01:05:01,527 --> 01:05:04,607
What do you do in this error?
What are the different lights

1118
01:05:04,607 --> 01:05:07,843
mean? and then we also put, QR
codes on all of the components.

1119
01:05:07,843 --> 01:05:08,000
and

1120
01:05:08,000 --> 01:05:10,000
Audrow Nash: Oh, that's

1121
01:05:10,000 --> 01:05:14,047
Saman Farid: they can scan the
operators who are on site can

1122
01:05:14,047 --> 01:05:17,824
scan a part and then, it
notifies us, right? And, we can

1123
01:05:17,824 --> 01:05:22,073
log in via telepresence and we
can ask that operator questions.

1124
01:05:22,073 --> 01:05:25,917
and we can also like surface
little video files that say,

1125
01:05:25,917 --> 01:05:30,031
Hey, here are all the things
that you can do to, resolve this

1126
01:05:30,031 --> 01:05:34,010
issue, right? if a suction cup
is blocked, Check and see if

1127
01:05:34,010 --> 01:05:37,922
there's something, if there's
debris in there, or if there

1128
01:05:37,922 --> 01:05:42,104
isn't, hit this reset button or
call us or we'll come on site,

1129
01:05:42,104 --> 01:05:46,016
or whatever it might be. we
general, so we, generate those

1130
01:05:46,016 --> 01:05:50,197
procedures upfront. and then we
have a bunch of software, that

1131
01:05:50,197 --> 01:05:53,704
we call Colony, which is
basically our kind of robot

1132
01:05:53,704 --> 01:05:57,684
management tool set, which
allows us to remotely manage and

1133
01:05:57,684 --> 01:06:01,595
monitor all of these systems
that we've deployed. we every

1134
01:06:01,595 --> 01:06:02,000
system

1135
01:06:02,000 --> 01:06:04,542
Audrow Nash: Did you build that?
That's internal, like it's a

1136
01:06:04,542 --> 01:06:06,000
tool you guys have developed.
Okay.

1137
01:06:06,000 --> 01:06:09,366
Saman Farid: Yeah. We built it
internally. we tried to not have

1138
01:06:09,366 --> 01:06:12,679
to, we tried to evaluate other,
other ones that there were out

1139
01:06:12,679 --> 01:06:13,000
there,

1140
01:06:13,000 --> 01:06:14,000
Audrow Nash: didn't meet your
needs.

1141
01:06:14,000 --> 01:06:17,767
Saman Farid: they didn't meet
exactly the things that we

1142
01:06:17,767 --> 01:06:21,803
needed. Yeah, exactly. because
our use case is a little bit,

1143
01:06:21,803 --> 01:06:25,973
complex, right? one is, we need
to not just monitor the robot,

1144
01:06:25,973 --> 01:06:29,807
we need to monitor all the
peripherals, right? so all the

1145
01:06:29,807 --> 01:06:33,776
server servos and PLCs and
conveyance, all that needs to be

1146
01:06:33,776 --> 01:06:37,744
monitored as well. there's
complexity there. Number two, we

1147
01:06:37,744 --> 01:06:41,578
kinda need video data, from
multiple cameras on that work

1148
01:06:41,578 --> 01:06:45,211
cell, at all times. Three, we
need to be able to inter

1149
01:06:45,211 --> 01:06:49,381
intervene when there's an issue.
and so about 60 or 70% of the

1150
01:06:49,381 --> 01:06:53,283
time, if there is an error on
the robot, we can resolve it

1151
01:06:53,283 --> 01:06:57,453
remotely, without having to come
on site. and so that saves, a

1152
01:06:57,453 --> 01:06:59,000
bunch of time and cost.

1153
01:06:59,000 --> 01:07:01,000
Audrow Nash: oh, for sure. Yeah.

1154
01:07:01,000 --> 01:07:05,091
Saman Farid: and we just had all
of these kind of relatively

1155
01:07:05,091 --> 01:07:07,000
unique needs, on our, robot,

1156
01:07:07,000 --> 01:07:08,000
Audrow Nash: didn't exist that
well.

1157
01:07:08,000 --> 01:07:11,000
Saman Farid: Yeah.

1158
01:07:11,000 --> 01:07:14,940
Audrow Nash: very cool. Do you
have, when someone has to

1159
01:07:14,940 --> 01:07:19,092
intervene, is it like, so do you
know plus one robotics and

1160
01:07:19,092 --> 01:07:23,455
they're set up so they have a
pretty sophisticated, to me, the

1161
01:07:23,455 --> 01:07:27,817
core competence, maybe not core
competence, but one of the big

1162
01:07:27,817 --> 01:07:31,899
differentiating things about
them is that they have a very

1163
01:07:31,899 --> 01:07:36,120
good workflow around human in
the loop for when their robots

1164
01:07:36,120 --> 01:07:40,202
get stuck. so they're trying to
pick up a box that I taped

1165
01:07:40,202 --> 01:07:44,564
together, to ship back an Amazon
package, and it's total chaos

1166
01:07:44,564 --> 01:07:48,997
and it's falling apart. And so
then they have a human say where

1167
01:07:48,997 --> 01:07:53,501
you can pick it up or should it
be discarded or whatever. Or not

1168
01:07:53,501 --> 01:07:57,652
discarded, but reprocessed or
something. do you guys have a

1169
01:07:57,652 --> 01:08:01,874
similar human in the loop system
or how do you, handle these

1170
01:08:01,874 --> 01:08:03,000
kinds of things?

1171
01:08:03,000 --> 01:08:07,008
Saman Farid: Yeah. So for
handling edge cases, we do have,

1172
01:08:07,008 --> 01:08:11,154
humans in the loop. So we have
a, our service engineers that

1173
01:08:11,154 --> 01:08:15,438
are, kinda monitoring all the
systems 24 7. If there's an edge

1174
01:08:15,438 --> 01:08:19,791
case that needs human attention,
it pings one of them. They log

1175
01:08:19,791 --> 01:08:23,385
in and they can remotely
support. the, the amount of

1176
01:08:23,385 --> 01:08:27,738
things that a remote, person can
do is quite limited though, so

1177
01:08:27,738 --> 01:08:32,022
we do have to kind of design for
it upfront, right? we need to

1178
01:08:32,022 --> 01:08:36,306
put in place procedures for what
happens if different kinds of

1179
01:08:36,306 --> 01:08:40,383
errors are tripped. to your
example about tape, like one of

1180
01:08:40,383 --> 01:08:44,598
the most common failure modes
for a palletizer, which is, one

1181
01:08:44,598 --> 01:08:48,744
of the simplest system types
that we deploy, the most common

1182
01:08:48,744 --> 01:08:53,167
failure mode is that the box has
not been taped properly. So the

1183
01:08:53,167 --> 01:08:56,000
robot picks up the box, but if
there's no

1184
01:08:56,000 --> 01:08:57,000
Audrow Nash: lips up or

1185
01:08:57,000 --> 01:09:00,751
Saman Farid: Yeah. Or if there's
no tape on the bottom, all the

1186
01:09:00,751 --> 01:09:04,323
contents just spill out. so
that's a common, that's a common

1187
01:09:04,323 --> 01:09:07,598
occurrence. and it's, not the
robot's fault per se, but

1188
01:09:07,598 --> 01:09:11,051
because we're committed to
overall uptime and performance,

1189
01:09:11,051 --> 01:09:14,505
like we still need to detect
that situation, and avoid it,

1190
01:09:14,505 --> 01:09:18,196
before the other stuff. So we,
put, we, we have a vision model

1191
01:09:18,196 --> 01:09:21,292
that, trains tape on tape
detection and or flap open

1192
01:09:21,292 --> 01:09:24,984
detection. and if it detects the
tape is, missing or the flaps

1193
01:09:24,984 --> 01:09:28,735
are open, it either pauses the
system or if there's a place for

1194
01:09:28,735 --> 01:09:32,426
it to push to the side and keep
running, then it will do that.

1195
01:09:32,426 --> 01:09:35,641
And there's a, there, every
system type has a bunch of

1196
01:09:35,641 --> 01:09:39,452
different failure modes you need
to account for. I need to build

1197
01:09:39,452 --> 01:09:41,000
it, build into the system.

1198
01:09:41,000 --> 01:09:45,570
Audrow Nash: Now, let's see.
Going back a little bit, when

1199
01:09:45,570 --> 01:09:50,455
you said you build a document
like a technical or some sort of

1200
01:09:50,455 --> 01:09:55,418
document for the workers so that
they have exact information on

1201
01:09:55,418 --> 01:10:00,145
the system you've set up, I
imagine that's fairly similar to

1202
01:10:00,145 --> 01:10:04,952
a number of the things we've
talked about, where it's largely

1203
01:10:04,952 --> 01:10:09,206
template, template driven, and
then also like you have

1204
01:10:09,206 --> 01:10:13,697
templated text that you put
throughout, but then you also

1205
01:10:13,697 --> 01:10:18,503
have custom text using LLMs for
this kind of thing. Is it the

1206
01:10:18,503 --> 01:10:20,000
same kind of thing?

1207
01:10:20,000 --> 01:10:23,161
Saman Farid: yes, there's a
little bit of it that's a

1208
01:10:23,161 --> 01:10:26,144
automatically generated,
although there's a human,

1209
01:10:26,144 --> 01:10:29,782
heavily involved in that. but
I'll say that the challenge is,

1210
01:10:29,782 --> 01:10:32,944
in, a real production
environment, it's very unlikely

1211
01:10:32,944 --> 01:10:36,463
that somebody's gonna stop what
they're doing and go read a

1212
01:10:36,463 --> 01:10:37,000
handbook.

1213
01:10:37,000 --> 01:10:39,835
Audrow Nash: Yeah, but you
probably want quick reference

1214
01:10:39,835 --> 01:10:41,000
points for this kind of

1215
01:10:41,000 --> 01:10:45,609
Saman Farid: so the way we
generally do that is with video.

1216
01:10:45,609 --> 01:10:46,000
so we

1217
01:10:46,000 --> 01:10:47,000
Audrow Nash: Oh,

1218
01:10:46,000 --> 01:10:50,346
Saman Farid: a. 10 or so, 10 to
20 different kind of common

1219
01:10:50,346 --> 01:10:54,470
failure modes, that we've
accounted for. and we just pop

1220
01:10:54,470 --> 01:10:58,963
up a little video on the screen
of that robot that says, Hey,

1221
01:10:58,963 --> 01:11:03,530
here are here. Here's what we
think might be happening. here's

1222
01:11:03,530 --> 01:11:08,096
the steps that, that you take,
and, hit this button if you, if

1223
01:11:08,096 --> 01:11:12,000
it's not resolved, and it'll
call us. and then we can

1224
01:11:12,000 --> 01:11:15,395
Audrow Nash: That's so cool. So
you like pull up the YouTube

1225
01:11:15,395 --> 01:11:19,016
video they need to see, but it's
internal probably. That's super

1226
01:11:19,016 --> 01:11:22,127
cool. So yeah, you're not
focusing on the documentation

1227
01:11:22,127 --> 01:11:25,465
because they're not gonna use
that to solve things. But you

1228
01:11:25,465 --> 01:11:28,803
have uptime requirements and so
you have a very streamlined

1229
01:11:28,803 --> 01:11:32,198
process for them, which often
is, this is a video of exactly

1230
01:11:32,198 --> 01:11:35,310
what you need. And you have
through your data analysis,

1231
01:11:35,310 --> 01:11:38,648
through collecting data, you
understand a lot of the common

1232
01:11:38,648 --> 01:11:42,042
things that may go wrong. And
you probably have some ability

1233
01:11:42,042 --> 01:11:45,211
to introspect the robot to
understand what may go wrong.

1234
01:11:45,211 --> 01:11:48,888
And then you pull up the correct
video, they watch it, they clear

1235
01:11:48,888 --> 01:11:52,000
it out, or they call you guys,
if they cannot figure it

1236
01:11:52,000 --> 01:11:54,797
Saman Farid: The most common
sequence of events is that the

1237
01:11:54,797 --> 01:11:57,641
first thing that people do when
the system goes down is they

1238
01:11:57,641 --> 01:12:00,628
call us, right? So they'll call,
we have a one 800 number, they

1239
01:12:00,628 --> 01:12:03,424
call us, one of our technicians
answers. they ask 'em a few

1240
01:12:03,424 --> 01:12:06,174
questions and they say, okay,
here's the issue. and we can

1241
01:12:06,174 --> 01:12:08,876
remotely trigger the video to
start getting played on the

1242
01:12:08,876 --> 01:12:11,767
screen. So we say, go to the
screen, look at this video, it's

1243
01:12:11,767 --> 01:12:13,000
gonna tell you what to do.

1244
01:12:13,000 --> 01:12:16,889
Audrow Nash: When are you gonna
do LLMs, voice for your

1245
01:12:16,889 --> 01:12:20,000
technician to just direct them
to the video?

1246
01:12:20,000 --> 01:12:22,291
Saman Farid: I think we're, I
think we're a little bit of a

1247
01:12:22,291 --> 01:12:24,000
ways away from that still, but
hopefully one

1248
01:12:24,000 --> 01:12:26,000
Audrow Nash: Yeah. Yeah.
Hopefully one day

1249
01:12:26,000 --> 01:12:31,394
Audrow Nash: . I wonder, for, so
one thing that is interesting to

1250
01:12:31,394 --> 01:12:36,704
me about you guys so far is that
it seems like you're doing your

1251
01:12:36,704 --> 01:12:41,185
technical support for your
customers in-House. several

1252
01:12:41,185 --> 01:12:46,247
other companies that I, don't
have scaled and maybe are a bit

1253
01:12:46,247 --> 01:12:51,143
further along, they seem to
outsource technical support for

1254
01:12:51,143 --> 01:12:56,287
this kind of thing. And I'm just
wondering about, I guess your

1255
01:12:56,287 --> 01:13:01,598
view on technical support and is
it something that should always

1256
01:13:01,598 --> 01:13:05,000
be in-house or how do you think
about it?

1257
01:13:05,000 --> 01:13:08,086
Saman Farid: yeah, I, am, I'm
gonna sound a little bit like an

1258
01:13:08,086 --> 01:13:11,222
asshole, saying this, but, who's
further along? I don't know of

1259
01:13:11,222 --> 01:13:13,910
any other robotics startups.
They're in a hundred plus

1260
01:13:13,910 --> 01:13:16,000
manufacturing or a hundred plus
facilities

1261
01:13:16,000 --> 01:13:18,000
Audrow Nash: Not in
manufacturing. I'm thinking

1262
01:13:18,000 --> 01:13:21,619
Saman Farid: in right? They're
typically in three to five

1263
01:13:21,619 --> 01:13:25,429
facilities. Maybe 10, maybe 20
if they're really pushing it.

1264
01:13:25,429 --> 01:13:28,730
There's very few robotics
startups that actually are

1265
01:13:28,730 --> 01:13:32,540
managing a fleet of systems
across a hundred, a hundred plus

1266
01:13:32,540 --> 01:13:36,349
facilities. and so, I'm not,
saying that to push, right? But

1267
01:13:36,349 --> 01:13:38,000
what I'm saying to say is,

1268
01:13:38,000 --> 01:13:39,000
Audrow Nash: yeah,

1269
01:13:39,000 --> 01:13:41,789
Saman Farid: I think that
outsourcing service and

1270
01:13:41,789 --> 01:13:45,147
maintenance is one of those
things that sounds nice, right?

1271
01:13:45,147 --> 01:13:48,391
A lot of people are like, yeah,
I, can't scale if I do it

1272
01:13:48,391 --> 01:13:51,977
myself. and that's just a very
common like fallacy that I think

1273
01:13:51,977 --> 01:13:55,619
a lot of robotics companies fall
into where they say, I wanna do

1274
01:13:55,619 --> 01:13:58,807
the things that scale, which
leads them to, I just wanna

1275
01:13:58,807 --> 01:14:02,278
provide an API and I won't let
somebody else do all the work.

1276
01:14:02,278 --> 01:14:05,466
And it's that's not the hard
part. That's the hard part,

1277
01:14:05,466 --> 01:14:08,539
right? doing service and
maintenance is the hard part.

1278
01:14:08,539 --> 01:14:12,239
That's the, whole key, right? If
you don't do that, your solution

1279
01:14:12,239 --> 01:14:15,483
is worthless. you can't just
drop it off and give it to a

1280
01:14:15,483 --> 01:14:19,069
third party. and I think this is
where, if I have any advice to

1281
01:14:19,069 --> 01:14:22,711
give to robotics companies, it's
don't, think you're too good to

1282
01:14:22,711 --> 01:14:25,728
do maintenance. don't think
you're too good to write,

1283
01:14:25,728 --> 01:14:29,086
documentation, don't think
you're too good, to go in custom

1284
01:14:29,086 --> 01:14:32,558
design, end of arm tool. 'cause
those are the things that are

1285
01:14:32,558 --> 01:14:35,972
necessary to drive, massive
adoption. Deploying systems in a

1286
01:14:35,972 --> 01:14:39,330
hundred plus facilities is a
herculean task. that is, there

1287
01:14:39,330 --> 01:14:42,859
may be other companies that have
more robots deployed than us,

1288
01:14:42,859 --> 01:14:46,331
but they may, have a hundred
robots deployed at one facility,

1289
01:14:46,331 --> 01:14:49,461
for example, right? that's a
totally different class of

1290
01:14:49,461 --> 01:14:52,478
problem than having robots
deployed at a hundred plus

1291
01:14:52,478 --> 01:14:55,665
facilities, right? every
facility has a bunch of its own

1292
01:14:55,665 --> 01:14:58,739
idiosyncrasies. You have
different people, you have to

1293
01:14:58,739 --> 01:15:01,869
interact with, different
operators with smaller amounts

1294
01:15:01,869 --> 01:15:05,056
of training, less onsite
support. So as a, I think we've

1295
01:15:05,056 --> 01:15:08,130
solved, we've gone much farther
on solving the kind of

1296
01:15:08,130 --> 01:15:11,659
post-deployment issues than I
think anybody else on the planet

1297
01:15:11,659 --> 01:15:12,000
today.

1298
01:15:12,000 --> 01:15:15,835
Audrow Nash: Hell yeah. Okay.
Super cool. Yeah. So you view it

1299
01:15:15,835 --> 01:15:19,485
as a core competence of the
company to basically be able to

1300
01:15:19,485 --> 01:15:23,134
service its robots over time.
'cause you're right, if they,

1301
01:15:23,134 --> 01:15:26,907
you can't really just drop 'em
off and if the people that are

1302
01:15:26,907 --> 01:15:30,619
supposed to maintain them are
not super good and you may get

1303
01:15:30,619 --> 01:15:34,392
some variation in quality, then
that's a mixed experience for

1304
01:15:34,392 --> 01:15:36,000
customers for this kind of

1305
01:15:36,000 --> 01:15:39,357
Saman Farid: And I think if it
depends what your goal is,

1306
01:15:39,357 --> 01:15:42,774
right? If your goal is to sell
robots and get them out the

1307
01:15:42,774 --> 01:15:46,484
door, then maybe you don't need
to do what we're doing. If your

1308
01:15:46,484 --> 01:15:50,195
goal is to have robots deployed
and be useful for a long period

1309
01:15:50,195 --> 01:15:53,906
of time, and like what, 'cause
the, metric that we track is not

1310
01:15:53,906 --> 01:15:57,263
sold units. The metric that we
track is usage hours every

1311
01:15:57,263 --> 01:16:00,915
month, every day, every morning
we look at what are the number

1312
01:16:00,915 --> 01:16:04,626
of usage hours from yesterday on
our robots? Did they, actually

1313
01:16:04,626 --> 01:16:08,277
do useful things for, 10 hours a
day or one hour a day? and if

1314
01:16:08,277 --> 01:16:11,812
your metric is that, then you
have to be responsible for the

1315
01:16:11,812 --> 01:16:15,581
service of that system. if your
metric is, did I sell things and

1316
01:16:15,581 --> 01:16:18,938
get them out the door, then
yeah, you don't need to track

1317
01:16:18,938 --> 01:16:21,000
how they did once you've sold
them.

1318
01:16:21,000 --> 01:16:25,105
Audrow Nash: Yeah. Yeah. I,
really like the idea of using

1319
01:16:25,105 --> 01:16:29,570
usage hours as your proxy for
success, because a lot of times,

1320
01:16:29,570 --> 01:16:34,034
I think, a lot of times with new
technology, what happens is a

1321
01:16:34,034 --> 01:16:38,499
lot of bigger companies, like
huge companies in manufacturing,

1322
01:16:38,499 --> 01:16:42,532
but agriculture and other
places, they'll buy technology

1323
01:16:42,532 --> 01:16:47,141
just to try it out. And it looks
promising for the startups that

1324
01:16:47,141 --> 01:16:51,606
are getting these early trials,
especially if it's a prominent

1325
01:16:51,606 --> 01:16:55,926
company. But then. It's not
always a land and expand. It may

1326
01:16:55,926 --> 01:17:00,463
be a try and, or we'll just keep
you at the same level for this

1327
01:17:00,463 --> 01:17:04,712
kind of thing to see how it
develops over time or something

1328
01:17:04,712 --> 01:17:05,000
like

1329
01:17:05,000 --> 01:17:09,136
Saman Farid: Yeah. I think the
most damning thing is a lot of

1330
01:17:09,136 --> 01:17:13,000
these robots get deployed into
innovation centers, right?

1331
01:17:13,000 --> 01:17:14,000
Audrow Nash: ha,

1332
01:17:14,000 --> 01:17:17,782
Saman Farid: environments. And
it, they think of it as a point

1333
01:17:17,782 --> 01:17:21,076
of pride where it's oh, I got
deployed in, so and so's

1334
01:17:21,076 --> 01:17:24,675
innovation center, like Proctor
and Gamble is notorious for

1335
01:17:24,675 --> 01:17:28,580
this, right? Proctor and Gamble
has this giant innovation center

1336
01:17:28,580 --> 01:17:32,362
of stuff that they will buy and
show off and then never use in

1337
01:17:32,362 --> 01:17:35,717
production. and so we have
flatly refused any, anything

1338
01:17:35,717 --> 01:17:39,255
like that. Like our systems are
only ever deployed in real

1339
01:17:39,255 --> 01:17:42,000
production environments. we
never do proof of

1340
01:17:42,000 --> 01:17:43,000
Audrow Nash: Love it.

1341
01:17:42,000 --> 01:17:46,384
Saman Farid: if it's not being
used 20 hours a day plus, we're

1342
01:17:46,384 --> 01:17:49,000
not interested in doing that
project.

1343
01:17:49,000 --> 01:17:54,097
Audrow Nash: Do you, so at your
current stage being in a hundred

1344
01:17:54,097 --> 01:17:59,035
facilities, are you, I, imagine
you're quite, are, you careful

1345
01:17:59,035 --> 01:18:03,973
about who you deploy to or who
you end up working with? do you

1346
01:18:03,973 --> 01:18:08,912
vet them for similarity or how
do you choose, in your scaling,

1347
01:18:08,912 --> 01:18:13,531
how do you choose to scale? or
like what decisions are you

1348
01:18:13,531 --> 01:18:16,000
making while scaling, I suppose

1349
01:18:16,000 --> 01:18:20,532
Saman Farid: Yeah. we talked a
little bit about what are the

1350
01:18:20,532 --> 01:18:24,612
task types that we, are
comfortable with. so the first

1351
01:18:24,612 --> 01:18:28,993
thing and the most important
thing is to validate that the

1352
01:18:28,993 --> 01:18:33,601
task is a fit for the things
that we feel like we can hit 99%

1353
01:18:33,601 --> 01:18:37,000
uptime on. so there's a list of
those things.

1354
01:18:37,000 --> 01:18:40,000
Audrow Nash: is that your 99% is
what you're going for

1355
01:18:40,000 --> 01:18:44,000
Saman Farid: 99 point, nine,
eight is our,

1356
01:18:44,000 --> 01:18:46,000
Audrow Nash: Ah, that's so
exciting.

1357
01:18:46,000 --> 01:18:50,247
Saman Farid: past 12 months. Our
uptime on all of our deployed

1358
01:18:50,247 --> 01:18:51,000
systems is,

1359
01:18:51,000 --> 01:18:52,000
Audrow Nash: Oh, hell yeah.

1360
01:18:52,000 --> 01:18:54,642
Saman Farid: point, 99.85, I
think is, the last number that

1361
01:18:54,642 --> 01:18:55,000
we have.

1362
01:18:55,000 --> 01:18:58,000
Audrow Nash: 9, 9, 8, 5. That's
awesome.

1363
01:18:58,000 --> 01:19:01,372
Saman Farid: and that's across
all of our deployed systems, or

1364
01:19:01,372 --> 01:19:04,145
the trailing 12 months number.
we're, tracking that

1365
01:19:04,145 --> 01:19:07,299
religiously. That's another part
where, a lot of, robotics

1366
01:19:07,299 --> 01:19:10,290
companies fall short, right?
With, it meets performance

1367
01:19:10,290 --> 01:19:13,662
targets, and then they have a
week of downtime, right? And all

1368
01:19:13,662 --> 01:19:16,000
of a sudden, that number goes
out the door.

1369
01:19:16,000 --> 01:19:17,000
Audrow Nash: Yeah.

1370
01:19:17,000 --> 01:19:20,872
Saman Farid: sorry. you you're
asking about what kinds of

1371
01:19:20,872 --> 01:19:24,879
customers do we go after? So
one, like we have to make sure

1372
01:19:24,879 --> 01:19:29,294
that the tasks that fit for what
we can accomplish. number two is

1373
01:19:29,294 --> 01:19:33,574
we do have a kind of, a specific
type of ICP that we're working

1374
01:19:33,574 --> 01:19:35,000
with today. typically

1375
01:19:35,000 --> 01:19:36,000
Audrow Nash: What's ICP.

1376
01:19:36,000 --> 01:19:38,000
Saman Farid: Ideal customer
profile. Sorry. so

1377
01:19:38,000 --> 01:19:41,538
Saman Farid: our ideal customer
profile today are, small to

1378
01:19:41,538 --> 01:19:45,317
medium sized manufacturers. So
between a hundred and a thousand

1379
01:19:45,317 --> 01:19:49,035
employees, in the facility, is a
good fit for us. So we're not

1380
01:19:49,035 --> 01:19:52,753
selling to, the fortune fifties
of the world. we're selling to

1381
01:19:52,753 --> 01:19:56,351
all of their suppliers. and we
have a kind of very specific,

1382
01:19:56,351 --> 01:20:00,130
reason for that. primarily it's
about speed of decision making.

1383
01:20:00,130 --> 01:20:03,488
but, all that to say, like we we
qualify from a business

1384
01:20:03,488 --> 01:20:07,206
perspective, we qualify from a
technical perspective, and then

1385
01:20:07,206 --> 01:20:10,445
we qualify also from a
operational perspective, right?

1386
01:20:10,445 --> 01:20:14,043
Like, how many hours a day do
they run their plant today? If

1387
01:20:14,043 --> 01:20:17,821
they increase that, would it be
valuable to 'em or not? Like we

1388
01:20:17,821 --> 01:20:21,000
have to validate all the
business side of things too.

1389
01:20:21,000 --> 01:20:25,200
Audrow Nash: so you look for a
good opportunity in a sense with

1390
01:20:25,200 --> 01:20:28,933
the company. So if they're
already say, whatever they're

1391
01:20:28,933 --> 01:20:32,600
doing is like super efficient
already and they probably

1392
01:20:32,600 --> 01:20:36,533
wouldn't benefit that much from
this. then you don't go for

1393
01:20:36,533 --> 01:20:40,667
them. Or they wouldn't be in as
high a priority, a customer to

1394
01:20:40,667 --> 01:20:44,467
deploy to for this kind of
thing. Okay. What do you think

1395
01:20:44,467 --> 01:20:48,267
your, growth curve will look
like over the next year, two

1396
01:20:48,267 --> 01:20:51,000
years, five years, these kinds
of things?

1397
01:20:51,000 --> 01:20:56,034
Saman Farid: yeah, I think
we're, growing, probably like

1398
01:20:56,034 --> 01:20:59,000
three to four X a year right
now.

1399
01:20:59,000 --> 01:21:00,000
Audrow Nash: That's

1400
01:21:00,000 --> 01:21:02,000
Saman Farid: our goal is to
improve

1401
01:21:02,000 --> 01:21:03,000
Audrow Nash: Keep it doing that.

1402
01:21:03,000 --> 01:21:04,000
Saman Farid: go,

1403
01:21:04,000 --> 01:21:05,000
Audrow Nash: Improve that. Hell
yeah.

1404
01:21:05,000 --> 01:21:07,000
Saman Farid: yeah, We wanna get
higher than that.

1405
01:21:07,000 --> 01:21:09,000
Audrow Nash: do you

1406
01:21:09,000 --> 01:21:12,689
Saman Farid: the rate of growth
is increasing because Our sales

1407
01:21:12,689 --> 01:21:16,261
team continues to sell kind of
new systems, to new customers.

1408
01:21:16,261 --> 01:21:19,892
But, one of the areas of a lot
of growth for us exactly, is in

1409
01:21:19,892 --> 01:21:23,347
our existing customers, the
more. And deployed systems that

1410
01:21:23,347 --> 01:21:26,450
we have. Each of those
opportunities starts to triple

1411
01:21:26,450 --> 01:21:30,140
or quadruple as they expand that
solution to multiple lines and

1412
01:21:30,140 --> 01:21:33,126
then move, the upstream or
downstream, in that same

1413
01:21:33,126 --> 01:21:35,000
production line to do new tasks.

1414
01:21:35,000 --> 01:21:38,719
Audrow Nash: That's super cool.
Yeah. Yeah. The land and expand

1415
01:21:38,719 --> 01:21:42,438
is a really powerful thing. It's
interesting to me that you are

1416
01:21:42,438 --> 01:21:45,802
going for smaller companies,
like a hundred to a thousand

1417
01:21:45,802 --> 01:21:49,226
people. 'cause, and I, bet you
the decision making is much

1418
01:21:49,226 --> 01:21:52,000
faster. So that's probably
better for you guys.

1419
01:21:52,000 --> 01:21:53,000
Saman Farid: Yeah,

1420
01:21:53,000 --> 01:21:57,436
Audrow Nash: where instead of
like really long sales cycles

1421
01:21:57,436 --> 01:22:01,947
for these large companies, you
get to ship much faster after

1422
01:22:01,947 --> 01:22:03,000
you sell them.

1423
01:22:03,000 --> 01:22:06,750
Saman Farid: the other thing is
that these smaller companies

1424
01:22:06,750 --> 01:22:10,000
don't have innovation centers,
right? So there's no,

1425
01:22:10,000 --> 01:22:11,000
Audrow Nash: Oh, you don't wanna
be pigeonholed

1426
01:22:11,000 --> 01:22:14,868
Saman Farid: no, it's less than,
it's less that and more, we need

1427
01:22:14,868 --> 01:22:17,784
people who are actively
concerned with production

1428
01:22:17,784 --> 01:22:21,057
outcomes, not, innovation
theater. And I think a lot of

1429
01:22:21,057 --> 01:22:24,627
the bigger companies have teams
of people that do innovation

1430
01:22:24,627 --> 01:22:28,078
theater where they say, oh,
let's do a Silicon Valley tour

1431
01:22:28,078 --> 01:22:31,887
and let's have a facility where
we demo all this stuff and let's

1432
01:22:31,887 --> 01:22:35,398
show off to our executives and
we have a earnings call next

1433
01:22:35,398 --> 01:22:39,206
week, and we're gonna talk about
ai. most of our customers don't

1434
01:22:39,206 --> 01:22:42,836
do any of that stuff. It's I
produced 200 parts to yesterday.

1435
01:22:42,836 --> 01:22:46,228
I need to produce 210 parts
today, otherwise we're out of

1436
01:22:46,228 --> 01:22:49,620
business. and so like we're
focused on customers that are

1437
01:22:49,620 --> 01:22:52,000
much more practical in their,
worldview.

1438
01:22:52,000 --> 01:22:56,214
Audrow Nash: Oh hell yeah.
That's great. That's so funny.

1439
01:22:56,214 --> 01:23:00,724
The innovation hubs and the, or
the innovation, facilities or

1440
01:23:00,724 --> 01:23:05,381
whatever it is that they have.
So silly. But yeah, it's they're

1441
01:23:05,381 --> 01:23:09,817
just looking like they're being
innovative, not really being

1442
01:23:09,817 --> 01:23:11,000
that innovative.

1443
01:23:11,000 --> 01:23:12,000
Saman Farid: Yeah.

1444
01:23:12,000 --> 01:23:15,664
Audrow Nash: So if you grow
three times this year, you'll

1445
01:23:15,664 --> 01:23:19,714
have, do you think you'll have
300? Deploy, facilities deployed

1446
01:23:19,714 --> 01:23:21,000
in a year, or is it,

1447
01:23:21,000 --> 01:23:22,000
Saman Farid: we may

1448
01:23:22,000 --> 01:23:24,000
Audrow Nash: might have three
times the number of robots

1449
01:23:24,000 --> 01:23:25,000
Saman Farid: Yeah. I think it'll
be

1450
01:23:25,000 --> 01:23:26,000
Audrow Nash: land and

1451
01:23:26,000 --> 01:23:29,310
Saman Farid: of the Exactly.
Yeah. We may not be in three

1452
01:23:29,310 --> 01:23:32,735
times the number of facilities,
but we should have, roughly

1453
01:23:32,735 --> 01:23:35,000
three times the number of
robots. Yeah.

1454
01:23:35,000 --> 01:23:40,648
Audrow Nash: Let's see. Looking
forward, what are some of the

1455
01:23:40,648 --> 01:23:45,648
hard challenges that you guys
still are gonna be going

1456
01:23:45,648 --> 01:23:50,000
through? like where, are the
risks to you guys,

1457
01:23:50,000 --> 01:23:53,651
Saman Farid: I think that there
are a lot of areas where we

1458
01:23:53,651 --> 01:23:57,425
could be doing better, in the,
the first area is what are the

1459
01:23:57,425 --> 01:24:01,323
kinds of tasks that we're trying
to automate. and as we, expand

1460
01:24:01,323 --> 01:24:04,540
that list, the level of
complexity goes up, more and

1461
01:24:04,540 --> 01:24:08,253
more, right? So each incremental
task requires us to, change

1462
01:24:08,253 --> 01:24:11,718
everything in our process,
right? what are the questions

1463
01:24:11,718 --> 01:24:15,554
that we ask upfront? what are
the, site evaluation things that

1464
01:24:15,554 --> 01:24:19,205
need to be validated? What are
the kinds of robots that are

1465
01:24:19,205 --> 01:24:22,608
available for that task? What
are the, post-deployment,

1466
01:24:22,608 --> 01:24:26,320
considerations? what are the
training videos that we need to

1467
01:24:26,320 --> 01:24:30,157
create, et cetera. So like every
new task, has a bunch of work

1468
01:24:30,157 --> 01:24:33,683
associated with it. so we're
getting better and faster at

1469
01:24:33,683 --> 01:24:37,148
doing that. But, like our
customers expect us to be able

1470
01:24:37,148 --> 01:24:40,923
to solve all problems. Like when
we show up and we deploy one

1471
01:24:40,923 --> 01:24:44,759
robot, typically the immediate
next thing that happens is, oh,

1472
01:24:44,759 --> 01:24:48,348
while you're here, can you do
this? And that and the other

1473
01:24:48,348 --> 01:24:52,308
thing. and, as it's, not, quite
that simple yet. but that is one

1474
01:24:52,308 --> 01:24:55,773
of our goals is to be able to,
solve as many problems as

1475
01:24:55,773 --> 01:24:58,000
possible for our existing
customers.

1476
01:24:58,000 --> 01:25:01,921
Audrow Nash: How do you stay
focused with that? Because it's

1477
01:25:01,921 --> 01:25:06,038
a hard thing. 'cause I know, a
lot of times when people talk to

1478
01:25:06,038 --> 01:25:10,285
customers, those customers don't
really have a good model of what

1479
01:25:10,285 --> 01:25:14,533
robotics can do. And so they may
suggest outlandish things. And I

1480
01:25:14,533 --> 01:25:18,453
suppose you probably wouldn't
take on any of the like, super

1481
01:25:18,453 --> 01:25:22,570
difficult robotics things that
probably are like, it might even

1482
01:25:22,570 --> 01:25:26,556
be just specific to them too. So
like super hard and very low

1483
01:25:26,556 --> 01:25:29,758
value. but how, do you
prioritize the things that

1484
01:25:29,758 --> 01:25:31,000
customers may want?

1485
01:25:31,000 --> 01:25:36,757
Saman Farid: yeah. so we, do a
census every once in a while On

1486
01:25:36,757 --> 01:25:42,514
all the things that people have
asked us to do that we've said

1487
01:25:42,514 --> 01:25:44,000
no to. and we're

1488
01:25:44,000 --> 01:25:45,000
Audrow Nash: you look for low
hanging fruit

1489
01:25:45,000 --> 01:25:48,874
Saman Farid: Yeah. We say, okay,
if we could provide this type of

1490
01:25:48,874 --> 01:25:51,974
case packer, here are 20
customers that have already

1491
01:25:51,974 --> 01:25:54,000
asked for it, and here are 10
more

1492
01:25:54,000 --> 01:25:55,000
Audrow Nash: Oh.

1493
01:25:55,000 --> 01:25:58,256
Saman Farid: we could offer it
to. and so this is like a

1494
01:25:58,256 --> 01:26:00,000
worthwhile opportunity for us.

1495
01:26:00,000 --> 01:26:04,348
Audrow Nash: Cool. I really like
that way of prioritizing or

1496
01:26:04,348 --> 01:26:08,623
keeping track of things, for
things you can expand to next.

1497
01:26:08,623 --> 01:26:10,000
This kind of thing,

1498
01:26:10,000 --> 01:26:13,591
Saman Farid: Yeah. Because of
what we're not trying to do is

1499
01:26:13,591 --> 01:26:16,942
go and sell to new customer
types. I think, the focus is

1500
01:26:16,942 --> 01:26:20,173
who, are we currently
successfully selling to and what

1501
01:26:20,173 --> 01:26:23,764
are the additional needs that
they have that we haven't been

1502
01:26:23,764 --> 01:26:27,175
able to meet? and so this is
where that, the kind of site

1503
01:26:27,175 --> 01:26:30,706
scan, for example, comes in
really handy. ' cause we have a

1504
01:26:30,706 --> 01:26:34,476
3D LIDAR scan of every facility
that we've walked. and it's not

1505
01:26:34,476 --> 01:26:38,066
just of we, we don't just scan
that one part, right? We scan

1506
01:26:38,066 --> 01:26:41,836
the whole facility. so we can go
back in time and say, what was

1507
01:26:41,836 --> 01:26:45,008
the, what, were they doing
upstream of, this robot or

1508
01:26:45,008 --> 01:26:48,000
downstream? we have that, full
scan on our, in our

1509
01:26:48,000 --> 01:26:52,984
Audrow Nash: Yeah. That's nice.
'cause then you don't have to

1510
01:26:52,984 --> 01:26:57,967
get the data multiple times. And
that makes things a bit more

1511
01:26:57,967 --> 01:27:03,114
efficient. Okay. Hell yeah. Any
other points of risk to you? So

1512
01:27:03,114 --> 01:27:07,608
improving your new process I
suppose is something, but,

1513
01:27:07,608 --> 01:27:12,428
anything else that you view as
like large challenges in the

1514
01:27:12,428 --> 01:27:13,000
future?

1515
01:27:13,000 --> 01:27:17,345
Saman Farid: I would say another
big one is, like you mentioned

1516
01:27:17,345 --> 01:27:20,931
earlier, like scaling up the
service and maintenance

1517
01:27:20,931 --> 01:27:25,069
footprint. that clearly has a
lot of operational complexity,

1518
01:27:25,069 --> 01:27:29,276
associated with it, right? We
need to have technicians across

1519
01:27:29,276 --> 01:27:33,345
the country and outside the
country. At some point. We need

1520
01:27:33,345 --> 01:27:37,483
to have, spare parts available.
We need to have, density, in

1521
01:27:37,483 --> 01:27:41,621
different regions to support,
those maintenance technicians.

1522
01:27:41,621 --> 01:27:45,828
we need to have better remote,
error resolution tools. So all

1523
01:27:45,828 --> 01:27:50,103
of those things, each one, has a
lot of, operational questions

1524
01:27:50,103 --> 01:27:54,448
associated with it. luckily we
have a fantastic COO, who's much

1525
01:27:54,448 --> 01:27:58,310
better at kind of those
operational questions than I am.

1526
01:27:58,310 --> 01:28:01,000
and making great progress
towards them.

1527
01:28:01,000 --> 01:28:04,000
Audrow Nash: Hell yeah.

1528
01:28:04,000 --> 01:28:08,185
Audrow Nash: So I have the
feeling that, you have some

1529
01:28:08,185 --> 01:28:12,989
interesting opinions about, what
a bunch of robotics companies

1530
01:28:12,989 --> 01:28:17,716
are doing wrong in terms of like
how they go to market and or

1531
01:28:17,716 --> 01:28:22,133
what they're prioritizing. one
that we've already said is

1532
01:28:22,133 --> 01:28:26,937
focusing too much on getting the
robot to do the thing and not

1533
01:28:26,937 --> 01:28:31,664
enough on how to configure the
robot for different customers.

1534
01:28:31,664 --> 01:28:36,701
but what if you, could reach the
robotics community and tell them

1535
01:28:36,701 --> 01:28:41,351
to focus on something or do
something differently, what kind

1536
01:28:41,351 --> 01:28:46,000
of advice do you have in general
for the robotics community?

1537
01:28:46,000 --> 01:28:50,369
Saman Farid: I, think, I. One of
them you nailed, right? Which

1538
01:28:50,369 --> 01:28:54,949
is, I think we need to make sure
that whoever the, people who are

1539
01:28:54,949 --> 01:28:59,389
designing new robots to do new
tasks are very, focused on, like

1540
01:28:59,389 --> 01:29:03,546
the, practical, immediately
necessary tasks, right? I think

1541
01:29:03,546 --> 01:29:07,774
a lot of people get obsessed
with the problem. They're like,

1542
01:29:07,774 --> 01:29:11,932
what if we had a robot that
could, pick up a fabric and, do

1543
01:29:11,932 --> 01:29:16,089
these kind of very complicated
motions and there's actually

1544
01:29:16,089 --> 01:29:20,176
just not really as big of a
market as they might think for

1545
01:29:20,176 --> 01:29:23,911
that. and so this is the
challenge of engineers being

1546
01:29:23,911 --> 01:29:27,998
engineers is we love to solve
technical problems. and we'd

1547
01:29:27,998 --> 01:29:32,437
love to make, solve really hard
engineering problems. but a lot

1548
01:29:32,437 --> 01:29:36,736
of the time in robotics, like
that's not where the bottleneck

1549
01:29:36,736 --> 01:29:41,034
is, right? The bottleneck is in
all the boring stuff, that we

1550
01:29:41,034 --> 01:29:45,403
talked about. I think the other
thing I would say is for a lot

1551
01:29:45,403 --> 01:29:49,702
of, I. Robotics companies, they
don't think about this uptime

1552
01:29:49,702 --> 01:29:53,859
metric that you and I talked
about, right? we have a target

1553
01:29:53,859 --> 01:29:57,805
around 99.9% uptime. that, is
something that far too few

1554
01:29:57,805 --> 01:30:01,963
robotics companies think about
on a daily basis. So they'll

1555
01:30:01,963 --> 01:30:06,261
ship products with all kinds of
bugs and, and they'll realize

1556
01:30:06,261 --> 01:30:10,560
that that robot goes down all
the time in the field. and like

1557
01:30:10,560 --> 01:30:14,717
they'll, it's infuriating. but
the reality is, if you don't

1558
01:30:14,717 --> 01:30:18,311
solve all of the little
mechanical things, oh, this

1559
01:30:18,311 --> 01:30:21,834
screw rattles and then
eventually falls out, or my

1560
01:30:21,834 --> 01:30:25,921
suction cup breaks after 2000
cycles, or, the facility air

1561
01:30:25,921 --> 01:30:30,290
pressure, fluctuates more than I
expected. And my vacuum tool,

1562
01:30:30,290 --> 01:30:34,730
suddenly loses pressure and my
system goes down every once in a

1563
01:30:34,730 --> 01:30:39,239
while. these are all very common
occurrences and, If you want to

1564
01:30:39,239 --> 01:30:42,833
have a robust solution that's
actually useful for a

1565
01:30:42,833 --> 01:30:47,202
manufacturer or any kind of end
user, like you need to account

1566
01:30:47,202 --> 01:30:51,501
for all that stuff. You can't
come back and complain and say,

1567
01:30:51,501 --> 01:30:55,588
you provided me, varying air
pressure and so my system was

1568
01:30:55,588 --> 01:31:00,027
down. Like it just doesn't fly
in a manufacturing facility like

1569
01:31:00,027 --> 01:31:02,000
you need, to be bulletproof.

1570
01:31:02,000 --> 01:31:04,000
Audrow Nash: Hell yeah.

1571
01:31:04,000 --> 01:31:09,600
Audrow Nash: And then this is a
little off topic, but, or maybe

1572
01:31:09,600 --> 01:31:14,667
not exactly, not robots, but,
what is your read on the US

1573
01:31:14,667 --> 01:31:20,267
manufacturing system, I suppose
overall and how can people help

1574
01:31:20,267 --> 01:31:26,044
or get involved or I don't know,
like where are we headed and how

1575
01:31:26,044 --> 01:31:28,000
can we make it better?

1576
01:31:28,000 --> 01:31:31,818
Saman Farid: Yeah. I think that
American manufacturing like

1577
01:31:31,818 --> 01:31:35,960
world is facing like a crucible
moment, right? this is, I. right

1578
01:31:35,960 --> 01:31:39,713
now it's do or die, right? I
think, American manufacturing

1579
01:31:39,713 --> 01:31:43,661
has stagnated for the last 20
years and, it's gotten less and

1580
01:31:43,661 --> 01:31:47,220
less capable. and that is, a
function of, less and less

1581
01:31:47,220 --> 01:31:51,362
skilled, skilled, people coming
into the industry, less and less

1582
01:31:51,362 --> 01:31:54,921
labor available, to do
production, and the constant out

1583
01:31:54,921 --> 01:31:58,221
offshoring that's been
happening, right? People are

1584
01:31:58,221 --> 01:32:01,910
moving production to, other
regions, because that's where

1585
01:32:01,910 --> 01:32:05,922
they can, meet the performance
specs that they need to. the, I

1586
01:32:05,922 --> 01:32:09,870
think that, the manufacturing
industry needs number one. Like

1587
01:32:09,870 --> 01:32:13,882
skilled people to come and join
it, right? And in, in whatever

1588
01:32:13,882 --> 01:32:17,053
fashion, whether that's
engineers, whether that's

1589
01:32:17,053 --> 01:32:20,871
operations people, whether
that's, deployment capabilities.

1590
01:32:20,871 --> 01:32:24,948
there's a beauty and there's a
romance to making things, that I

1591
01:32:24,948 --> 01:32:28,960
think is just not a part of our
culture that much today. and I

1592
01:32:28,960 --> 01:32:32,973
think that needs to change. so
I'm very enthusiastic about any

1593
01:32:32,973 --> 01:32:36,597
efforts to improve that like
your podcast. number two, I

1594
01:32:36,597 --> 01:32:40,221
think, utilization needs to be
top of mind. I think that

1595
01:32:40,221 --> 01:32:43,521
there's this fallacy that we
hear a lot in American

1596
01:32:43,521 --> 01:32:47,275
manufacturing, which is, let's
let other people do the low

1597
01:32:47,275 --> 01:32:51,028
value stuff and we'll just do
the high value stuff, right?

1598
01:32:51,028 --> 01:32:55,170
Let's let other people make all
the car parts and we'll make the

1599
01:32:55,170 --> 01:32:58,988
aircraft parts, let's let other
people make all the plastic

1600
01:32:58,988 --> 01:33:02,676
parts and we'll make the
high-end medical devices, right?

1601
01:33:02,676 --> 01:33:06,365
Like we want, we only want the
high margin things, right?

1602
01:33:06,365 --> 01:33:09,730
There's this sense of
superiority. the problem is, a

1603
01:33:09,730 --> 01:33:13,807
manufacturing industry does not
exist if you can't make the low

1604
01:33:13,807 --> 01:33:17,755
end stuff too, right? Like the
people that are good at making

1605
01:33:17,755 --> 01:33:21,443
the low end stuff are very
likely gonna be the people who

1606
01:33:21,443 --> 01:33:25,326
are good at making the high end
stuff right. If you look at,

1607
01:33:25,326 --> 01:33:29,533
China's evolution, today, there,
you cannot make an iPhone in the

1608
01:33:29,533 --> 01:33:33,545
us. No matter how much money you
put towards that problem, you

1609
01:33:33,545 --> 01:33:37,428
cannot make. the most advanced
chips in the us no matter how

1610
01:33:37,428 --> 01:33:41,246
much money you put towards the
problem, it's not because we

1611
01:33:41,246 --> 01:33:44,935
don't have great engineers, but
if you don't have all the

1612
01:33:44,935 --> 01:33:49,012
ecosystem of the kind of the low
end manufacturing, the mid end

1613
01:33:49,012 --> 01:33:52,959
manufacturing and the high end
manufacturing, like you end up

1614
01:33:52,959 --> 01:33:56,454
missing all of the building
blocks to do that high-end

1615
01:33:56,454 --> 01:34:00,401
manufacturing. So I think it's
this like logical fallacy that

1616
01:34:00,401 --> 01:34:04,478
we've told ourselves in America,
which is we're focusing on the

1617
01:34:04,478 --> 01:34:08,620
high-end high value stuff. Like
it's bullshit. if you can't make

1618
01:34:08,620 --> 01:34:12,179
a cheap component, what makes
you think you can make an

1619
01:34:12,179 --> 01:34:15,933
expensive component? If you
can't make a simple component,

1620
01:34:15,933 --> 01:34:19,686
what makes you think you can
make a complex component? and

1621
01:34:19,686 --> 01:34:23,698
like the American manufacturing
industry just needs to be able

1622
01:34:23,698 --> 01:34:27,452
to support the full spectrum of
things. and I am, I'm very

1623
01:34:27,452 --> 01:34:31,658
enthusiastic on, things that are
helping smaller and medium sized

1624
01:34:31,658 --> 01:34:35,347
manufacturers. 'cause that's
really where the majority of

1625
01:34:35,347 --> 01:34:38,000
manufacturing happens in the US
today. I.

1626
01:34:38,000 --> 01:34:42,651
Audrow Nash: yeah. Now if you,
if our listeners and watchers

1627
01:34:42,651 --> 01:34:47,302
take away only one thing from
this episode, what do you want

1628
01:34:47,302 --> 01:34:48,000
it to be?

1629
01:34:48,000 --> 01:34:51,460
Saman Farid: I would say like
the number one thing that

1630
01:34:51,460 --> 01:34:54,732
American, like people who are
interested in American

1631
01:34:54,732 --> 01:34:58,507
manufacturing need to spend a
lot of time thinking about is,

1632
01:34:58,507 --> 01:35:02,344
how to increase the utilization
of our existing manufacturing

1633
01:35:02,344 --> 01:35:05,868
base. I think, we currently have
in America 300,000 plus

1634
01:35:05,868 --> 01:35:09,642
manufacturing facilities, that
are churning out all kinds of

1635
01:35:09,642 --> 01:35:13,669
products on a daily basis. And,
the average utilization of these

1636
01:35:13,669 --> 01:35:17,695
facilities is so low. so I would
say, like for any, whether it's

1637
01:35:17,695 --> 01:35:21,659
a kind of aspiring entrepreneur,
whether it's a business owner,

1638
01:35:21,659 --> 01:35:25,623
whether it's a manufacturer, to
the extent that we can get more

1639
01:35:25,623 --> 01:35:29,523
utilization from our existing,
base, there's so much potential

1640
01:35:29,523 --> 01:35:32,983
there. and I think that really
is the path to, like the

1641
01:35:32,983 --> 01:35:36,444
promised Land, right? Of having
a really strong, robust

1642
01:35:36,444 --> 01:35:40,344
manufacturing base that creates
a world of abundance. it comes

1643
01:35:40,344 --> 01:35:44,245
from we already have a lot of
these assets, let's get the most

1644
01:35:44,245 --> 01:35:45,000
out of them.

1645
01:35:45,000 --> 01:35:50,155
Audrow Nash: Hell yeah. Love it.
All right. I think we'll end

1646
01:35:50,155 --> 01:35:51,000
with that.

1647
01:35:51,000 --> 01:35:52,000
Saman Farid: Awesome.

1648
01:35:52,000 --> 01:35:53,000
Audrow Nash: having you.

1649
01:35:53,000 --> 01:35:55,000
Saman Farid: Thanks so much for
having me, Audrow.

1650
01:35:55,000 --> 01:35:57,000
Audrow Nash: All right. Bye
everyone. ​

1651
01:35:57,000 --> 01:36:00,937
Audrow Nash: What an insightful
conversation with Saman Farid

1652
01:36:00,937 --> 01:36:05,068
from Formic. It's fascinating to
see how they're revolutionizing

1653
01:36:05,068 --> 01:36:08,877
the accessibility of robotics
and American manufacturing. A

1654
01:36:08,877 --> 01:36:12,814
few things that really stood out
to me. First, their focus on

1655
01:36:12,814 --> 01:36:15,912
increasing utilization of
existing manufacturing

1656
01:36:15,912 --> 01:36:19,462
facilities. It's eyeopening to
learn that many American

1657
01:36:19,462 --> 01:36:23,206
factories are running at far
below their capacity. Second,

1658
01:36:23,206 --> 01:36:27,079
their commitment to solving real
world problems by deploying

1659
01:36:27,079 --> 01:36:31,210
robots only in actual production
environments and tracking usage

1660
01:36:31,210 --> 01:36:34,825
hours they're ensuring their
technology makes a tangible

1661
01:36:34,825 --> 01:36:38,439
impact. And finally, their
emphasis on the importance of

1662
01:36:38,439 --> 01:36:42,570
maintaining and servicing robots
after deployment. It's not just

1663
01:36:42,570 --> 01:36:46,766
about selling robots, it's about
keeping them running efficiently

1664
01:36:46,766 --> 01:36:50,574
long term. As we wrap up, I
wanna leave you with a question

1665
01:36:50,574 --> 01:36:53,931
to ponder. How might increased
automation and better

1666
01:36:53,931 --> 01:36:57,481
utilization of existing
facilities change the landscape

1667
01:36:57,481 --> 01:37:01,483
of manufacturing in your local
community? Thanks for listening

1668
01:37:01,483 --> 01:37:04,000
and I'll catch you in the next
episode.
