1
00:00:00,000 --> 00:00:03,816
Audrow Nash: Hi everyone, Audrow
here. Housing is expensive. A

2
00:00:03,816 --> 00:00:07,386
big and growing part of that
expense is the cost of labor.

3
00:00:07,386 --> 00:00:11,141
People are expensive, and the
laborforce is shrinking as more

4
00:00:11,141 --> 00:00:15,019
people retire than join. A big
way to drive the cost down is to

5
00:00:15,019 --> 00:00:18,650
automate parts of the process
that are labor intensive. But

6
00:00:18,650 --> 00:00:22,405
robotics in its current state is
only good at certain things.

7
00:00:22,405 --> 00:00:25,975
Therefore there's an art in
picking things to automate. If

8
00:00:25,975 --> 00:00:29,668
the task is too obvious, you
have tons of competitors and it

9
00:00:29,668 --> 00:00:33,669
becomes a race to the bottom. If
you pick a task that's too hard,

10
00:00:33,669 --> 00:00:37,116
you run out of money trying to
solve the problem and the

11
00:00:37,116 --> 00:00:40,624
company fails. Today I'm talking
to Barrett Ames, who's a

12
00:00:40,624 --> 00:00:44,317
cofounder of Botbuilt. I think
Botbuilt has found that sweet

13
00:00:44,317 --> 00:00:48,134
spot. They solve a few really
thorny problems and it lets them

14
00:00:48,134 --> 00:00:52,011
automate build wood frames for
houses and other buildings. This

15
00:00:52,011 --> 00:00:56,012
conversation will be interesting
to - any aspiring entreprenur as

16
00:00:56,012 --> 00:00:59,767
a great example of how different
challenges can be overcome -

17
00:00:59,767 --> 00:01:03,583
people curious about robotics
construction - and any investors

18
00:01:03,583 --> 00:01:07,153
looking to learn more about a
strong robotics company with

19
00:01:07,153 --> 00:01:10,908
what seems to me to be a good
moat Before we dive in, a small

20
00:01:10,908 --> 00:01:14,231
secret since I'm making an
announcement in a couple of

21
00:01:14,231 --> 00:01:17,986
days, but I'm excited to that
I've create Audrow Nash Podcast

22
00:01:17,986 --> 00:01:21,556
Jobs board. The board grew out
of a problem I kept seeing:

23
00:01:21,556 --> 00:01:25,557
robotics companies struggling to
find talent, and talented people

24
00:01:25,557 --> 00:01:29,189
struggling to find the right
robotics opportunities. I hope

25
00:01:29,189 --> 00:01:33,066
this jobs board can help here.
Several great robotics companies

26
00:01:33,066 --> 00:01:36,821
are already posting positions,
including this episode's guest

27
00:01:36,821 --> 00:01:40,514
BotBuilt, along with Formic,
Luxonis, and Rerun. BotBuilt is

28
00:01:40,514 --> 00:01:43,961
still finalizing their hiring
plans and should have more

29
00:01:43,961 --> 00:01:47,839
engineering roles posted in the
coming weeks. If you're looking

30
00:01:47,839 --> 00:01:51,409
for a job or maybe thinking
about making a change, you can

31
00:01:51,409 --> 00:01:55,102
sign up for free email updates
to be the first to know about

32
00:01:55,102 --> 00:01:58,733
new positions. If you're at
robotics company that's looking

33
00:01:58,733 --> 00:02:02,119
to hire new people, consider
posting on the Audrow Nash

34
00:02:02,119 --> 00:02:05,319
Podcast Jobs board to gain
visibility with our great

35
00:02:05,319 --> 00:02:08,582
community of robotics
professionals. Job postings are

36
00:02:08,582 --> 00:02:12,275
free. You can learn more at
jobs.audrownashpodcast.com. I'll

37
00:02:12,275 --> 00:02:15,783
have a link in the show notes.
Alright, grab a snack, get

38
00:02:15,783 --> 00:02:19,353
comfortable, and here's my
conversation with Barrett Ames.

39
00:02:19,353 --> 00:02:22,000
Hey, Barrett. Would you
introduce yourself?

40
00:02:22,000 --> 00:02:25,869
Barrett Ames: Yeah. My name is
Barrett Ames. I'm one of the

41
00:02:25,869 --> 00:02:29,541
co-founders of BotBuilt. At
BotBuilt we build homes with

42
00:02:29,541 --> 00:02:30,000
robots.

43
00:02:30,000 --> 00:02:32,182
Audrow Nash: Tell me more about
that. What are you doing to help

44
00:02:32,182 --> 00:02:33,000
build homes with robots?

45
00:02:33,000 --> 00:02:36,977
Barrett Ames: Yeah. So we do a
bunch of construction in a

46
00:02:36,977 --> 00:02:41,095
warehouse. Right? So, somebody
sends us a floor plan that's

47
00:02:41,095 --> 00:02:45,491
just like a PDF drawing of the
kind of house that they want. In

48
00:02:45,491 --> 00:02:50,026
fact, it doesn't even have to be
a PDF anymore. We just literally

49
00:02:50,026 --> 00:02:54,074
did a napkin drawing, like
somebody drew out on a piece of

50
00:02:54,074 --> 00:02:58,540
paper. Like, this is how I want
my house to be. And we were able

51
00:02:58,540 --> 00:03:03,005
to design that from the piece of
paper. But anyways, however you

52
00:03:03,005 --> 00:03:07,332
have it, you give us your floor,
floor plan design, your house

53
00:03:07,332 --> 00:03:11,449
design. We ingest that, design
out the 3D aspects of it. So

54
00:03:11,449 --> 00:03:15,496
where does every piece of
lumber, where does every nail go

55
00:03:15,496 --> 00:03:19,683
in order to build this house?
That gets transitioned over to

56
00:03:19,683 --> 00:03:24,009
our robots. And we figure out
first kind of a high level plan,

57
00:03:24,009 --> 00:03:27,847
how do we go and do each of
these pieces? So where does

58
00:03:27,847 --> 00:03:32,174
where, you know, do we put this
piece of lumber in before that

59
00:03:32,174 --> 00:03:36,360
one? Do we put these nails in
before those nails. We do that

60
00:03:36,360 --> 00:03:40,408
and then we execute it on the
robots. So that's, you know,

61
00:03:40,408 --> 00:03:44,455
dealing with all of the gnarly
ness that is lumber, right?

62
00:03:44,455 --> 00:03:48,781
It'll be Bo. It'll be twisted
all sorts of, bad lumber. And we

63
00:03:48,781 --> 00:03:53,108
do all of that in a warehouse,
in big chunks. Build your house

64
00:03:53,108 --> 00:03:57,155
over a period of time in the
warehouse, stack it up on the

65
00:03:57,155 --> 00:04:01,342
back of a truck, ship it out to
the jobsite. And the big win

66
00:04:01,342 --> 00:04:05,459
there is. It lets the framers
go. The rough carpenters that

67
00:04:05,459 --> 00:04:09,925
are on site lets them go from a
14 to 21 day build cycle down to

68
00:04:09,925 --> 00:04:14,391
a four hours. Wow. So there's a
big labor savings, in particular

69
00:04:14,391 --> 00:04:18,647
on the jobsite, right? Which is
just like, not job sites are,

70
00:04:18,647 --> 00:04:23,043
like, not a great place to be,
right? Like, you're out there in

71
00:04:23,043 --> 00:04:27,370
the cold, you're out there in
the sun. You're out there in the

72
00:04:27,370 --> 00:04:31,836
heat. We just got our site video
back from our last, build, and,

73
00:04:31,836 --> 00:04:36,023
one of our panels didn't make it
because they tore it up and

74
00:04:36,023 --> 00:04:40,000
burned it to stay warm, right?
Oh, my gosh. Like, that's,

75
00:04:40,000 --> 00:04:43,436
Audrow Nash: That's hilarious.
Oh, man. Yeah. So you don't want

76
00:04:43,436 --> 00:04:46,545
to be out there. And this is
specifically for the framing

77
00:04:46,545 --> 00:04:49,000
part, which is the walls that go
up. Correct?

78
00:04:49,000 --> 00:04:52,029
Barrett Ames: Correct, correct.
Yeah. So you can think about it

79
00:04:52,029 --> 00:04:54,000
as just like the, the bones of
the house.

80
00:04:54,000 --> 00:04:57,503
Audrow Nash: Okay. And so how is
this done now and then. I'd love

81
00:04:57,503 --> 00:05:00,359
to hear about how you guys are
doing it. And then the

82
00:05:00,359 --> 00:05:03,000
challenges you come into by
doing it with robots.

83
00:05:03,000 --> 00:05:07,160
Barrett Ames: Yeah. So right now
it's all done by hand, right.

84
00:05:07,160 --> 00:05:11,454
It's just, if you go out to the
job site, there will be like one

85
00:05:11,454 --> 00:05:15,480
really tattered piece of paper
on every job site. And that's

86
00:05:15,480 --> 00:05:19,774
the floor plan. Every framer is
going and looking at it, drawing

87
00:05:19,774 --> 00:05:23,598
on it, spilling their, their
beer or their monster on it.

88
00:05:23,598 --> 00:05:27,758
Right. Like, and they're just
they're just picking up pieces a

89
00:05:27,758 --> 00:05:31,449
two by four, cutting them down
to size and nailing them

90
00:05:31,449 --> 00:05:35,541
together. And it's, it's, it's
surprisingly dangerous, right?

91
00:05:35,541 --> 00:05:39,097
Like, you don't, you know, you
kind of think about it

92
00:05:39,097 --> 00:05:43,392
abstractly and you're like, oh,
yeah, you're just putting two by

93
00:05:43,392 --> 00:05:47,619
fours together. Yeah, whatever.
But in order to build the house

94
00:05:47,619 --> 00:05:51,644
up. Right, like, you really have
to start doing some pretty,

95
00:05:51,644 --> 00:05:55,737
pretty, interesting gymnastics
pretty early. Right? So, like,

96
00:05:55,737 --> 00:05:59,830
on that site video that I was,
just referencing, we could see

97
00:05:59,830 --> 00:06:03,990
guys on other jobs standing on
top of the first floor, walking

98
00:06:03,990 --> 00:06:08,217
around on two by fours. Right.
That's all. No safety equipment.

99
00:06:08,217 --> 00:06:12,377
So it's just like, there's saws
everywhere, nail guns, and and

100
00:06:12,377 --> 00:06:16,000
it's just, it's just generally
not a safe place to be.

101
00:06:16,000 --> 00:06:18,892
Audrow Nash: There. Yeah, but
the accident rates are pretty

102
00:06:18,892 --> 00:06:21,931
high. For this and especially
once you get into multi storeys.

103
00:06:21,931 --> 00:06:24,873
Right. Because I mean first
floor it's like well you have to

104
00:06:24,873 --> 00:06:27,765
stand up these things and you
have to cut and you have nail

105
00:06:27,765 --> 00:06:30,657
guns and things like this I'm
sure accidents happen just on

106
00:06:30,657 --> 00:06:33,696
the first floor. But yeah I'm
sure with more than one floor it

107
00:06:33,696 --> 00:06:36,000
becomes very dicey very quickly.
Yeah probably.

108
00:06:36,000 --> 00:06:40,398
Barrett Ames: Yep yep. And so
it's yeah. The way that they do

109
00:06:40,398 --> 00:06:44,292
it right now is, pretty
rudimentary, right. It's just,

110
00:06:44,292 --> 00:06:48,329
like, you get the plan, you
build it. And then somebody,

111
00:06:48,329 --> 00:06:52,439
somebody else, comes by and
checks it, and that's, that's

112
00:06:52,439 --> 00:06:56,765
kind of the end of the story.
And you gotta you gotta do it.

113
00:06:56,765 --> 00:06:59,000
And whatever, whether there is.

114
00:06:59,000 --> 00:07:02,256
Audrow Nash: Yeah, that's sinks
a cot or cold or anything.

115
00:07:02,256 --> 00:07:05,513
Right. But so I think there are
some challenges because we

116
00:07:05,513 --> 00:07:08,938
talked before. And so I remember
some of the things there are

117
00:07:08,938 --> 00:07:12,419
some challenges that sound very
interesting with this one. The

118
00:07:12,419 --> 00:07:15,451
floor plan is very low
resolution, like it's it's just

119
00:07:15,451 --> 00:07:18,707
basically this is where the
walls are going to go. They're

120
00:07:18,707 --> 00:07:22,357
not saying the like the distance
between the studs, how much wood

121
00:07:22,357 --> 00:07:25,613
they need, all sorts of thing.
So I think there's like one

122
00:07:25,613 --> 00:07:28,926
there's a lot of ambiguity in
this. And especially for like

123
00:07:28,926 --> 00:07:32,070
custom homes or things like
this, a lot of variation and

124
00:07:32,070 --> 00:07:35,382
then probably a lot of waste
because of this. Tell me a bit

125
00:07:35,382 --> 00:07:36,000
about this.

126
00:07:36,000 --> 00:07:40,042
Barrett Ames: Yeah. So every
floor plan, when you look at it

127
00:07:40,042 --> 00:07:44,084
first you, you think like, oh,
this is a technical document,

128
00:07:44,084 --> 00:07:48,125
right? Like it's got lots of
lines and there's like a scale,

129
00:07:48,125 --> 00:07:52,302
right? Like there's all these
dimensions drawn everywhere. And

130
00:07:52,302 --> 00:07:56,546
what we've found is, it's not a
technical document at all. It's

131
00:07:56,546 --> 00:08:00,722
actually just providing designer
intent. Right? Like we want a

132
00:08:00,722 --> 00:08:05,034
window, and the window is here.
The door is there. Right? And we

133
00:08:05,034 --> 00:08:08,806
understand it to be intent
because, like, if you shift a

134
00:08:08,806 --> 00:08:12,915
door an eighth of an inch left
or right, no one cares. No one

135
00:08:12,915 --> 00:08:16,957
cares, right? They do care if
all of the doors, are lined up

136
00:08:16,957 --> 00:08:20,864
in a hallway or something,
right? Like, so there there are

137
00:08:20,864 --> 00:08:25,108
kind of these design guidelines,
and so the first thing that we

138
00:08:25,108 --> 00:08:29,217
have to do is, is really bring
all of that intent in and then

139
00:08:29,217 --> 00:08:33,528
go and design every little bit.
And because we do that, we get a

140
00:08:33,528 --> 00:08:37,233
really accurate accounting.
Right? Like we have to know

141
00:08:37,233 --> 00:08:41,275
where every nail goes in order
for the robots to come in, do

142
00:08:41,275 --> 00:08:45,182
that. And when they're building
a house right now, they're

143
00:08:45,182 --> 00:08:48,955
literally just looking at a
piece of paper and going, oh

144
00:08:48,955 --> 00:08:52,997
yeah, I need to do that over
there. And they walk over there

145
00:08:52,997 --> 00:08:56,836
and do it right. There's not a
lot of like, yeah, there's

146
00:08:56,836 --> 00:09:00,878
there's not a lot of planning
necessary. And so they're they

147
00:09:00,878 --> 00:09:04,920
don't do a lot of planning. And
so when we go through and do

148
00:09:04,920 --> 00:09:08,894
every single stud in every
single nail ahead of time, we've

149
00:09:08,894 --> 00:09:13,138
get this, this really accurate
picture of the house well before

150
00:09:13,138 --> 00:09:17,247
it gets built. And because of
that we can order material more

151
00:09:17,247 --> 00:09:21,491
accurately, drastically reducing
waste. Right. So on the houses

152
00:09:21,491 --> 00:09:25,870
that we built, we've had our our
contract partners say we usually

153
00:09:25,870 --> 00:09:29,979
have 3 to 5 dumpster dumpsters
of waste, but with you we only

154
00:09:29,979 --> 00:09:32,000
have half a dumpster of waste.

155
00:09:32,000 --> 00:09:33,000
Audrow Nash: Right. So that's a
big reduction.

156
00:09:33,000 --> 00:09:36,634
Barrett Ames: Big Reduction in
waste. And then on top of that,

157
00:09:36,634 --> 00:09:39,859
the way that they actually
currently order materials is

158
00:09:39,859 --> 00:09:43,669
really interesting. They'll take
the floor plan and they'll go to

159
00:09:43,669 --> 00:09:47,303
a lumber yard and they'll say
like, hey, lumber yard. How much

160
00:09:47,303 --> 00:09:50,000
material do you want or do we
want? And, that.

161
00:09:50,000 --> 00:09:51,000
Audrow Nash: Sounds cooked.

162
00:09:51,000 --> 00:09:55,166
Barrett Ames: Right. Yeah. And
so we've we've found 20 of all

163
00:09:55,166 --> 00:09:59,400
of it. Yeah, yeah, yeah. So we
find, we find, 20 to 40% saving

164
00:09:59,400 --> 00:10:03,839
in material, just by, by kind of
solving this asymmetry in it in,

165
00:10:03,839 --> 00:10:05,000
in in incentives.

166
00:10:05,000 --> 00:10:08,066
Audrow Nash: Okay. That's very
interesting. What about so

167
00:10:08,066 --> 00:10:11,187
another problem that I see
that's kind of difficult is the

168
00:10:11,187 --> 00:10:14,468
variation in wood quality. So
like you buy a two by four it's

169
00:10:14,468 --> 00:10:17,696
slightly twisted or a little
longer or the angles are weird.

170
00:10:17,696 --> 00:10:21,139
Or tell me a bit about that too,
because that that probably is a

171
00:10:21,139 --> 00:10:22,000
challenge. Also.

172
00:10:22,000 --> 00:10:26,524
Barrett Ames: Yeah, it's a
massive challenge. And it's it's

173
00:10:26,524 --> 00:10:31,048
a key challenge, right. There
are a ton of grades of lumber

174
00:10:31,048 --> 00:10:35,572
out there. You can go all the
way up to LSL, which is this,

175
00:10:35,572 --> 00:10:39,942
this beautiful engineered
lumber, and it will be straight

176
00:10:39,942 --> 00:10:44,466
and perfect, but it cost ten
times as much as what a normal

177
00:10:44,466 --> 00:10:46,000
house is built with.

178
00:10:46,000 --> 00:10:47,688
Audrow Nash: You're paying for
that precision, right? Right. Or

179
00:10:47,688 --> 00:10:49,000
this kind of thing, because it's
two spec, right?

180
00:10:49,000 --> 00:10:53,191
Barrett Ames: Okay. Right. And
so we're early on at BotBuilt we

181
00:10:53,191 --> 00:10:57,249
had this decision put before us,
like, do we decide to engage

182
00:10:57,249 --> 00:11:01,173
with the lumber that is on the
job site, or do we pick this

183
00:11:01,173 --> 00:11:05,165
much better engineered lumber?
Right. And it would have made

184
00:11:05,165 --> 00:11:09,289
the engineering problem much,
much easier. But we decided that

185
00:11:09,289 --> 00:11:12,948
the the fundamental thing is
that we need to build more

186
00:11:12,948 --> 00:11:16,939
houses in America, and that
means cheaper, you know, cheaper

187
00:11:16,939 --> 00:11:21,130
materials. So we went with what
homes are built with today. And

188
00:11:21,130 --> 00:11:24,988
if you look at what homes are
built with today, it's just,

189
00:11:24,988 --> 00:11:28,846
it's just the two by fours that
you can see at Home Depot,

190
00:11:28,846 --> 00:11:33,037
right? Like it's just going to
be twisted and borrowed. There's

191
00:11:33,037 --> 00:11:36,895
this thing called waning, which
is where big chunks of the

192
00:11:36,895 --> 00:11:40,487
corners are missing. Right.
There's going to be knots.

193
00:11:40,487 --> 00:11:44,013
There's a big pine beetle
problem right now. So there

194
00:11:44,013 --> 00:11:47,938
there will be holes in some of
the lumber. And that's okay.

195
00:11:47,938 --> 00:11:50,000
Right. Like there there's some.

196
00:11:50,000 --> 00:11:52,047
Audrow Nash: Amount that doesn't
compromise the structural

197
00:11:52,047 --> 00:11:53,000
integrity that much. Right?

198
00:11:53,000 --> 00:11:56,823
Barrett Ames: Okay. Right,
right. And so there's there's

199
00:11:56,823 --> 00:12:00,918
just a ton of variation within
one species of lumber. Right.

200
00:12:00,918 --> 00:12:05,150
Like we in the southeast build
with, what's called spruce pine

201
00:12:05,150 --> 00:12:09,109
fir. But if you go to the
Pacific Northwest, they're going

202
00:12:09,109 --> 00:12:13,410
to be more Douglas fir. And so
you'll get variations in, in all

203
00:12:13,410 --> 00:12:17,505
sorts of the texture, like the
visual texture, but then also

204
00:12:17,505 --> 00:12:21,191
the, the geometry of the two by
four, that are species

205
00:12:21,191 --> 00:12:25,355
dependent, location dependent,
harvest dependent. Time. Like,

206
00:12:25,355 --> 00:12:29,451
how old was this tree when they
chopped it down? All of that

207
00:12:29,451 --> 00:12:33,000
leads to just an immense amount
of variation, I bet.

208
00:12:33,000 --> 00:12:36,596
Audrow Nash: Okay, so how do
your, So, like, I feel like I

209
00:12:36,596 --> 00:12:40,502
have a good grasp of the problem
now. How do your systems work?

210
00:12:40,502 --> 00:12:44,346
And what have been kind of the
hard problems on the way? I can

211
00:12:44,346 --> 00:12:48,190
imagine a few and we've talked
earlier, but tell me about kind

212
00:12:48,190 --> 00:12:51,786
of like end to end. What are you
guys doing? I know you're

213
00:12:51,786 --> 00:12:55,196
building the frames, but what
does it look like in your

214
00:12:55,196 --> 00:12:58,606
operation? And then I'd love to
just quickly, as you're

215
00:12:58,606 --> 00:13:02,264
describing it, talk about the,
hard problems and then we'll

216
00:13:02,264 --> 00:13:04,000
dive more into each of them.

217
00:13:04,000 --> 00:13:07,812
Barrett Ames: Yeah. From a
operational standpoint. Right.

218
00:13:07,812 --> 00:13:11,758
Like working with lumber, starts
at, like a really material

219
00:13:11,758 --> 00:13:15,837
handling specific level. Right?
A lot of a lot of approaches.

220
00:13:15,837 --> 00:13:19,515
Don't think about just the
logistics of moving a two by

221
00:13:19,515 --> 00:13:23,394
four around. And so what we do
is we just stack all of our

222
00:13:23,394 --> 00:13:27,607
lumber right next to the robots.
And then the robots look over,

223
00:13:27,607 --> 00:13:31,820
grab the piece of lumber out of
the out of. It's called a bunk,

224
00:13:31,820 --> 00:13:36,167
which is just a big old thing, a
lumber freight. Just think about

225
00:13:36,167 --> 00:13:40,180
all the lumber that you see
stacked up in Home Depot. That's

226
00:13:40,180 --> 00:13:43,858
a bunk. We grab a piece out of
that, we put it in in, a

227
00:13:43,858 --> 00:13:48,004
measurement device at the back
of our system that allows us to

228
00:13:48,004 --> 00:13:52,084
position it precisely to a 32nd
of an inch. Then we pick up a

229
00:13:52,084 --> 00:13:56,096
saw, cut the lumber down to
size, and then pick it up out of

230
00:13:56,096 --> 00:13:59,707
our our. We call it the input
table, but basically our

231
00:13:59,707 --> 00:14:03,921
measurement device and put it on
the, put it on the work table.

232
00:14:03,921 --> 00:14:08,000
You do that a couple thousand
times and you've built a house.

233
00:14:08,000 --> 00:14:12,193
Audrow Nash: Right? Yeah. Okay.
So you process the lumber, you

234
00:14:12,193 --> 00:14:16,454
put it in a big bin, the bunk,
you grab it out, you cut it, you

235
00:14:16,454 --> 00:14:20,444
place it where you need it. How
do you. I guess so the main

236
00:14:20,444 --> 00:14:24,435
challenges that I see with this
one, getting an accurate 3D

237
00:14:24,435 --> 00:14:28,696
model of all the wood and then
doing something appropriate with

238
00:14:28,696 --> 00:14:32,889
it to cut it, and things like
this. Like, tell me a little bit

239
00:14:32,889 --> 00:14:36,000
about that. I suspect that's a
good challenge.

240
00:14:36,000 --> 00:14:40,344
Barrett Ames: Yeah. Yeah. So I
actually like even before we

241
00:14:40,344 --> 00:14:44,762
characterize the lumber, just
the, so the, the bunk actually

242
00:14:44,762 --> 00:14:49,032
comes to us just strapped.
Right. So there's okay, there's

243
00:14:49,032 --> 00:14:53,597
like, this nice 14 by 14, grid
of lumber. And when we, cut the

244
00:14:53,597 --> 00:14:58,088
straps, it's it's a bunch of
lumber all recently cut from the

245
00:14:58,088 --> 00:15:02,800
same tree, all laid out next to
each other. So I actually like t

246
00:15:02,800 --> 00:15:06,923
first problem is individual
weighting or simulating that

247
00:15:06,923 --> 00:15:11,193
piece of lumber. Yeah. Visually.
Right. And so, so there's

248
00:15:11,193 --> 00:15:15,685
actually, a really interesting
challenge there, that we solve

249
00:15:15,685 --> 00:15:20,249
with a deep convolutional neural
network, and a whole bunch of

250
00:15:20,249 --> 00:15:24,888
data that only we have. Right?
Like, there's not a whole lot of

251
00:15:24,888 --> 00:15:28,937
people out there that are
collecting massive amounts of

252
00:15:28,937 --> 00:15:33,281
two by four data, but we we're
one of them. Yeah. And so we

253
00:15:33,281 --> 00:15:37,478
that's fine. Yeah. We simulate
each of them. Just using a

254
00:15:37,478 --> 00:15:41,969
stereo pair, and then, pick it
up, we pick it up using vacuum

255
00:15:41,969 --> 00:15:43,000
grippers. Wow.

256
00:15:43,000 --> 00:15:45,051
Audrow Nash: Yeah. I'm
impressed. It's vacuum grippers

257
00:15:45,051 --> 00:15:47,405
for that. That must be a big arm
and big vacuum grippers. Then

258
00:15:47,405 --> 00:15:49,000
how heavy is each of these
pieces of wood?

259
00:15:49,000 --> 00:15:51,511
Barrett Ames: Two by four. It's
actually only 10 pounds, so

260
00:15:51,511 --> 00:15:53,000
they're actually pretty light.
And.

261
00:15:53,000 --> 00:15:54,000
Audrow Nash: Okay. Just awkward
and long.

262
00:15:54,000 --> 00:15:56,789
Barrett Ames: Yeah, yeah, it's
like a motion planning

263
00:15:56,789 --> 00:15:59,895
nightmare, right? Like, just
imagine you don't whack stuff,

264
00:15:59,895 --> 00:16:02,947
right? Yeah. Just imagine
everything's a bug trap problem,

265
00:16:02,947 --> 00:16:06,158
because that's what everything
turns into when you have this.

266
00:16:06,158 --> 00:16:07,000
What long piece.

267
00:16:07,000 --> 00:16:08,000
Audrow Nash: What is a bug trap?

268
00:16:08,000 --> 00:16:11,675
Barrett Ames: Oh, yeah. Yeah.
Just like, you know, the, the,

269
00:16:11,675 --> 00:16:15,534
it comes from the, the bug traps
that you'd use to catch, like,

270
00:16:15,534 --> 00:16:18,842
fruit flies. Right? Like you
take a cup, you put some,

271
00:16:18,842 --> 00:16:22,518
vinegar and some sugar in the
bottom of it, and then you put

272
00:16:22,518 --> 00:16:26,315
some Saran wrap over it, and you
poke hole in it. Right? Okay.

273
00:16:26,315 --> 00:16:30,297
And the fruit fly flies into it,
but it's the they're too dumb to

274
00:16:30,297 --> 00:16:34,156
get out. Like there's just one
very narrow passage that can get

275
00:16:34,156 --> 00:16:37,709
through. Yeah. That's that's
what in motion planning. Like

276
00:16:37,709 --> 00:16:41,446
that's, that's how we describe
kind of these narrow passages.

277
00:16:41,446 --> 00:16:45,243
Right. Where relatively easy to
get in one direction, but like

278
00:16:45,243 --> 00:16:48,000
very specific to get out the
other direction.

279
00:16:48,000 --> 00:16:51,026
Audrow Nash: To pull it out.
Right. Of the bunk. Right. For

280
00:16:51,026 --> 00:16:52,000
this kind of thing.

281
00:16:52,000 --> 00:16:55,260
Barrett Ames: Right. Okay. And
so yeah. So just moving with

282
00:16:55,260 --> 00:16:58,575
lumber in general kind of turns
everything into these narrow

283
00:16:58,575 --> 00:17:01,613
passage problems. Yeah. For
sure. Right. Because you're

284
00:17:01,613 --> 00:17:02,000
you're.

285
00:17:02,000 --> 00:17:05,450
Audrow Nash: Because you have
to, like, put it this way and

286
00:17:05,450 --> 00:17:09,016
bring it that way. Right? Like
backing up a car, kind of like

287
00:17:09,016 --> 00:17:12,759
turn the wood. Right. This kind
of thing. Right? Okay. So, yeah,

288
00:17:12,759 --> 00:17:16,442
you have to move all that. That
sounds painful or sounds like a

289
00:17:16,442 --> 00:17:19,892
challenging motion planning
problem. And segmenting it from

290
00:17:19,892 --> 00:17:23,400
all the other wood that looks
similar with, 3D sensor. Yeah,

291
00:17:23,400 --> 00:17:26,792
that sounds hard. But so you
have a data set for that that

292
00:17:26,792 --> 00:17:30,417
you've collected through your
own. Means. And that lets you do

293
00:17:30,417 --> 00:17:34,217
it. Okay. So that's kind of like
problem one is the segmenting it

294
00:17:34,217 --> 00:17:37,901
gets at. Problem two is moving
it. Then you move it to your saw

295
00:17:37,901 --> 00:17:41,000
and your position for measuring
it right. Everything.

296
00:17:41,000 --> 00:17:45,833
Barrett Ames: Right. And that's
that's a really tricky problem

297
00:17:45,833 --> 00:17:50,354
as well because, gripping on to
lumber. Lumber is actually

298
00:17:50,354 --> 00:17:55,266
pretty soft. Right. And so you
have to worry about deforming it

299
00:17:55,266 --> 00:18:00,099
while you're moving it. And how
that impacts your measurement.

300
00:18:00,099 --> 00:18:04,308
And so there are, there's a
bunch of how do we move it

301
00:18:04,308 --> 00:18:08,907
really quickly because we need
to move it really quickly in

302
00:18:08,907 --> 00:18:13,585
order to, to process it in, in
an economical way, and get it

303
00:18:13,585 --> 00:18:17,872
down to a 32nd of an inch.
Right. So we have to measure

304
00:18:17,872 --> 00:18:22,627
very precisely, to, to really,
get the, get the tolerances we

305
00:18:22,627 --> 00:18:27,382
need. Right. Those tolerances
are basically driven from like,

306
00:18:27,382 --> 00:18:32,294
how the lumber stacks up in, in
a building. And so there's like

307
00:18:32,294 --> 00:18:36,893
a really interesting controls
problem there, that we tackle

308
00:18:36,893 --> 00:18:41,726
with, with some in-house, kind
of combination of, of controls,

309
00:18:41,726 --> 00:18:45,000
sensor fusion and some custom,
mechanisms.

310
00:18:45,000 --> 00:18:49,018
Audrow Nash: Gotcha. So once
you're grabbing the wood and

311
00:18:49,018 --> 00:18:53,458
then you're placing it, cutting,
measuring it, cutting it, this

312
00:18:53,458 --> 00:18:57,758
kind of thing. Now there's still
the difficulty in my head of

313
00:18:57,758 --> 00:19:01,000
going from the napkin sketch to
the full plan.

314
00:19:01,000 --> 00:19:02,000
Barrett Ames: Yeah. Yeah.

315
00:19:02,000 --> 00:19:04,786
Audrow Nash: That has like
windows and doors and all the

316
00:19:04,786 --> 00:19:07,920
features that one might expect
and matches code and things like

317
00:19:07,920 --> 00:19:10,856
this. Tell me a bit about that
because that sounds hard and

318
00:19:10,856 --> 00:19:12,000
interesting also. Yeah.

319
00:19:12,000 --> 00:19:16,816
Barrett Ames: Yeah. So that very
difficult problem. And just,

320
00:19:16,816 --> 00:19:21,789
just absolutely key to getting
everything working, in the right

321
00:19:21,789 --> 00:19:26,605
direction. Right. And so the,
what, what we do is, is kind of

322
00:19:26,605 --> 00:19:30,474
this combination of, mixed
integer programing and

323
00:19:30,474 --> 00:19:34,974
generative AI. Right? So, we
need to take in all of these

324
00:19:34,974 --> 00:19:39,711
constraints and generate, the
true constraint. Right. And so

325
00:19:39,711 --> 00:19:42,000
we we use a generative model.

326
00:19:42,000 --> 00:19:43,000
Audrow Nash: What is your true
constraint. What do you mean.

327
00:19:43,000 --> 00:19:47,376
Barrett Ames: Well, like we have
to interpret all of these as an

328
00:19:47,376 --> 00:19:51,000
intent as opposed to like,
mathematically true dicta.

329
00:19:51,000 --> 00:19:52,000
Audrow Nash: Right.

330
00:19:51,000 --> 00:19:53,980
Barrett Ames: Yeah. And so
that's kind of that's that's the

331
00:19:53,980 --> 00:19:56,000
translation that I'm talking
about here.

332
00:19:56,000 --> 00:19:59,197
Audrow Nash: I see. Yeah. So
there will be a true constraint.

333
00:19:59,197 --> 00:20:01,555
And that is something that
matches all of the

334
00:20:01,555 --> 00:20:04,803
specifications dictated in the.
Okay. So you have to interpret

335
00:20:04,803 --> 00:20:08,000
that into something that's like
right, complete right. Right.

336
00:20:08,000 --> 00:20:12,332
Barrett Ames: And so okay, in
order to build that model, we

337
00:20:12,332 --> 00:20:16,810
actually have just just over
2,000,000ft² of floor plans that

338
00:20:16,810 --> 00:20:21,142
we've now processed. And that
lets us, train, I think we've

339
00:20:21,142 --> 00:20:25,547
got three, no, we've got four
generative models now that are

340
00:20:25,547 --> 00:20:29,805
built off of that, just picking
out all of these different

341
00:20:29,805 --> 00:20:33,696
constraints and synthesizing
them together to give us

342
00:20:33,696 --> 00:20:37,000
something, that can go into the
optimization.

343
00:20:37,000 --> 00:20:39,514
Audrow Nash: Right. Okay. Why?
For models, what are they doing

344
00:20:39,514 --> 00:20:40,000
differently?

345
00:20:40,000 --> 00:20:44,519
Barrett Ames: Well, we kind of
segmented it up into different

346
00:20:44,519 --> 00:20:48,593
problems. Like finding windows
and doors, as opposed to

347
00:20:48,593 --> 00:20:52,963
dimensions and like, there
there's just some nice, areas to

348
00:20:52,963 --> 00:20:54,000
segment it up.

349
00:20:54,000 --> 00:20:57,536
Audrow Nash: I see, yeah. Okay.
Tell me more about that because

350
00:20:57,536 --> 00:21:01,128
that's very interesting. So you
trained on a bunch of floor plan

351
00:21:01,128 --> 00:21:04,214
data. What did that data look
like? And where were your

352
00:21:04,214 --> 00:21:05,000
sources? Yeah.

353
00:21:05,000 --> 00:21:10,012
Barrett Ames: For that. Yeah.
Our our sources, mostly come

354
00:21:10,012 --> 00:21:15,284
from builders that we've been
able to talk to. And it's just,

355
00:21:15,284 --> 00:21:20,210
you know, anyone and everyone
that we can get to give us,

356
00:21:20,210 --> 00:21:25,395
floor plan. What? We'll we'll
use the data to help train our

357
00:21:25,395 --> 00:21:26,000
models.

358
00:21:26,000 --> 00:21:29,158
Audrow Nash: Okay, okay. And
then so do you have to annotate

359
00:21:29,158 --> 00:21:32,000
it? Yeah. Initially. Or how did
how did all that work?

360
00:21:32,000 --> 00:21:35,933
Barrett Ames: Yeah. So we've
got, a labeling team that goes

361
00:21:35,933 --> 00:21:40,133
and annotate, really difficult
to outsource that, because I bet

362
00:21:40,133 --> 00:21:43,333
it's not. It's not your
standard, like, labeling

363
00:21:43,333 --> 00:21:47,333
problem. Right? It's not just a
classification problem. It's

364
00:21:47,333 --> 00:21:51,400
it's I mean, in some parts it
is, but, you know, there's just

365
00:21:51,400 --> 00:21:52,000
a lot of.

366
00:21:52,000 --> 00:21:54,919
Audrow Nash: Code you need to be
sort of subject experts, right?

367
00:21:54,919 --> 00:21:57,747
Some some, like you need some
training. Right? So I'm sure you

368
00:21:57,747 --> 00:22:00,575
can't like, or Amazon Turk or
whatever it is, that thing where

369
00:22:00,575 --> 00:22:03,267
you have, like, random people do
your task for a few cents,

370
00:22:03,267 --> 00:22:05,000
right? Or whatever it might be.
Right?

371
00:22:05,000 --> 00:22:08,749
Barrett Ames: Yeah. So we spent
six months training up a

372
00:22:08,749 --> 00:22:12,700
labeling team, to be able to
read construction plans and go

373
00:22:12,700 --> 00:22:16,449
and label everything. And
they're like, they're very key

374
00:22:16,449 --> 00:22:20,400
to the beginning of the whole
process. Right? Like, without

375
00:22:20,400 --> 00:22:24,618
them, we wouldn't have been able
to get to this point where now

376
00:22:24,618 --> 00:22:28,300
they're more supervisors, right?
They're doing a lot of

377
00:22:28,300 --> 00:22:32,251
validation. And, the models are
doing most of the work. But

378
00:22:32,251 --> 00:22:36,000
early on they were providing all
of the annotations for.

379
00:22:36,000 --> 00:22:40,424
Audrow Nash: That's cool. Okay.
So you have a bunch, you have a

380
00:22:40,424 --> 00:22:44,568
bunch of floor plan data. Yep.
You use that to learn how to

381
00:22:44,568 --> 00:22:48,992
generate plans that your robots
can make. You've had a bunch of

382
00:22:48,992 --> 00:22:53,347
people bootstrap this by adding
the data in and then you learn

383
00:22:53,347 --> 00:22:57,420
from that. What? It just sounds
hard to me like I wouldn't

384
00:22:57,420 --> 00:23:01,915
naively know that that would be,
a tractable problem in a sense,

385
00:23:01,915 --> 00:23:06,058
or that the data wouldn't be
just, like, total garbage. The

386
00:23:06,058 --> 00:23:09,991
output wouldn't be total
garbage. Yeah. For this kind of

387
00:23:09,991 --> 00:23:14,345
thing. Like, how do you, like,
it seems like an optimistic bet

388
00:23:14,345 --> 00:23:18,278
to me. And I'm glad that it
seems like it's working. But

389
00:23:18,278 --> 00:23:22,000
tell me a bit more about the
challenges around there.

390
00:23:22,000 --> 00:23:26,279
Barrett Ames: Yeah, I, I think,
first of all, you're not alone,

391
00:23:26,279 --> 00:23:30,491
right? Basically every engineer
that I've talked to about this

392
00:23:30,491 --> 00:23:34,566
problem has been like, well, why
why did you even solve this

393
00:23:34,566 --> 00:23:38,845
problem? Right? Like, can't we
just make, make architects build

394
00:23:38,845 --> 00:23:40,000
it in CAD, right.

395
00:23:40,000 --> 00:23:41,000
Audrow Nash: And expensive and
slow.

396
00:23:41,000 --> 00:23:45,578
Barrett Ames: Right, right. And,
the fact of the matter is that

397
00:23:45,578 --> 00:23:50,157
most PDFs are or most homes are,
are coming out of designs that

398
00:23:50,157 --> 00:23:54,372
are already exist, and 99% of
them exist in paper. So that

399
00:23:54,372 --> 00:23:58,732
that really provided the impetus
for going and building this

400
00:23:58,732 --> 00:24:03,092
thing. And it's, it's really
this combination, of our, like,

401
00:24:03,092 --> 00:24:07,162
ripping out all the information
or extracting all of the

402
00:24:07,162 --> 00:24:11,304
information from the PDF with
our optimization. That that

403
00:24:11,304 --> 00:24:15,228
makes it all work. Right?
Because, there's this, this,

404
00:24:15,228 --> 00:24:19,371
feedback cycle of, you know,
here, here's everything that

405
00:24:19,371 --> 00:24:23,949
we've learned from the plan. All
right, run it through the, the

406
00:24:23,949 --> 00:24:28,091
more procedural optimization
constrained approach. Right.

407
00:24:28,091 --> 00:24:32,015
And, the way that we built that
optimization is really

408
00:24:32,015 --> 00:24:36,739
important, because it can, well,
we we knew upfront that builders

409
00:24:36,739 --> 00:24:41,172
were going to have different
ways of building things. We knew

410
00:24:41,172 --> 00:24:45,750
that, every state you go to is
going to have different building

411
00:24:45,750 --> 00:24:50,329
codes. Definitely. So we had to
build that optimization in like

412
00:24:50,329 --> 00:24:52,000
a very, very nice, way.

413
00:24:52,000 --> 00:24:55,005
Audrow Nash: To, to include
those features. Right. Or like

414
00:24:55,005 --> 00:24:58,269
labels or something so that you
can use them. It's like so your

415
00:24:58,269 --> 00:25:00,756
model understands the
differences that it may be

416
00:25:00,756 --> 00:25:03,710
saying, do this with the
constraint of 16in between stud.

417
00:25:03,710 --> 00:25:06,819
Right. I'm making it up. But
like something like that, where

418
00:25:06,819 --> 00:25:09,617
one place might be 18in,
something like that, ten foot

419
00:25:09,617 --> 00:25:12,000
walls versus eight foot walls
versus whatever.

420
00:25:12,000 --> 00:25:15,837
Barrett Ames: Right. And that's
why we went with like a more

421
00:25:15,837 --> 00:25:19,866
classical optimization approach
there, because we can just swap

422
00:25:19,866 --> 00:25:23,895
in and out the constraints and
those, those give us like these,

423
00:25:23,895 --> 00:25:27,669
these hard, hard guarantees that
we're going to satisfy the

424
00:25:27,669 --> 00:25:31,378
building code. Even though
we're, we're shoving in a whole

425
00:25:31,378 --> 00:25:35,471
bunch of stuff up front. And if,
you know, if, if the labels are

426
00:25:35,471 --> 00:25:39,180
wrong or the annotations wrong,
which at this point, like,

427
00:25:39,180 --> 00:25:42,698
they're like 99%, which is
probably pretty good, right?

428
00:25:42,698 --> 00:25:45,000
Yeah. It's great. Oh, totally.
Then.

429
00:25:45,000 --> 00:25:48,176
Audrow Nash: And then hopefully
if they're 99%, that 1% is just

430
00:25:48,176 --> 00:25:51,000
like a little tweaking or you
run it again or something.

431
00:25:51,000 --> 00:25:55,274
Barrett Ames: Right, right,
right. So like the first time we

432
00:25:55,274 --> 00:25:59,689
built a plan, I went through, we
didn't we didn't know that we

433
00:25:59,689 --> 00:26:03,607
needed the software. Right. I
went through the plan and

434
00:26:03,607 --> 00:26:08,165
designed every single panel by.
Right. Yeah. And like, that took

435
00:26:08,165 --> 00:26:12,368
me a week. Wow. And then I was
like, this is nuts. I should

436
00:26:12,368 --> 00:26:16,499
find somebody else to do this.
So I went and asked a civil

437
00:26:16,499 --> 00:26:20,986
engineering firm like, hey, will
you guys design panels for us?

438
00:26:20,986 --> 00:26:24,832
They gave me a quote for
$20,000. Wow. I was like, all

439
00:26:24,832 --> 00:26:29,390
right, so we need to build this
software. And now we're down to,

440
00:26:29,390 --> 00:26:33,450
just under two hours with
validation. Right. And the dam,

441
00:26:33,450 --> 00:26:37,581
the goal is 30s. I want to be
able to put a plan in, get a

442
00:26:37,581 --> 00:26:41,000
code out while you're on the
phone with someone.

443
00:26:41,000 --> 00:26:44,719
Audrow Nash: Oh, that would be
so sick. Hell, yeah. What is it?

444
00:26:44,719 --> 00:26:48,260
Do you have any idea about the
cost reduction in that? So if

445
00:26:48,260 --> 00:26:51,920
they quoted you at $20,000, this
probably if you don't include

446
00:26:51,920 --> 00:26:55,108
all the R&D cost and time
developing it, your your run

447
00:26:55,108 --> 00:26:58,000
cost is probably like super
cheap. Yeah. I run a.

448
00:26:58,000 --> 00:27:01,364
Barrett Ames: Dollar or a few
dollars. Yeah I've run it. Our

449
00:27:01,364 --> 00:27:04,000
run cost with validation is
about 20 bucks now.

450
00:27:04,000 --> 00:27:09,000
Audrow Nash: It's amazing. Yeah.
So a thousand reduction.

451
00:27:09,000 --> 00:27:10,000
Barrett Ames: Yep.

452
00:27:10,000 --> 00:27:12,617
Audrow Nash: That's bonkers.
That's so fun. Yeah. That's

453
00:27:12,617 --> 00:27:15,000
that's, Beautiful. Yeah. Oh,
yeah. Yeah, I love it.

454
00:27:15,000 --> 00:27:18,315
Barrett Ames: Yeah. It's it's
it's, it's it's like one of the

455
00:27:18,315 --> 00:27:20,000
key driving things here, right?

456
00:27:20,000 --> 00:27:23,377
Audrow Nash: Yeah. Big
differentiator I imagine too,

457
00:27:23,377 --> 00:27:27,338
because it's a, it sounds like a
thorny problem and it sounds

458
00:27:27,338 --> 00:27:30,649
like. So when you say
generative, solutions, you're

459
00:27:30,649 --> 00:27:34,351
not talking like ChatGPT, you
know, generative AI, you're

460
00:27:34,351 --> 00:27:37,597
talking more like classical
optimization or. Yeah,

461
00:27:37,597 --> 00:27:40,000
constraints or. No. Like, what
is it.

462
00:27:40,000 --> 00:27:41,000
Barrett Ames: Like like.

463
00:27:41,000 --> 00:27:42,000
Audrow Nash: Evolutionary
strategy?

464
00:27:42,000 --> 00:27:45,785
Barrett Ames: No. You know what?
What kind of thing? When I mean

465
00:27:45,785 --> 00:27:49,334
generative, I really do mean
generative. Like there is. Like

466
00:27:49,334 --> 00:27:52,292
like there there is an
underlying distribution and

467
00:27:52,292 --> 00:27:56,018
we're learning that underlying
distribution. And, and you know,

468
00:27:56,018 --> 00:27:59,508
you can think about new things
coming in, and then sampling

469
00:27:59,508 --> 00:28:02,879
from that distribution. So you
can kind of condition your

470
00:28:02,879 --> 00:28:06,014
distribution based on
information that you've brought

471
00:28:06,014 --> 00:28:09,799
in. I've seen and then sampling
from that to give you an idea of

472
00:28:09,799 --> 00:28:13,171
confidence in something. Right.
So if you make a bunch of

473
00:28:13,171 --> 00:28:16,838
samples and they're all the same
thing, you should have pretty

474
00:28:16,838 --> 00:28:20,564
high confidence, like, oh, yeah,
this. Totally. Right. So it's,

475
00:28:20,564 --> 00:28:23,758
that I, when I, when I say
generative, I do mean like,

476
00:28:23,758 --> 00:28:25,000
literally generative.

477
00:28:25,000 --> 00:28:28,977
Audrow Nash: You're making a
distribution, you're sampling

478
00:28:28,977 --> 00:28:33,434
from it. Right. I see with, with
that, what kinds of decisions is

479
00:28:33,434 --> 00:28:37,000
the model making? Like what what
all is it deciding.

480
00:28:37,000 --> 00:28:40,000
Barrett Ames: It's it's, you
know.

481
00:28:40,000 --> 00:28:43,175
Audrow Nash: It's like stud
placement or like, configuration

482
00:28:43,175 --> 00:28:46,402
of wood when you're putting,
like, an actor somewhere or like

483
00:28:46,402 --> 00:28:49,735
all these things. Like what what
what all is it deciding. Yeah,

484
00:28:49,735 --> 00:28:50,000
it's.

485
00:28:50,000 --> 00:28:54,306
Barrett Ames: It's really
synthesizing all of the various

486
00:28:54,306 --> 00:28:58,838
constraints. Right. So it's it's
like, what's the, you know,

487
00:28:58,838 --> 00:29:03,295
given given window placement
here and dimensions here. Like

488
00:29:03,295 --> 00:29:07,827
what what where do we really
want that window? And so that's

489
00:29:07,827 --> 00:29:11,000
the it's kind of, it's taking
all of that,

490
00:29:11,000 --> 00:29:15,436
Audrow Nash: Hell yeah. And then
how do you. So you're getting a

491
00:29:15,436 --> 00:29:19,733
higher fidelity model with all
the studs and things like this,

492
00:29:19,733 --> 00:29:24,099
in their position. How do you go
from there to what your robots

493
00:29:24,099 --> 00:29:25,000
are building?

494
00:29:25,000 --> 00:29:30,206
Barrett Ames: Yeah, yeah, yeah.
The great question. It's, yeah,

495
00:29:30,206 --> 00:29:35,081
deeply technical question. But
at a high level, like, we're

496
00:29:35,081 --> 00:29:40,204
just thinking about the geometry
of the robot, the geometry of

497
00:29:40,204 --> 00:29:45,328
the problem, putting those two
together and saying, all right,

498
00:29:45,328 --> 00:29:50,451
we need this two by four to go
here. How does that impact what

499
00:29:50,451 --> 00:29:55,244
the robot can do? And vice
versa. It it's the the way that

500
00:29:55,244 --> 00:30:00,119
we've implemented it very much
falls in line with, with, an

501
00:30:00,119 --> 00:30:04,416
area in robotics called task in
motion planning. And

502
00:30:04,416 --> 00:30:09,704
specifically, like a lot of the
details are just stuff that come

503
00:30:09,704 --> 00:30:14,910
directly from my PhD. So I did a
lot of work on how you do this

504
00:30:14,910 --> 00:30:19,538
kind of high level task
planning, but it's parameterized

505
00:30:19,538 --> 00:30:24,000
by some behavior that the robot
has. And so that that,

506
00:30:24,000 --> 00:30:26,973
Audrow Nash: What do you mean? I
don't quite understand.

507
00:30:26,973 --> 00:30:30,212
Parameterize. So task I think
of. So you have something to do

508
00:30:30,212 --> 00:30:33,398
that requires many steps. And
then parameterized with by the

509
00:30:33,398 --> 00:30:36,000
behavior that the robot has, I
don't quite. Yeah.

510
00:30:36,000 --> 00:30:39,258
Barrett Ames: And so I kind of
got my words flip there. The

511
00:30:39,258 --> 00:30:42,294
behaviors are oh no are
parameterized in the sense that

512
00:30:42,294 --> 00:30:45,607
like, you know, if, if, you want
to put a two by four on the

513
00:30:45,607 --> 00:30:48,975
table, there's a whole bunch of
spots on the table, you could

514
00:30:48,975 --> 00:30:52,288
put it down, or you need to
figure out what the where, where

515
00:30:52,288 --> 00:30:54,000
on the table is it going to go.

516
00:30:54,000 --> 00:30:57,570
Audrow Nash: Yep. So your
parameters are dictating what to

517
00:30:57,570 --> 00:31:01,448
do next. So you have a bunch of
simple behaviors like place the

518
00:31:01,448 --> 00:31:05,203
thing here, but then you have
the parameters that specify the

519
00:31:05,203 --> 00:31:09,204
what should go in this location.
And then another piece should go

520
00:31:09,204 --> 00:31:12,897
in that location. Right for this
kind of thing. So you break

521
00:31:12,897 --> 00:31:16,775
down, but so that the process of
breaking down your output from

522
00:31:16,775 --> 00:31:20,468
the, the the blueprint to like
PDF to, a more detailed frame

523
00:31:20,468 --> 00:31:23,792
design, that part's very
interesting. And maybe that's

524
00:31:23,792 --> 00:31:27,608
very hard to talk about for this
kind of thing. Yeah. But, I'm

525
00:31:27,608 --> 00:31:31,055
curious about that trends,
because I imagine you have to

526
00:31:31,055 --> 00:31:34,871
figure out where all the nails
go. You have to figure out what

527
00:31:34,871 --> 00:31:38,811
order tasks must be done. Yeah.
Like, there's a lot there that I

528
00:31:38,811 --> 00:31:42,627
imagine is very hard. And the
way that I would naively suspect

529
00:31:42,627 --> 00:31:46,628
you implemented it is a bunch of
heuristics. Have been specified,

530
00:31:46,628 --> 00:31:50,506
and then you kind of run it over
this, like basically make sure

531
00:31:50,506 --> 00:31:54,199
you have something placed so you
can attach something to it.

532
00:31:54,199 --> 00:31:57,954
Nails go every this often this
kind of thing. But tell me how

533
00:31:57,954 --> 00:31:59,000
you've done that.

534
00:31:59,000 --> 00:32:04,497
Barrett Ames: Yeah. So at at a
design point, right. Like at the

535
00:32:04,497 --> 00:32:09,906
end of the, floor plan to design
stage, we end up with, just a

536
00:32:09,906 --> 00:32:12,000
CAD of the house, right?

537
00:32:12,000 --> 00:32:13,000
Audrow Nash: Yeah. That's so
cool.

538
00:32:13,000 --> 00:32:17,813
Barrett Ames: It has every two
by four, every nail. And so from

539
00:32:17,813 --> 00:32:22,703
that, we can we can take a panel
out, right? Like, take our Lego

540
00:32:22,703 --> 00:32:27,135
block of the house out of the
house and say, all right, we

541
00:32:27,135 --> 00:32:31,490
want to build this chunk. And,
and building that chunk is

542
00:32:31,490 --> 00:32:36,380
really just a planning problem.
Right. We know what our end goal

543
00:32:36,380 --> 00:32:40,888
is. And we know a bunch of
initial conditions which are the

544
00:32:40,888 --> 00:32:45,243
we know we have lumber and we
know where nails come from.

545
00:32:45,243 --> 00:32:47,000
Right. Yeah. They they.

546
00:32:47,000 --> 00:32:48,000
Audrow Nash: I like your
framing.

547
00:32:47,000 --> 00:32:51,369
Barrett Ames: Of this. Right.
Yeah. And then we just we have,

548
00:32:51,369 --> 00:32:55,094
you say go, right. We have a
classical planner, that

549
00:32:55,094 --> 00:32:59,463
interacts with the geometry of
the robot and searches through

550
00:32:59,463 --> 00:33:03,832
that space to find, you know,
given our set of behaviors, and

551
00:33:03,832 --> 00:33:07,700
the parameters for those
behaviors, how do we get from

552
00:33:07,700 --> 00:33:12,069
pile of lumber and the ability
to magically make nails appear

553
00:33:12,069 --> 00:33:13,000
to end panel?

554
00:33:13,000 --> 00:33:16,385
Audrow Nash: That's awesome. And
do you, do you do you kind of go

555
00:33:16,385 --> 00:33:18,000
through it in simulation first?

556
00:33:18,000 --> 00:33:19,000
Barrett Ames: Yeah. Yeah.

557
00:33:19,000 --> 00:33:20,000
Audrow Nash: So how does all
that work?

558
00:33:20,000 --> 00:33:23,436
Barrett Ames: Yeah, we do it all
first in simulation, to figure

559
00:33:23,436 --> 00:33:26,000
out like what's the feasible
path, right? Yeah.

560
00:33:26,000 --> 00:33:27,000
Audrow Nash: Because that sounds
very hard.

561
00:33:27,000 --> 00:33:30,511
Barrett Ames: Right. And a
feasible path. We get it in

562
00:33:30,511 --> 00:33:34,282
about like a minute and a half
to two minutes. And then we

563
00:33:34,282 --> 00:33:38,118
spend, like another 25 minutes
optimizing. And then once we

564
00:33:38,118 --> 00:33:41,759
have that optimized path, it's
giving us like high level

565
00:33:41,759 --> 00:33:45,464
behaviors and, and and their,
their parameters. Right. So

566
00:33:45,464 --> 00:33:49,365
it's, it's put down this two by
four, which for us is almost

567
00:33:49,365 --> 00:33:53,071
always the top plate goes in
first top plates. The two by

568
00:33:53,071 --> 00:33:57,102
four that goes to the top of the
wall. And then you put down a

569
00:33:57,102 --> 00:34:01,133
bunch of studs and then you put
down the bottom plate, and how

570
00:34:01,133 --> 00:34:05,099
you put the order in which you
put all of those studs down is

571
00:34:05,099 --> 00:34:08,675
very important because the
geometry of our end effector

572
00:34:08,675 --> 00:34:09,000
like.

573
00:34:09,000 --> 00:34:10,000
Audrow Nash: Can't get
everywhere, right.

574
00:34:10,000 --> 00:34:14,000
Barrett Ames: Unless you put it
in just the right order. And so

575
00:34:14,000 --> 00:34:18,000
we figure out that order, just
by simulating everything. Right.

576
00:34:18,000 --> 00:34:21,471
Audrow Nash: I would imagine.
Okay. That's really cool. And

577
00:34:21,471 --> 00:34:25,059
you basically is it like you're
trying to simulate it forward

578
00:34:25,059 --> 00:34:28,765
and then it's like blocked. And
then you throw out that one and

579
00:34:28,765 --> 00:34:32,353
you generate again and you do
that a few times. Or how do you

580
00:34:32,353 --> 00:34:35,765
I or it's an optimization
problem where you keep moving. I

581
00:34:35,765 --> 00:34:39,412
just I'm curious how you frame
the optimization problem to get

582
00:34:39,412 --> 00:34:42,941
the targets I suppose. Or like
the path that you're going to

583
00:34:42,941 --> 00:34:45,000
end up doing. Yeah. For
everything.

584
00:34:45,000 --> 00:34:49,532
Barrett Ames: Yeah. So we
because we know what we want at

585
00:34:49,532 --> 00:34:54,462
the end and we know everything
that we have, both material and

586
00:34:54,462 --> 00:34:59,471
behavior wise. Right. You can in
parallel look at a whole bunch

587
00:34:59,471 --> 00:35:04,560
of different configurations that
we might run into along the way

588
00:35:04,560 --> 00:35:08,774
and decide whether or not
they're valid. And that's a

589
00:35:08,774 --> 00:35:11,000
really fast simulation shot.

590
00:35:11,000 --> 00:35:12,000
Audrow Nash: Oh, that's cool.

591
00:35:11,000 --> 00:35:15,402
Barrett Ames: And so it just
shunts off whole branches, of of

592
00:35:15,402 --> 00:35:18,000
the planning problem really
quickly.

593
00:35:18,000 --> 00:35:21,084
Audrow Nash: Yeah. So it's a big
branching thing and you keep

594
00:35:21,084 --> 00:35:23,966
going. Oh, this one doesn't
work. Right. Cut that off and

595
00:35:23,966 --> 00:35:27,051
then keep proceeding down other
ones. And so you'll have this

596
00:35:27,051 --> 00:35:30,135
big explored space of possible
solutions. And then from there

597
00:35:30,135 --> 00:35:33,067
maybe that's where you're like
25 minute optimizing. Yeah.

598
00:35:33,067 --> 00:35:36,000
Yeah. You're like trying them
scoring them whatever it is.

599
00:35:36,000 --> 00:35:39,306
Barrett Ames: Right. Yeah.
Because the first the first plan

600
00:35:39,306 --> 00:35:42,780
through. Well it'll like pick up
and put down the tool like 25

601
00:35:42,780 --> 00:35:46,198
times or something. So it's free
because it's just doing like

602
00:35:46,198 --> 00:35:49,000
depth first search through this
massive, you know.

603
00:35:49,000 --> 00:35:53,491
Audrow Nash: Huge space. Right.
That's wild. What a cool thing.

604
00:35:53,491 --> 00:35:57,626
Okay. Hell, yeah. So then you
assemble your pallet and you

605
00:35:57,626 --> 00:36:01,761
load it on a truck and then ship
it out, and then builders

606
00:36:01,761 --> 00:36:05,682
assemble it. That's awesome.
Okay, so how do you handle

607
00:36:05,682 --> 00:36:10,173
exceptions with this? Like, I'm
sure that things don't go right

608
00:36:10,173 --> 00:36:14,736
all the time. Yeah, yeah. Like,
maybe sometimes you drop a piece

609
00:36:14,736 --> 00:36:19,299
of wood or a nail doesn't go in
the correct play. Probably go in

610
00:36:19,299 --> 00:36:23,719
the correct spot, but like, say
it chips the wood or something

611
00:36:23,719 --> 00:36:26,000
like I'm sure exceptions happen.

612
00:36:26,000 --> 00:36:27,000
Barrett Ames: Oh, yeah. What.

613
00:36:27,000 --> 00:36:29,213
Audrow Nash: What are some of
the exceptions and how do you

614
00:36:29,213 --> 00:36:30,000
deal with that. Yeah.

615
00:36:30,000 --> 00:36:34,137
Barrett Ames: Yeah, yeah. So the
exceptions are, they are

616
00:36:34,137 --> 00:36:37,911
exceptional, but not as
exceptional as I would like.

617
00:36:37,911 --> 00:36:42,484
Right. Of course, there's, you
know, if you think about a house

618
00:36:42,484 --> 00:36:46,694
being a thousand pieces of
lumber and 2000 joints, there's

619
00:36:46,694 --> 00:36:51,339
going to be like, we have to be
really, really good in order for

620
00:36:51,339 --> 00:36:56,056
there to be no errors. Yeah. And
so what we do is actually pretty

621
00:36:56,056 --> 00:37:00,048
interesting. And I've never,
I've never seen anybody in

622
00:37:00,048 --> 00:37:03,968
robotics do this. I hope I'm
wrong. And there are just

623
00:37:03,968 --> 00:37:08,250
smarter people out there. But
but we use this, thing called

624
00:37:08,250 --> 00:37:12,532
the good Turing estimator. So.
Okay. So Ian good was one of

625
00:37:12,532 --> 00:37:16,960
Alan Turing's, colleagues while
they were building the Enigma

626
00:37:16,960 --> 00:37:21,387
machine. And he developed this
estimator that just allows you

627
00:37:21,387 --> 00:37:25,597
to estimate the number of unseen
bugs. Right. Okay. Really

628
00:37:25,597 --> 00:37:29,879
useful. Interesting tool. And
it's actually how we quantify

629
00:37:29,879 --> 00:37:34,379
our exceptions, even though we
haven't seen all of them. So we

630
00:37:34,379 --> 00:37:39,024
just, write them down, as we see
them, there's a few kind of key

631
00:37:39,024 --> 00:37:43,306
hotspots, and then there's a
long tail of stuff. And that's

632
00:37:43,306 --> 00:37:47,661
one of the really, really key
things to construction is that

633
00:37:47,661 --> 00:37:51,000
there's not one big like, hairy
problem. It's.

634
00:37:51,000 --> 00:37:52,000
Audrow Nash: A bunch of little
one.

635
00:37:52,000 --> 00:37:55,983
Barrett Ames: Thousand little
ones. Yeah. Totally. And so I

636
00:37:55,983 --> 00:38:00,033
actually early on, I was really
lucky to find this guy who I

637
00:38:00,033 --> 00:38:03,948
consider to be the oracle for
the problem that BotBuilt is

638
00:38:03,948 --> 00:38:08,133
solving. He he spent the first
part of his career as a Seabee,

639
00:38:08,133 --> 00:38:12,385
in the Navy. So he is building
stuff. And then when he left the

640
00:38:12,385 --> 00:38:15,828
Navy, he went from a laborer to
a superintendent of

641
00:38:15,828 --> 00:38:20,013
construction. And then just, you
know, around that he was like

642
00:38:20,013 --> 00:38:24,333
40, 45 and decided like, I don't
want to do this anymore. Robots

643
00:38:24,333 --> 00:38:28,653
are cool. I'm going to go build
robots. And I did all the way up

644
00:38:28,653 --> 00:38:32,231
through the DARPA Grand
Challenge. Right. So he has a

645
00:38:32,231 --> 00:38:36,011
vehicle entered in the DARPA
Grand Challenge. Like, just

646
00:38:36,011 --> 00:38:39,926
awesome. Just like how how do
you find people like this? I

647
00:38:39,926 --> 00:38:44,044
don't know, I just got really
lucky and found this one, dude.

648
00:38:44,044 --> 00:38:48,161
Yeah. And so I invited worlds.
Amazing, right? I invited Mark

649
00:38:48,161 --> 00:38:52,414
to BotBuilt, and I said, like,
look, how do we die? And his his

650
00:38:52,414 --> 00:38:56,396
his, advice was exactly what I
just said. The construction.

651
00:38:56,396 --> 00:39:00,581
There's not one big problem that
kills you. It's not doing the

652
00:39:00,581 --> 00:39:04,631
thousand things well. And so
that's how we are. We're always

653
00:39:04,631 --> 00:39:08,614
trying to figure out what are
the thousand things, and then

654
00:39:08,614 --> 00:39:12,597
how do we how do we prioritize
those thousand things to do?

655
00:39:12,597 --> 00:39:16,849
Well, because, you know, we're
we're, we're constrained, right?

656
00:39:16,849 --> 00:39:21,102
Like, we can't solve them all at
once. So, like, the big things

657
00:39:21,102 --> 00:39:25,017
for us, over the last couple of
couple of months have been

658
00:39:25,017 --> 00:39:28,662
taking out those really high
frequency things like the

659
00:39:28,662 --> 00:39:32,780
computer vision, simulating all
of the lumber, and doing a, a

660
00:39:32,780 --> 00:39:36,965
high enough of every, degree.
Those, those kinds of errors are

661
00:39:36,965 --> 00:39:40,880
actually pretty nice in that,
when we do have a problem in

662
00:39:40,880 --> 00:39:45,200
labeling, we just have an image,
and we just need it labeled, so

663
00:39:45,200 --> 00:39:49,452
that we can actually just ship
out, get it labeled, comes back,

664
00:39:49,452 --> 00:39:53,502
robots keep going. And now we
have more labeled data. That's

665
00:39:53,502 --> 00:39:56,000
that's like, super nice problem.
For.

666
00:39:56,000 --> 00:39:58,436
Audrow Nash: Sure. Yeah.
Computer vision problems tend to

667
00:39:58,436 --> 00:40:01,000
be nice, right? A lot of times
if you get them. Right. Okay.

668
00:40:01,000 --> 00:40:05,715
Barrett Ames: But other like
other classes of problem are

669
00:40:05,715 --> 00:40:09,686
like, we, we have this
unobservable problem that

670
00:40:09,686 --> 00:40:14,981
happens, about like once every 5
or 6 homes, a nail will go into

671
00:40:14,981 --> 00:40:20,275
a piece of lumber and there's a
knot, hidden beneath the face of

672
00:40:20,275 --> 00:40:25,321
the lumber. And so the nail will
actually hit the knot, curve

673
00:40:25,321 --> 00:40:30,616
down around it, because it's not
more dense. Right? It's harder.

674
00:40:30,616 --> 00:40:35,000
Yeah. And it'll actually go down
into our work table.

675
00:40:35,000 --> 00:40:36,000
Audrow Nash: Then it gets stuck.

676
00:40:35,000 --> 00:40:38,760
Barrett Ames: And then it's
stuck on your work, and we have

677
00:40:38,760 --> 00:40:42,775
no way, like the first time it
happened. We're, like, trying to

678
00:40:42,775 --> 00:40:46,598
pull this panel up off the. What
happens. Right? And then we

679
00:40:46,598 --> 00:40:50,549
figured it out. So that's just
like, an exception that we have

680
00:40:50,549 --> 00:40:54,373
to, to, to sense, right? Like,
oh, we're trying to pull this

681
00:40:54,373 --> 00:40:58,387
thing up, and it's not, it's not
coming up. But we really don't

682
00:40:58,387 --> 00:41:01,000
have a great way of solving that
one yet.

683
00:41:01,000 --> 00:41:04,615
Audrow Nash: Can you put
everything on, like, a not hard,

684
00:41:04,615 --> 00:41:08,293
surface? Like, I'm thinking,
like, a spongy thing. But you

685
00:41:08,293 --> 00:41:12,288
may also want the rigidity for
your framing. So that's actually

686
00:41:12,288 --> 00:41:14,000
probably pretty hard. Yeah.

687
00:41:14,000 --> 00:41:17,235
Barrett Ames: Yeah, it's it's
definitely an interesting

688
00:41:17,235 --> 00:41:20,706
problem. I think, we haven't
even really tried to tackle it

689
00:41:20,706 --> 00:41:23,706
yet because it's so low
frequency, but, if I had to

690
00:41:23,706 --> 00:41:27,059
guess, you know, we'd use the
force torque sensors in the

691
00:41:27,059 --> 00:41:29,000
wrist and, and we'd notice,
like.

692
00:41:29,000 --> 00:41:30,000
Audrow Nash: Oh, hey, we.

693
00:41:30,000 --> 00:41:33,965
Barrett Ames: Can tell we're or
we can't pull this up. But there

694
00:41:33,965 --> 00:41:37,681
is like, when I pull it up, you
can feel that there's, like,

695
00:41:37,681 --> 00:41:41,212
this direction that it wants to
go. Yeah. And if you just

696
00:41:41,212 --> 00:41:44,000
allowed it to kind of follow
along that path,

697
00:41:44,000 --> 00:41:47,335
Audrow Nash: It might be able to
pull it up that way, but

698
00:41:47,335 --> 00:41:50,904
interesting. What a funny thing,
that the notches are harder.

699
00:41:50,904 --> 00:41:54,649
Yeah. And then, you get the nail
deforming for all of this. What

700
00:41:54,649 --> 00:41:58,160
a crazy thing. That's so funny.
You're like, why the hell is

701
00:41:58,160 --> 00:42:01,904
this nailed to the table? Like,
yeah, yeah, it's wild. Okay, any

702
00:42:01,904 --> 00:42:05,239
other classes of different
exceptions? You've kind of run

703
00:42:05,239 --> 00:42:06,000
across? Yeah.

704
00:42:06,000 --> 00:42:09,553
Barrett Ames: We see,
constrained motion planning

705
00:42:09,553 --> 00:42:13,541
problems a lot, right? Like,
that's just, fundamentally

706
00:42:13,541 --> 00:42:17,529
difficult field still. Yeah.
Right. Like you're, you're

707
00:42:17,529 --> 00:42:21,952
trying to, to sample from this
manifold in, seven dimensional

708
00:42:21,952 --> 00:42:25,722
space, but it's, it's, zero
volume manifold from the

709
00:42:25,722 --> 00:42:30,000
samplers standpoint. So, like,
yeah, that's just difficult.

710
00:42:30,000 --> 00:42:31,000
Audrow Nash: Super hard.

711
00:42:31,000 --> 00:42:35,063
Barrett Ames: And we we've made
some improvements there. And

712
00:42:35,063 --> 00:42:39,057
it's, it's all driven around,
the fact that we have to move

713
00:42:39,057 --> 00:42:42,984
these big pieces of lumber in
and, you know, very specific

714
00:42:42,984 --> 00:42:44,000
ways sometimes.

715
00:42:44,000 --> 00:42:47,658
Audrow Nash: Okay. Very
interesting. So for the end

716
00:42:47,658 --> 00:42:52,105
result, how does this compare?
Like how are, I guess, how many

717
00:42:52,105 --> 00:42:56,266
houses have you done so far and
how does it compare to the

718
00:42:56,266 --> 00:43:00,785
traditional building method of
like the people on site building

719
00:43:00,785 --> 00:43:01,000
it.

720
00:43:01,000 --> 00:43:08,000
Barrett Ames: Yeah, yeah. So,
we're just finishing up 27, so.

721
00:43:08,000 --> 00:43:10,000
Audrow Nash: 27 houses or.

722
00:43:10,000 --> 00:43:14,816
Barrett Ames: 27 houses. That's
awesome. Yeah. And, there are

723
00:43:14,816 --> 00:43:19,000
just people living in them,
right? Like, That's wild.

724
00:43:19,000 --> 00:43:20,000
Audrow Nash: What, what I bet
that feels.

725
00:43:20,000 --> 00:43:23,600
Barrett Ames: Yeah, yeah. It's
great. I mean, there's nothing

726
00:43:23,600 --> 00:43:27,141
like writing code. And then
somebody lives inside your code,

727
00:43:27,141 --> 00:43:30,564
right? Like. Oh, so definitely,
definitely a lot more like

728
00:43:30,564 --> 00:43:34,164
substance to that. It's it's
pretty cool. And they don't know

729
00:43:34,164 --> 00:43:38,000
that they're robot built, right?
Like it. It's just another home.

730
00:43:38,000 --> 00:43:42,603
Audrow Nash: Yeah, they're just
home. Right? It's great. So how

731
00:43:42,603 --> 00:43:47,279
does it compare to. So you have
27 homes. That's super exciting.

732
00:43:47,279 --> 00:43:51,662
How does it compare in like
efficiency or cost or things to,

733
00:43:51,662 --> 00:43:54,000
when you're building it on site?

734
00:43:54,000 --> 00:43:59,225
Barrett Ames: Yeah, yeah. So,
the robots are slower right now.

735
00:43:59,225 --> 00:44:04,449
But more cost effective because
one person can monitor a whole

736
00:44:04,449 --> 00:44:09,000
bunch of them. Right. And I
actually like the grocery.

737
00:44:09,000 --> 00:44:11,710
Audrow Nash: Store right thing
where you have one checkout

738
00:44:11,710 --> 00:44:14,000
person manning ten kiosks or
something like that.

739
00:44:14,000 --> 00:44:19,200
Barrett Ames: Exactly, exactly.
So our kiosk checker, his name

740
00:44:19,200 --> 00:44:24,065
is Tim. Tim used to work at,
chick fil A. He's just one of

741
00:44:24,065 --> 00:44:27,000
the guys his sons. He's 16. And
he.

742
00:44:27,000 --> 00:44:28,000
Audrow Nash: It's so good. Hell,
yeah.

743
00:44:28,000 --> 00:44:30,908
Barrett Ames: He can operate all
of the robots, right? And what

744
00:44:30,908 --> 00:44:31,000
a.

745
00:44:31,000 --> 00:44:32,000
Audrow Nash: Cool job for a 16
year old.

746
00:44:32,000 --> 00:44:35,364
Barrett Ames: Yeah, yeah, it's
it's he loves it. And we love

747
00:44:35,364 --> 00:44:38,897
having him because he. Yeah, he
pushes it way harder and faster

748
00:44:38,897 --> 00:44:42,486
than I will. Right. Like I know
too much 60. And of course, like

749
00:44:42,486 --> 00:44:44,000
he'll just go for it. Yeah.

750
00:44:44,000 --> 00:44:46,400
Audrow Nash: And so it's like
any good 16 year old right. So

751
00:44:46,400 --> 00:44:47,000
that's awesome.

752
00:44:47,000 --> 00:44:52,386
Barrett Ames: Exactly, exactly.
So we we Yeah, he can he can run

753
00:44:52,386 --> 00:44:57,688
all the robots. And that's where
we get a lot of our efficiency

754
00:44:57,688 --> 00:45:02,485
is just scaling out, Tim's
ability to build a whole bunch

755
00:45:02,485 --> 00:45:04,000
of panels at once.

756
00:45:04,000 --> 00:45:07,655
Audrow Nash: Nice. Yeah. And
that's awesome to Tim being 16.

757
00:45:07,655 --> 00:45:10,883
So I, I suppose, like, maybe he
has some construction

758
00:45:10,883 --> 00:45:14,294
background. But, like, that's
exciting that it's someone

759
00:45:14,294 --> 00:45:18,071
who's, like, relatively new to
the workforce. Right. And, it's

760
00:45:18,071 --> 00:45:21,543
a good model for that kind of
thing that probably is very

761
00:45:21,543 --> 00:45:25,320
favorable for you guys. Scaling
eventually. Yeah, that Tim can

762
00:45:25,320 --> 00:45:28,000
be right. Manning all these
machines. Right?

763
00:45:28,000 --> 00:45:32,450
Barrett Ames: Yeah. All right.
My co-founder, Brant was, in, in

764
00:45:32,450 --> 00:45:36,900
the Army, and he said that I had
to make the robots army proof.

765
00:45:36,900 --> 00:45:41,208
Right? So, like, you just have
to be able to have anyone, any

766
00:45:41,208 --> 00:45:45,729
old infantry, be able to run it,
and, that's, that's that's what

767
00:45:45,729 --> 00:45:47,000
we've tried to do.

768
00:45:47,000 --> 00:45:50,502
Audrow Nash: That's hilarious.
Yeah. One of my good friends is

769
00:45:50,502 --> 00:45:53,608
in the Navy, and he says the
same thing. But Navy brew,

770
00:45:53,608 --> 00:45:56,997
right? For this kind of thing.
Right? Hilarious. Okay. Hell,

771
00:45:56,997 --> 00:46:00,385
yeah. Do you have any idea? So
with this parallelization, do

772
00:46:00,385 --> 00:46:03,492
you have any idea in terms of
like time or cost, how it

773
00:46:03,492 --> 00:46:04,000
compares?

774
00:46:04,000 --> 00:46:08,366
Barrett Ames: Yeah, we're still
getting, better numbers on all

775
00:46:08,366 --> 00:46:09,000
of those.

776
00:46:09,000 --> 00:46:12,242
Audrow Nash: And maybe the
trajectory of your, improvements

777
00:46:12,242 --> 00:46:14,000
would be interesting to mention.

778
00:46:14,000 --> 00:46:18,545
Barrett Ames: Yeah, the
trajectory has been pretty

779
00:46:18,545 --> 00:46:24,273
awesome. Year ago to now, we've
improved our throughput by five

780
00:46:24,273 --> 00:46:29,000
x. So that's yeah, big, big
improvements there. The.

781
00:46:29,000 --> 00:46:32,000
Audrow Nash: Throughput, is that
correct? Yeah. Yep. Okay. Yep.

782
00:46:32,000 --> 00:46:36,065
Barrett Ames: Yep. And we're
about to drop a big feature

783
00:46:36,065 --> 00:46:39,839
that's going to double our
speed. So that that'll be

784
00:46:39,839 --> 00:46:41,000
awesome as well.

785
00:46:41,000 --> 00:46:44,000
Audrow Nash: And so be ten times
better than it was a year ago.

786
00:46:44,000 --> 00:46:46,714
Roughly. Yeah. Yeah, yeah, I
like that. Yeah. That growth

787
00:46:46,714 --> 00:46:47,000
curve.

788
00:46:47,000 --> 00:46:51,373
Barrett Ames: That's awesome.
Right? Right. Yeah. You don't

789
00:46:51,373 --> 00:46:55,598
you that's that's the kind of
growth curve you need for a

790
00:46:55,598 --> 00:47:00,046
venture backed company. Right.
Yeah. So and then the kind of

791
00:47:00,046 --> 00:47:04,419
the other big wins that we see
are that our panels are our,

792
00:47:04,419 --> 00:47:09,163
like, frankly beautiful, in that
they're, they're very precisely

793
00:47:09,163 --> 00:47:13,610
made, and put together. We, we
were just walking one of them

794
00:47:13,610 --> 00:47:18,131
recently with, an inspector here
in Durham. And he said, wow,

795
00:47:18,131 --> 00:47:22,430
this is like, the best framing
that I have seen in Durham.

796
00:47:22,430 --> 00:47:26,952
Right? Which, like, feels really
good. But it also just means

797
00:47:26,952 --> 00:47:31,399
that at the end of the day, the
consumer is getting a better

798
00:47:31,399 --> 00:47:35,624
product. And it actually takes
significantly less time to

799
00:47:35,624 --> 00:47:40,146
inspect. Right. So, we've seen
inspection times drop from two

800
00:47:40,146 --> 00:47:44,000
and a half to three hours, down
to 30 to 45 minutes.

801
00:47:44,000 --> 00:47:45,000
Audrow Nash: That's awesome,
because.

802
00:47:45,000 --> 00:47:47,597
Barrett Ames: The inspectors
just gain a lot more trust, a

803
00:47:47,597 --> 00:47:48,000
lot more.

804
00:47:48,000 --> 00:47:51,835
Audrow Nash: Quickly. Now, not
to the, like, I'm sure that's

805
00:47:51,835 --> 00:47:55,223
true, but to play devil's
advocate. Yeah. With, like,

806
00:47:55,223 --> 00:47:58,867
ChatGPT and things, they can
seem very sophisticated. And

807
00:47:58,867 --> 00:48:02,702
it's kind of easy to trust for a
lot of things. But there is

808
00:48:02,702 --> 00:48:06,346
definitely cases where they,
like, we'll just hallucinate

809
00:48:06,346 --> 00:48:10,309
random stuff. And so the trust
can occasionally be betrayed in

810
00:48:10,309 --> 00:48:14,144
some sense, or just like it
wasn't well founded. Right. This

811
00:48:14,144 --> 00:48:17,788
sounds a lot more structured
than that. But how would you

812
00:48:17,788 --> 00:48:21,687
think about that kind of thing
if there is kind of a downside

813
00:48:21,687 --> 00:48:25,650
to decreasing the, increasing
the trust so much that they just

814
00:48:25,650 --> 00:48:28,974
let you go on past, or on
uncosted if something is a

815
00:48:28,974 --> 00:48:32,937
little bit wrong because they
were just like, yeah, everything

816
00:48:32,937 --> 00:48:36,069
looks so good because they
assume a high level of

817
00:48:36,069 --> 00:48:40,160
competence for the builders that
are the machines. But you still

818
00:48:40,160 --> 00:48:44,124
can do something wonky somewhere
and they just may not see it.

819
00:48:44,124 --> 00:48:47,000
How would you think about that
kind of thing?

820
00:48:47,000 --> 00:48:54,000
Barrett Ames: Yeah, that's an
interesting question.

821
00:48:54,000 --> 00:48:58,000
Audrow Nash: And obviously it's
like you're on your way to doing

822
00:48:58,000 --> 00:49:02,000
very good work. Right? So in the
path of doing good work, you'll

823
00:49:02,000 --> 00:49:05,750
come like you're on the right
path for sure. Right? But this

824
00:49:05,750 --> 00:49:07,000
may be a thing that,

825
00:49:07,000 --> 00:49:10,745
Barrett Ames: Yeah. What could
occur. Yeah. So one, one thing,

826
00:49:10,745 --> 00:49:14,128
you know, like, there are
they're just pattern matching,

827
00:49:14,128 --> 00:49:17,691
right? Like they're when they
look through a house, they're

828
00:49:17,691 --> 00:49:21,436
just kind of looking for things
that are odd and out of place.

829
00:49:21,436 --> 00:49:24,638
And our stuff tends to be
extremely regular, orderly.

830
00:49:24,638 --> 00:49:25,000
Right.

831
00:49:25,000 --> 00:49:27,122
Audrow Nash: And so that so
they're looking for exceptions.

832
00:49:27,122 --> 00:49:29,137
Right. And you don't have many
exceptions. Right. That's

833
00:49:29,137 --> 00:49:30,000
exciting right. Oh yeah.

834
00:49:30,000 --> 00:49:34,358
Barrett Ames: Right. And so that
that gives them the confidence.

835
00:49:34,358 --> 00:49:38,172
But it is it is a human
inspection process. Right. From,

836
00:49:38,172 --> 00:49:42,394
from from the building inspector
standpoint, we actually do QA

837
00:49:42,394 --> 00:49:46,616
internally as well. Right. So
because we we have this 3D model

838
00:49:46,616 --> 00:49:49,000
of what we expect to build, we
can.

839
00:49:49,000 --> 00:49:50,000
Audrow Nash: And you've added
against that.

840
00:49:50,000 --> 00:49:54,503
Barrett Ames: Right. We can just
look at that. And so that gives

841
00:49:54,503 --> 00:49:58,653
us, higher confidence and I
think I'm not positive yet, but

842
00:49:58,653 --> 00:50:03,156
I think it will also help, with
insurance matters down the road.

843
00:50:03,156 --> 00:50:04,000
Oh, there's.

844
00:50:04,000 --> 00:50:05,000
Audrow Nash: That's super cool.

845
00:50:04,000 --> 00:50:07,948
Barrett Ames: Like, building
insurance is kind of this really

846
00:50:07,948 --> 00:50:12,090
messy place, like, you know. Oh,
yeah, a wall falls down and all

847
00:50:12,090 --> 00:50:16,102
the contractors just go, not me.
And then the judge just gives

848
00:50:16,102 --> 00:50:20,050
everybody an equal share. And
because we capture so much data

849
00:50:20,050 --> 00:50:23,868
on every joint and every piece
of lumber will be it will be

850
00:50:23,868 --> 00:50:28,010
able to very clearly say, like,
look, we did our part. And if we

851
00:50:28,010 --> 00:50:32,023
messed up, like, yeah, we'll,
we'll also be able to say, yeah,

852
00:50:32,023 --> 00:50:35,000
of course, clearly like, yeah,
that's our bad.

853
00:50:35,000 --> 00:50:38,453
Audrow Nash: Yeah. Because you
have like a 3D camera and you

854
00:50:38,453 --> 00:50:41,791
have like a camera image. Right.
And probably a 3D scan of

855
00:50:41,791 --> 00:50:45,360
everything. And so they could be
like, this wasn't done right.

856
00:50:45,360 --> 00:50:48,986
And it's like, well, leaving our
facility, here's the thing. It

857
00:50:48,986 --> 00:50:52,439
was perfect. Right. Which yeah,
I view that as a big win for

858
00:50:52,439 --> 00:50:56,065
insurance. Yep. For sure. Yep.
It's so interesting to me that a

859
00:50:56,065 --> 00:50:59,460
lot of these automation, a lot
of the output or a result of

860
00:50:59,460 --> 00:51:02,971
increasing automation is better
monitoring and that it's very

861
00:51:02,971 --> 00:51:06,137
interesting to me that like
insurance seems to be a big

862
00:51:06,137 --> 00:51:09,417
winner. Yeah. Through this
because and just like business

863
00:51:09,417 --> 00:51:12,871
tracking right. And things like
you mentioned like inventory

864
00:51:12,871 --> 00:51:15,000
tracking and things very
interesting.

865
00:51:15,000 --> 00:51:18,750
Barrett Ames: Yeah. I think so
I've been, I've been thinking

866
00:51:18,750 --> 00:51:22,500
about this like, framework or
paradigm for like how to build

867
00:51:22,500 --> 00:51:25,563
robots. Right. And I call it
teleoperation driven

868
00:51:25,563 --> 00:51:29,375
development. And the idea is
basically like first you have to

869
00:51:29,375 --> 00:51:33,188
design the mechanism, then you
have to run the mechanism, but

870
00:51:33,188 --> 00:51:36,500
you want to run it with
teleoperation. And that helps

871
00:51:36,500 --> 00:51:38,000
you get this, this idea.

872
00:51:38,000 --> 00:51:39,000
Audrow Nash: Problem fit.

873
00:51:39,000 --> 00:51:43,200
Barrett Ames: Right like problem
fit, but also like you have to

874
00:51:43,200 --> 00:51:47,333
have really good perception in
order to operate it. Right. And

875
00:51:47,333 --> 00:51:50,667
so you get this like,
observability from a control

876
00:51:50,667 --> 00:51:54,400
standpoint, very early in the
process. And because we're

877
00:51:54,400 --> 00:51:58,600
building control algorithms that
are subhuman, they always seem

878
00:51:58,600 --> 00:52:02,533
to need more perception than
humans, right? Like they can't

879
00:52:02,533 --> 00:52:04,000
keep this in. What do.

880
00:52:04,000 --> 00:52:05,000
Audrow Nash: You mean? Subhuman?

881
00:52:05,000 --> 00:52:08,000
Barrett Ames: Just like like
below, below human care.

882
00:52:08,000 --> 00:52:10,568
Audrow Nash: But they're not as
sophisticated as us, I guess.

883
00:52:10,568 --> 00:52:13,011
Yeah, yeah. Or they may be in
other ways, but for the most

884
00:52:13,011 --> 00:52:15,453
part, they're less sophisticated
than us, right? Right. In

885
00:52:15,453 --> 00:52:16,000
sensing. Say.

886
00:52:16,000 --> 00:52:18,852
Barrett Ames: Right. Right.
Well, I'm kind of like keeping

887
00:52:18,852 --> 00:52:21,852
an internal model, right? Like
where we kind of sub out a lot

888
00:52:21,852 --> 00:52:24,951
of our perception for like, this
really rich internal model, at

889
00:52:24,951 --> 00:52:28,000
least, I think. I don't know,
I'm not a neuroscientist, but I.

890
00:52:28,000 --> 00:52:31,000
Audrow Nash: It sounds right to
me. Yeah.

891
00:52:31,000 --> 00:52:33,718
Barrett Ames: And then, you
know, once you've done the

892
00:52:33,718 --> 00:52:36,587
teleoperation stuff, you're,
you're you've now guaranteed

893
00:52:36,587 --> 00:52:39,607
kind of two things. One, that
you have enough perception and

894
00:52:39,607 --> 00:52:42,477
two, that, you have enough
actuation to get the job done,

895
00:52:42,477 --> 00:52:45,497
then you can go and build your
own. Yeah, build your control

896
00:52:45,497 --> 00:52:46,000
algorithm.

897
00:52:46,000 --> 00:52:49,847
Audrow Nash: I like that, yeah.
I've been thinking of, like, one

898
00:52:49,847 --> 00:52:53,094
thing through all these
interviews that I've seen as a

899
00:52:53,094 --> 00:52:56,760
big pattern and something that
I've often said, I suppose, is

900
00:52:56,760 --> 00:53:00,007
like a human in the loop.
Development is very good for

901
00:53:00,007 --> 00:53:03,614
startups, for the reasons you
described. But also, like, you

902
00:53:03,614 --> 00:53:07,341
can gradually automate parts of
it. You're getting data you're

903
00:53:07,341 --> 00:53:11,068
checking for, unknown unknowns
because you're actually getting

904
00:53:11,068 --> 00:53:14,855
out there and solving it. Yeah.
And using Tim or similar people

905
00:53:14,855 --> 00:53:18,522
where it's like you can slowly
scale the number of robots one

906
00:53:18,522 --> 00:53:22,189
person is able to manage, then
eventually you can get to full

907
00:53:22,189 --> 00:53:26,096
autonomy. Right? And at the same
time, you're proving your market

908
00:53:26,096 --> 00:53:29,222
right, which is really
effective, so that you can go

909
00:53:29,222 --> 00:53:32,949
get investment and, you can go
start working with and pleasing

910
00:53:32,949 --> 00:53:36,737
customers. Yeah, for this kind
of thing. So it's they seem very

911
00:53:36,737 --> 00:53:40,524
similar, these two ideas. And I
really like your perspective on

912
00:53:40,524 --> 00:53:44,010
it, where you have enough
sensing and then you can confirm

913
00:53:44,010 --> 00:53:47,677
that the actuation can do it. So
now you can start automating

914
00:53:47,677 --> 00:53:49,000
right for this. Right.

915
00:53:49,000 --> 00:53:53,643
Barrett Ames: Right. Yeah. Yeah.
That's your your thought process

916
00:53:53,643 --> 00:53:57,643
there I think exactly. Mirrors
kind of how I think about

917
00:53:57,643 --> 00:54:00,000
teleoperation driven
development.

918
00:54:00,000 --> 00:54:02,919
Audrow Nash: Yeah. Yeah. And
it's so interesting to me

919
00:54:02,919 --> 00:54:06,000
because a lot of the like
hardcore roboticists don't like

920
00:54:06,000 --> 00:54:09,405
this approach. But it is I think
a very fast way to find value.

921
00:54:09,405 --> 00:54:12,595
Right. Like where you can add
value. Yeah. And to me that's

922
00:54:12,595 --> 00:54:16,000
more interesting. Yeah. You know
to make something really work.

923
00:54:16,000 --> 00:54:17,000
Barrett Ames: Well it also
really.

924
00:54:17,000 --> 00:54:19,000
Audrow Nash: And solve
something. Yeah. Yeah. Go ahead.

925
00:54:19,000 --> 00:54:22,062
Barrett Ames: I was just gonna
say it also really helps you

926
00:54:22,062 --> 00:54:25,176
quantify the problems to solve,
right. Like totally the good

927
00:54:25,176 --> 00:54:28,031
Turing test that we do just
really helps us figure out,

928
00:54:28,031 --> 00:54:31,197
like, okay, these are the things
with high frequency and high

929
00:54:31,197 --> 00:54:34,000
severity. Those are the things
that are worth solving.

930
00:54:34,000 --> 00:54:37,028
Audrow Nash: Yeah. Right. It's a
good way of weighting the

931
00:54:37,028 --> 00:54:40,056
problems to solve for sure.
Right. Yeah. Because frequency

932
00:54:40,056 --> 00:54:43,241
time severity is the urgency.
Right, right. I don't know that

933
00:54:43,241 --> 00:54:46,373
thing but I'm gonna look at
Turing and good. That seems like

934
00:54:46,373 --> 00:54:47,000
a wonderful.

935
00:54:47,000 --> 00:54:51,317
Barrett Ames: Yeah it's, it's a
model. It's a pretty interesting

936
00:54:51,317 --> 00:54:55,500
way to think about how robust
your system should be. The other

937
00:54:55,500 --> 00:54:59,615
thing that I've done kind of,
kind of with it is I've applied

938
00:54:59,615 --> 00:55:03,393
it to scheduling, which is
pretty, pretty interesting as

939
00:55:03,393 --> 00:55:04,000
well. We.

940
00:55:04,000 --> 00:55:05,000
Audrow Nash: What do you mean by
schedule like.

941
00:55:05,000 --> 00:55:09,101
Barrett Ames: Like engineering
scheduling. So, so if you have,

942
00:55:09,101 --> 00:55:13,070
if you've got, some, some big
engineering task that you want

943
00:55:13,070 --> 00:55:17,304
to do, you get a whole bunch of
engineers together, you all kind

944
00:55:17,304 --> 00:55:21,272
of lay out the path that you
would want to do. And the areas

945
00:55:21,272 --> 00:55:22,000
where like.

946
00:55:22,000 --> 00:55:24,762
Audrow Nash: Highest technical
risk, right, kind of thing.

947
00:55:24,762 --> 00:55:25,000
Well.

948
00:55:25,000 --> 00:55:28,523
Barrett Ames: Those will show up
as, like a whole bunch of

949
00:55:28,523 --> 00:55:32,350
different ways of doing the same
thing, and where people are in

950
00:55:32,350 --> 00:55:35,995
agreement. It kind of collapses
down to like, one thing. And

951
00:55:35,995 --> 00:55:38,000
the, the cool thing about is
the.

952
00:55:38,000 --> 00:55:41,013
Audrow Nash: Frequency, right?
Dimension is the consensus.

953
00:55:41,013 --> 00:55:42,000
Right. Interesting.

954
00:55:42,000 --> 00:55:45,916
Barrett Ames: And so the, the
cool thing that you get out of

955
00:55:45,916 --> 00:55:49,637
the good Turing estimation is
you get now instead of just

956
00:55:49,637 --> 00:55:53,358
saying, like, hey, to the
schedule, you get to scale your

957
00:55:53,358 --> 00:55:57,274
schedule estimate based off of
uncertainty. Oh, I love that.

958
00:55:57,274 --> 00:56:01,060
And so the, the, the awesome
thing that we were able to do

959
00:56:01,060 --> 00:56:04,911
with it was I took a six month
project. We laid it out with

960
00:56:04,911 --> 00:56:07,000
this, and we were within a week.

961
00:56:07,000 --> 00:56:10,130
Audrow Nash: Damn. That's
awesome. Yeah, yeah, it's like

962
00:56:10,130 --> 00:56:13,093
how a GPS will take into it.
Like, when you're taking

963
00:56:13,093 --> 00:56:16,000
directions on a map, it'll take
into account all the

964
00:56:16,000 --> 00:56:18,907
uncertainty. But by the
distributions of how traffic

965
00:56:18,907 --> 00:56:22,093
progresses and whatever, on
average, you'll arrive pretty

966
00:56:22,093 --> 00:56:25,503
close to the time you want it,
right? That sounds super cool.

967
00:56:25,503 --> 00:56:28,466
If you could do that with
engineering effort, because

968
00:56:28,466 --> 00:56:31,820
that's not like everything is
always happening later than it

969
00:56:31,820 --> 00:56:35,118
was expected. And things like
this. And it's interesting if

970
00:56:35,118 --> 00:56:38,528
you can codify your problem in
this way and then get a really

971
00:56:38,528 --> 00:56:41,379
good estimate, that takes into
account a lot of the

972
00:56:41,379 --> 00:56:43,000
uncertainty. Yep. Super cool.

973
00:56:43,000 --> 00:56:46,125
Barrett Ames: Yeah, it's been
fun. I'm still still

974
00:56:46,125 --> 00:56:49,938
experimenting with it, but at
least the early tests I've done

975
00:56:49,938 --> 00:56:52,000
with it. It seems really
helpful.

976
00:56:52,000 --> 00:56:55,629
Audrow Nash: I would absolutely
love if you make like a blog

977
00:56:55,629 --> 00:56:59,137
post or a video or something
once you have it more fleshed

978
00:56:59,137 --> 00:57:02,525
out. Yeah, yeah. Or just hell,
we can do another podcast

979
00:57:02,525 --> 00:57:06,093
interview. Yeah. And, I would
love to learn more about that

980
00:57:06,093 --> 00:57:09,662
because that that's a hard
problem. Yes. And, yeah, I mean,

981
00:57:09,662 --> 00:57:13,352
people do or companies do scrum
and stuff like this and yeah,

982
00:57:13,352 --> 00:57:16,799
not that satisfying in my
experience. And this, this kind

983
00:57:16,799 --> 00:57:20,489
of uncertainty because I think
that's always funny with scrum

984
00:57:20,489 --> 00:57:24,421
is it's like, well, I don't know
the task. So therefore it's hard

985
00:57:24,421 --> 00:57:28,171
to do it for this kind of thing.
Right. And so this is like it

986
00:57:28,171 --> 00:57:31,256
takes into account the
uncertainty. Right. Which is

987
00:57:31,256 --> 00:57:34,764
quite cool. Right. Yep. Very
cool. Do you have any idea of

988
00:57:34,764 --> 00:57:38,453
the potential cost? I know I
keep asking around the same kind

989
00:57:38,453 --> 00:57:41,659
of things, but yeah, I'm
wondering like, do you think

990
00:57:41,659 --> 00:57:45,107
it'll be a good bit faster
because like one of the things

991
00:57:45,107 --> 00:57:48,676
that comes to my mind through
most of my interviews is that

992
00:57:48,676 --> 00:57:51,458
there are massive labor
shortages, and I think

993
00:57:51,458 --> 00:57:54,845
construction is one of those
areas particularly strongly

994
00:57:54,845 --> 00:57:58,656
affected. Yeah. And so I wonder
how, like just getting the task

995
00:57:58,656 --> 00:58:02,587
done at all is fantastic. But if
it's done really efficiently and

996
00:58:02,587 --> 00:58:06,519
you make a lot of gains in terms
of speed and it allows people to

997
00:58:06,519 --> 00:58:10,148
use it for hours instead of 20
days or whatever. You said at

998
00:58:10,148 --> 00:58:13,414
the beginning. Yeah. I maybe
that's the good metric to

999
00:58:13,414 --> 00:58:16,802
assess. I'm curious. Kind of
like an apples to apples in

1000
00:58:16,802 --> 00:58:20,612
terms of like, you build a house
this way and you build a house

1001
00:58:20,612 --> 00:58:24,483
the traditional way. I'm curious
if there's like a cost estimate

1002
00:58:24,483 --> 00:58:28,415
and then a person time estimate,
maybe 20 weeks to the other one.

1003
00:58:28,415 --> 00:58:32,105
But can you talk a little bit to
this? And it sounds like the

1004
00:58:32,105 --> 00:58:35,976
trajectory is improving rapidly.
Right. But I'd be interested to

1005
00:58:35,976 --> 00:58:39,000
hear where you are and where you
think it's going.

1006
00:58:39,000 --> 00:58:42,240
Barrett Ames: Yeah, we're we're
right at parity right now. Like

1007
00:58:42,240 --> 00:58:45,480
it depends a little bit on the
complexity of the job. But we're

1008
00:58:45,480 --> 00:58:48,000
like up and down right around
parity with humans.

1009
00:58:48,000 --> 00:58:50,000
Audrow Nash: That's exciting.

1010
00:58:50,000 --> 00:58:55,185
Barrett Ames: And so I think,
you know, we will be we still

1011
00:58:55,185 --> 00:59:00,194
haven't turned the dial up on
the actual industrial arms,

1012
00:59:00,194 --> 00:59:05,467
right. Like they're still set at
20% speed. So we can we can

1013
00:59:05,467 --> 00:59:10,915
certainly get a lot more speed
out there. And the big the, the

1014
00:59:10,915 --> 00:59:16,012
big thing that we're, we're
releasing shortly will be what

1015
00:59:16,012 --> 00:59:19,000
will help improve speed even
more.

1016
00:59:19,000 --> 00:59:22,600
Audrow Nash: Can you speak to
that at all or is that, secret

1017
00:59:22,600 --> 00:59:25,000
thing that will be announced
eventually.

1018
00:59:25,000 --> 00:59:29,962
Barrett Ames: Well, just, stay
tuned. It'll it'll be it'll be a

1019
00:59:29,962 --> 00:59:35,082
pretty cool video. And I'll, Oh,
hell yeah. So we'll we'll, we'll

1020
00:59:35,082 --> 00:59:39,493
show it, for sure. But, yeah,
it's there, there's a lot,

1021
00:59:39,493 --> 00:59:44,455
efficiency still to be squeezed
out. And we're, you know, we're

1022
00:59:44,455 --> 00:59:48,000
we're making good progress on
that every day.

1023
00:59:48,000 --> 00:59:51,708
Audrow Nash: Yep. So cost is
about the same. And so if you

1024
00:59:51,708 --> 00:59:55,735
make it faster, that'll make it
cost less. Yeah. I suppose with

1025
00:59:55,735 --> 00:59:59,379
all of this and I mean, you guys
from my. So you guys are

1026
00:59:59,379 --> 01:00:02,000
approaching series A. Yep.
Correct. Yeah.

1027
01:00:02,000 --> 01:00:05,429
Barrett Ames: Yeah yeah. We're
just about to do our series A

1028
01:00:05,429 --> 01:00:06,000
fundraise.

1029
01:00:06,000 --> 01:00:11,000
Audrow Nash: Hell yeah. Want to
share any details with that or.

1030
01:00:11,000 --> 01:00:14,916
Barrett Ames: Yeah. You know,
we're, we're, always looking for

1031
01:00:14,916 --> 01:00:18,453
investors who are interested or
experienced in robotics.

1032
01:00:18,453 --> 01:00:22,368
There's, you know, I think an
ever growing group of people out

1033
01:00:22,368 --> 01:00:23,000
there who.

1034
01:00:23,000 --> 01:00:24,000
Audrow Nash: It is growing.

1035
01:00:24,000 --> 01:00:28,722
Barrett Ames: Yeah. Who who
understand? I call it the return

1036
01:00:28,722 --> 01:00:33,523
profile. But, like, it's, you
know, the moats are much bigger

1037
01:00:33,523 --> 01:00:38,481
in robotics. Totally. And so I
think the, the long term returns

1038
01:00:38,481 --> 01:00:43,597
will be higher. But it does take
it does take longer. Right. Like

1039
01:00:43,597 --> 01:00:48,162
there there's. I can't ask
ChatGPT to build me, robot code

1040
01:00:48,162 --> 01:00:52,963
yet. Well, you can. Yeah, yeah
yeah yeah yeah, yeah. But it's

1041
01:00:52,963 --> 01:00:58,000
not a it's it's certainly not as
easy as building a web app now.

1042
01:00:58,000 --> 01:01:01,668
Audrow Nash: Definitely. Yeah.
Yeah. It's, there's a new breed

1043
01:01:01,668 --> 01:01:04,980
of investor that's, I'm seeing
that's coming out, that's

1044
01:01:04,980 --> 01:01:08,648
thinking long term for robotics
and it's very exciting to see,

1045
01:01:08,648 --> 01:01:12,020
and they understand a lot of the
grievances that robotics

1046
01:01:12,020 --> 01:01:15,746
companies have had in the last
ten years working with investors

1047
01:01:15,746 --> 01:01:19,000
that try to turn everything into
software as a service.

1048
01:01:19,000 --> 01:01:20,000
Barrett Ames: Right.

1049
01:01:20,000 --> 01:01:21,000
Audrow Nash: With subscription.

1050
01:01:21,000 --> 01:01:24,779
Barrett Ames: Models and stuff.
And I think as a community,

1051
01:01:24,779 --> 01:01:28,750
we're also just doing a better
job of getting to value faster,

1052
01:01:28,750 --> 01:01:32,401
right? Like totally doing doing
this teleoperation driven

1053
01:01:32,401 --> 01:01:36,564
development type approach, like,
let's, provide and verify value,

1054
01:01:36,564 --> 01:01:40,535
much, much earlier. Whereas, you
know, like, I totally used to

1055
01:01:40,535 --> 01:01:44,314
think this way that like, yeah,
we just got to automate the

1056
01:01:44,314 --> 01:01:48,413
whole thing. But that's, that's,
that's clearly the wrong way to

1057
01:01:48,413 --> 01:01:52,000
go about building a robotics
startup. Right. Like, yeah.

1058
01:01:52,000 --> 01:01:54,000
Audrow Nash: Not very capital
efficient. Right. All right.

1059
01:01:54,000 --> 01:01:58,895
Barrett Ames: For this. Right.
So I think, you know, they're

1060
01:01:58,895 --> 01:02:03,546
they're the kind of investor
community pairing I think is

1061
01:02:03,546 --> 01:02:08,278
really close to being at, I kind
of think about it as like

1062
01:02:08,278 --> 01:02:13,500
getting ever closer to, like 20.
Well, like 2008, 2009, like web

1063
01:02:13,500 --> 01:02:18,395
dev or SAS, right? Where you're
like, yeah, Ruby on Rails is

1064
01:02:18,395 --> 01:02:23,209
just starting to come out.
People are like, oh, yeah, I can

1065
01:02:23,209 --> 01:02:28,349
build all these cool things now.
And yeah, I think, I think the

1066
01:02:28,349 --> 01:02:33,000
category as a whole is about to
have an inflection point.

1067
01:02:33,000 --> 01:02:36,502
Audrow Nash: I think that's
true. And I think that is going

1068
01:02:36,502 --> 01:02:40,123
to be brought on even more by
labor shortages and like, aging

1069
01:02:40,123 --> 01:02:43,447
demographics causing labor
shortages. Yeah. Where things

1070
01:02:43,447 --> 01:02:46,000
are going to need robots to
continue. Yeah.

1071
01:02:46,000 --> 01:02:49,755
Barrett Ames: Yeah. Construction
for sure has that problem,

1072
01:02:49,755 --> 01:02:53,700
right. Where like the. Oh, yeah.
The crazy the crazy anecdotes

1073
01:02:53,700 --> 01:02:57,327
that I have. And data for
construction here are just like

1074
01:02:57,327 --> 01:03:00,000
40% of contractors are over the
age of 50.

1075
01:03:00,000 --> 01:03:02,432
Audrow Nash: I know, right? Oh,
my God. And new ones are not

1076
01:03:02,432 --> 01:03:03,000
being trained.

1077
01:03:03,000 --> 01:03:04,000
Barrett Ames: Right. So it is.

1078
01:03:04,000 --> 01:03:07,056
Audrow Nash: Everyone went to
college and didn't want to go to

1079
01:03:07,056 --> 01:03:09,965
blue collar job, right? Which
now we need so many more blue

1080
01:03:09,965 --> 01:03:11,000
collar people, right?

1081
01:03:11,000 --> 01:03:14,594
Barrett Ames: For every day in
North Carolina, there are three

1082
01:03:14,594 --> 01:03:17,319
contractors who can do
electrical work that are

1083
01:03:17,319 --> 01:03:19,000
licensed under the age of 30.

1084
01:03:19,000 --> 01:03:20,000
Audrow Nash: Wow. That's the
whole state. There's a.

1085
01:03:20,000 --> 01:03:25,318
Barrett Ames: Whole state. Holy
cow. Yeah. That's wild. Yeah. So

1086
01:03:25,318 --> 01:03:30,470
there there's they're there has
to be automation. Right. Like.

1087
01:03:30,470 --> 01:03:35,622
Yeah. And one way to think about
it is that automation is just

1088
01:03:35,622 --> 01:03:40,275
taking this old construction
problem and making it sexy.

1089
01:03:40,275 --> 01:03:44,928
Because if you, if you do it
with a robot, it's hot. And

1090
01:03:44,928 --> 01:03:49,000
like, doing it by hand. Well,
let's say it's not.

1091
01:03:49,000 --> 01:03:52,523
Audrow Nash: Yeah. Well, I mean,
but it lets you there's so many

1092
01:03:52,523 --> 01:03:55,771
exciting things about doing it
with robots because one, the

1093
01:03:55,771 --> 01:03:59,073
robots get better and better.
It's like you can become a bit

1094
01:03:59,073 --> 01:04:02,541
more of like a technician. Yeah.
For the robots. Right. I don't

1095
01:04:02,541 --> 01:04:06,009
know. I think the jobs will be
changing. Definitely. For all of

1096
01:04:06,009 --> 01:04:07,000
the stuff, I mean.

1097
01:04:07,000 --> 01:04:11,932
Barrett Ames: Yeah. And then,
you know, there there are labor

1098
01:04:11,932 --> 01:04:16,783
shortages on the job site as
well. Like, on a recently I saw

1099
01:04:16,783 --> 01:04:21,957
somebody post they expected 27,
framers to show up for a job and

1100
01:04:21,957 --> 01:04:27,051
only three showed up. And like,
that's, that's part like that's

1101
01:04:27,051 --> 01:04:31,579
just the way contractors are,
right? Like they're a very

1102
01:04:31,579 --> 01:04:36,430
mercurial bunch. But then also
like the political climate is

1103
01:04:36,430 --> 01:04:40,715
also not great for the
contractor. Yeah, right. Like,

1104
01:04:40,715 --> 01:04:45,728
the, the, official numbers are
that a 30 to 40% of contractors

1105
01:04:45,728 --> 01:04:50,821
are undocumented. Our experience
is more like 90%. Holy cow. So

1106
01:04:50,821 --> 01:04:55,430
there's there is kind of this,
you know, very interesting

1107
01:04:55,430 --> 01:05:00,038
political aspect to it, that
that will, will, you know, I

1108
01:05:00,038 --> 01:05:04,000
think demand more more
automation, more robotics.

1109
01:05:04,000 --> 01:05:07,910
Audrow Nash: Yeah. That's just
what I'm saying. Yeah. Because

1110
01:05:07,910 --> 01:05:11,628
it's the solution. We can
automate ourselves into a better

1111
01:05:11,628 --> 01:05:15,538
situation. Right. I do think
there will be, I suspect, a wave

1112
01:05:15,538 --> 01:05:19,449
of blue collar workers. Yeah.
Because plumbers are making 300

1113
01:05:19,449 --> 01:05:23,423
K in the Bay area. So yeah, it's
like they're making more than

1114
01:05:23,423 --> 01:05:27,205
some software engineers. Yep.
And I think we're seeing that

1115
01:05:27,205 --> 01:05:31,244
kind of everywhere. Where are a
lot of places anyways that it's

1116
01:05:31,244 --> 01:05:34,897
like a lot of these jobs, it's
so scarce, that you have a

1117
01:05:34,897 --> 01:05:38,936
competent person that they can
command a wonderful wage. And so

1118
01:05:38,936 --> 01:05:43,038
that's driving people back into
it. Which will start to correct,

1119
01:05:43,038 --> 01:05:46,436
I expect, in the US. Yeah.
Specifically because we're

1120
01:05:46,436 --> 01:05:49,000
relatively young
demographically. Right.

1121
01:05:49,000 --> 01:05:52,985
Barrett Ames: Right, right. And
I think still will be tough. I

1122
01:05:52,985 --> 01:05:57,099
think you're right there. It'll
be a little bit slower in bigger

1123
01:05:57,099 --> 01:06:00,955
cities because they tend to be
more unionized and have a lot

1124
01:06:00,955 --> 01:06:04,876
more requirements. You know, you
gotta apprentice for 7 to 10

1125
01:06:04,876 --> 01:06:08,604
years or something like that,
right? Yeah. And so that'll,

1126
01:06:08,604 --> 01:06:12,653
that'll slow things down. But on
the flip side, unions are also

1127
01:06:12,653 --> 01:06:16,574
very good at adopting technology
quickly. Because, they, they

1128
01:06:16,574 --> 01:06:19,916
have to be able to like,
demonstrate, like, hey, are

1129
01:06:19,916 --> 01:06:23,709
higher wages, provide more
value, right. And so we actually

1130
01:06:23,709 --> 01:06:27,372
we're really concerned about
unions at first, but what we

1131
01:06:27,372 --> 01:06:31,293
found is that they're actually,
pretty interested in adopting

1132
01:06:31,293 --> 01:06:32,000
technology.

1133
01:06:32,000 --> 01:06:35,759
Audrow Nash: That's cool. Yeah.
One of the companies that I

1134
01:06:35,759 --> 01:06:39,582
talked with a while ago was
canvas, and they're doing robots

1135
01:06:39,582 --> 01:06:43,341
for drywall. Yep. And they work
very closely with unions in

1136
01:06:43,341 --> 01:06:47,036
California. Yeah. Last I talked
to them, and that was very

1137
01:06:47,036 --> 01:06:50,922
interesting to hear. Are you
guys doing any similar work with

1138
01:06:50,922 --> 01:06:54,554
unions or how are you guys
working with or thinking about

1139
01:06:54,554 --> 01:06:55,000
unions?

1140
01:06:55,000 --> 01:06:58,149
Barrett Ames: Yeah, unions
aren't a big thing. In the

1141
01:06:58,149 --> 01:07:01,000
southeast in general. So not
terribly important.

1142
01:07:01,000 --> 01:07:03,182
Audrow Nash: And you're in North
Carolina. Yeah, yeah. As you're

1143
01:07:03,182 --> 01:07:04,000
saying. Durham. Yep yep.

1144
01:07:04,000 --> 01:07:08,571
Barrett Ames: Yep. So we're
we're not terribly concerned.

1145
01:07:08,571 --> 01:07:13,383
Like we they just don't exist
here. In, in other places, you

1146
01:07:13,383 --> 01:07:18,436
know, it's definitely going to
have to be a conversation, but I

1147
01:07:18,436 --> 01:07:23,088
think, like Kevin from canvas,
has demonstrated, like, the

1148
01:07:23,088 --> 01:07:27,900
right way to approach, unions
and automation and just, like,

1149
01:07:27,900 --> 01:07:32,632
show that, like, yeah, this,
this, this is a tool. And just

1150
01:07:32,632 --> 01:07:36,000
like all tools, craftsmen should
use them.

1151
01:07:36,000 --> 01:07:40,006
Audrow Nash: Yeah. Totally.
Especially, like for canvas with

1152
01:07:40,006 --> 01:07:43,812
drywall work. One the labor
shortages are intense. There.

1153
01:07:43,812 --> 01:07:47,819
And to the people who do it get
like unbelievable repetitive

1154
01:07:47,819 --> 01:07:52,025
stress shoulder injuries because
they're just holding a sander.

1155
01:07:52,025 --> 01:07:56,299
Yep. And it's like finishing the
wall and it's just super, super

1156
01:07:56,299 --> 01:08:00,639
difficult on their body. So they
retire quickly from that kind of

1157
01:08:00,639 --> 01:08:04,779
job. Yeah. So that makes sense
for the union to really want to

1158
01:08:04,779 --> 01:08:08,986
help them. And I would bet that
it's similar for a lot of jobs,

1159
01:08:08,986 --> 01:08:12,925
like you were saying at the
beginning that, the injury rate

1160
01:08:12,925 --> 01:08:16,798
is quite high or accident rate
because it's like sores and

1161
01:08:16,798 --> 01:08:20,804
stepping on two by fours, one
floor up or more, this kind of

1162
01:08:20,804 --> 01:08:25,078
thing. Yeah. So I would bet that
this would be a big benefit. So

1163
01:08:25,078 --> 01:08:28,884
you guys have 27 homes. You're
going for your series A so

1164
01:08:28,884 --> 01:08:33,091
you're looking at scaling more.
Yep. Well what do you think the

1165
01:08:33,091 --> 01:08:37,297
future is going to look like for
you guys. Like kind of draw it

1166
01:08:37,297 --> 01:08:41,504
out for me because I think like
my impression is the last, like

1167
01:08:41,504 --> 01:08:45,176
getting to 27 homes now or
getting to 27 homes has been

1168
01:08:45,176 --> 01:08:49,517
kind of like a proof of concept.
You've gotten it working. You're

1169
01:08:49,517 --> 01:08:53,456
now making efficiency gains.
Tell me how we go forward from

1170
01:08:53,456 --> 01:08:57,730
here. And what do you think the
timing and what do you think the

1171
01:08:57,730 --> 01:09:00,000
bottlenecks are? Yeah.
Everything.

1172
01:09:00,000 --> 01:09:05,951
Barrett Ames: Yeah, yeah. So
forward, is is kind of along I

1173
01:09:05,951 --> 01:09:11,095
think three fronts. First is,
continuing to push on

1174
01:09:11,095 --> 01:09:17,248
efficiency, of robots. Second is
adding more value to panels.

1175
01:09:17,248 --> 01:09:22,997
Right. So if you if you think
about the what, what can be

1176
01:09:22,997 --> 01:09:29,049
prefab, the biggest piece of
that pie, is what we're already

1177
01:09:29,049 --> 01:09:35,000
doing. That's about 20% of the
cost of the home, right. It.

1178
01:09:35,000 --> 01:09:38,663
Audrow Nash: Is that factoring
and labor? Yeah. 20%. Yep. Okay,

1179
01:09:38,663 --> 01:09:42,151
so if you remove labor, if you
just look at. Oh, but I guess

1180
01:09:42,151 --> 01:09:45,756
okay, that's the fair comparison
for this okay. So that's 20%.

1181
01:09:45,756 --> 01:09:49,360
Yeah I guess finishing takes a
lot of it. Like the finished it

1182
01:09:49,360 --> 01:09:50,000
parts or I.

1183
01:09:50,000 --> 01:09:55,498
Barrett Ames: It's I don't know.
So like a roughly 20% of it is

1184
01:09:55,498 --> 01:10:00,996
lumber. Sorry framing. And then
the next two biggest chunks are

1185
01:10:00,996 --> 01:10:05,622
actually plumbing and
electrical. And yeah. And so we

1186
01:10:05,622 --> 01:10:10,771
have, we have the ability to
help them a lot there. In that

1187
01:10:10,771 --> 01:10:14,000
we can rule, plumbing and
electrical.

1188
01:10:14,000 --> 01:10:16,000
Audrow Nash: Oh, I like that a
lot.

1189
01:10:16,000 --> 01:10:19,448
Barrett Ames: We can save a lot
on on copper. Copper is

1190
01:10:19,448 --> 01:10:23,146
expensive. And then actually
plumbing in electrical are the

1191
01:10:23,146 --> 01:10:26,657
number one causes for framing
failures, when it comes to

1192
01:10:26,657 --> 01:10:30,731
inspection, because when you're,
when you're if you're drilling a

1193
01:10:30,731 --> 01:10:34,493
hole through two by four to run
a, pipe, the hole saw, if it

1194
01:10:34,493 --> 01:10:37,000
walks a little bit one way or
the other,

1195
01:10:37,000 --> 01:10:39,000
Audrow Nash: Kinda compromises
the integrity of the structure.

1196
01:10:39,000 --> 01:10:44,120
Barrett Ames: Right. Right. And
so being able to a root it and B

1197
01:10:44,120 --> 01:10:49,000
cut the holes very precisely
makes their job easier. And then

1198
01:10:49,000 --> 01:10:54,120
you know, there I think there is
kind of an interesting, I don't

1199
01:10:54,120 --> 01:10:59,080
know, 18 to 24 month problem of
like do we start to wire it in

1200
01:10:59,080 --> 01:11:01,000
in the warehouse. Right.

1201
01:11:01,000 --> 01:11:02,000
Audrow Nash: Like that would be
so great.

1202
01:11:02,000 --> 01:11:05,657
Barrett Ames: And there there's,
there's some interesting things

1203
01:11:05,657 --> 01:11:06,000
there.

1204
01:11:06,000 --> 01:11:08,791
Audrow Nash: Sounds really hard.
I imagine a lot of hard problems

1205
01:11:08,791 --> 01:11:11,067
around it because are
deformable, right? Materials. I

1206
01:11:11,067 --> 01:11:13,000
know they're pretty rigid.
Right. But, still.

1207
01:11:13,000 --> 01:11:16,508
Barrett Ames: Right, right. And
then that motion planning

1208
01:11:16,508 --> 01:11:17,000
problem.

1209
01:11:17,000 --> 01:11:18,000
Audrow Nash: Right. That'd be
much harder, I bet.

1210
01:11:18,000 --> 01:11:22,685
Barrett Ames: Yeah. Right.
Right, right. And then, the kind

1211
01:11:22,685 --> 01:11:27,608
of like third area that I think
is really important, for us is

1212
01:11:27,608 --> 01:11:31,896
we'll be pushing a lot on
learning from demonstration,

1213
01:11:31,896 --> 01:11:36,819
because, as you expand into more
and more of the house, things

1214
01:11:36,819 --> 01:11:41,663
get, more fabric, like, or more
squishy, and there's a ton of

1215
01:11:41,663 --> 01:11:46,268
experience locked in craftsmen.
That if we can get them to

1216
01:11:46,268 --> 01:11:50,000
demonstrate it, we can get the
robots to do it.

1217
01:11:50,000 --> 01:11:53,038
Audrow Nash: Oh, that's really
cool. That's a nice long term

1218
01:11:53,038 --> 01:11:54,000
perspective, right?

1219
01:11:54,000 --> 01:11:58,283
Barrett Ames: Right. So, like, a
really early, thing for us to

1220
01:11:58,283 --> 01:12:02,082
tackle is, is actually
insulation, right? Insulation is

1221
01:12:02,082 --> 01:12:06,020
this kind of fabric, right? When
it's when it's made from

1222
01:12:06,020 --> 01:12:10,303
fiberglass. It it feels kind of
just like a big blanket, being

1223
01:12:10,303 --> 01:12:14,516
able to put that in, it is not a
straightforward, like just a

1224
01:12:14,516 --> 01:12:15,000
motion.

1225
01:12:15,000 --> 01:12:18,913
Audrow Nash: Planning because
it's, it's a very difficult

1226
01:12:18,913 --> 01:12:23,376
compressible surface. Right. You
have to do it just right and not

1227
01:12:23,376 --> 01:12:27,290
have it overlap and stuff and
whatever else. Right. Okay.

1228
01:12:27,290 --> 01:12:31,478
Yeah. A lot of cool a lot of
very cool, problems and a lot of

1229
01:12:31,478 --> 01:12:35,528
high potential use cases that
sound like they leverage what

1230
01:12:35,528 --> 01:12:38,000
you already have pretty. Well.
Yeah.

1231
01:12:38,000 --> 01:12:41,902
Barrett Ames: Yeah, yeah. I
think one of, one of the, the

1232
01:12:41,902 --> 01:12:46,077
key things for us is that by
tackling framing first and doing

1233
01:12:46,077 --> 01:12:49,842
that in a warehouse, we really
set ourselves up because

1234
01:12:49,842 --> 01:12:54,018
everything else builds off of
framing. Yeah. Right. And so we

1235
01:12:54,018 --> 01:12:58,262
so we can leave ourselves kind
of crooked cookie crumbs where,

1236
01:12:58,262 --> 01:13:01,000
where necessary to help future
problems.

1237
01:13:01,000 --> 01:13:04,164
Audrow Nash: Yeah. Super cool I
would say like the super low

1238
01:13:04,164 --> 01:13:07,327
hanging fruit to me that adds a
lot of value is just cutting

1239
01:13:07,327 --> 01:13:10,702
those holes for pipes and wires.
Yeah. Right. As you're doing it

1240
01:13:10,702 --> 01:13:13,813
that's like I can't imagine
that's that much harder because

1241
01:13:13,813 --> 01:13:16,924
you could just drill it while
you're, you grab the piece of

1242
01:13:16,924 --> 01:13:20,087
wood you want, you drill the
hole through it. Yeah. And then

1243
01:13:20,087 --> 01:13:23,409
you go place it where you want
that holes there now. Yeah. That

1244
01:13:23,409 --> 01:13:26,678
that to me seems super, super
exciting. And then all the other

1245
01:13:26,678 --> 01:13:30,000
things are very, very cool. As
you guys grow as a company. Yep.

1246
01:13:30,000 --> 01:13:34,115
Barrett Ames: Yep. Yeah. Though
the holes are like our number

1247
01:13:34,115 --> 01:13:38,365
one requested feature right now.
We don't get a lot of requests

1248
01:13:38,365 --> 01:13:42,008
from from, the job site. They
tend to usually just say

1249
01:13:42,008 --> 01:13:46,123
everything's great. But that
that's one, that's one that that

1250
01:13:46,123 --> 01:13:47,000
we've gotten.

1251
01:13:47,000 --> 01:13:51,189
Audrow Nash: Oh, yeah. Okay.
Very cool. So tell me a bit more

1252
01:13:51,189 --> 01:13:55,584
about the construction space in
general. Because a lot of people

1253
01:13:55,584 --> 01:13:59,910
I did a poll on X a while ago,
and like, it was bonkers. It was

1254
01:13:59,910 --> 01:14:04,099
like 90% of people that are in
the audience are interested in

1255
01:14:04,099 --> 01:14:07,944
starting a robotics company at
some point, and like, 25%

1256
01:14:07,944 --> 01:14:12,064
already had or something like
bonkers. Yep. So tell me about

1257
01:14:12,064 --> 01:14:16,459
the construction space. From the
perspective of an entrepreneur,

1258
01:14:16,459 --> 01:14:20,923
what are like low hanging fruit?
Lots of opportunity. Yeah, yeah.

1259
01:14:20,923 --> 01:14:24,974
How do you think of it? Yeah,
it's an exciting space. And I

1260
01:14:24,974 --> 01:14:29,094
don't know too many robotics
startups that are going in. And

1261
01:14:29,094 --> 01:14:33,077
I think there's a lot of low
hanging fruit. Probably. Yes,

1262
01:14:33,077 --> 01:14:35,000
but what do you think? Yeah.

1263
01:14:35,000 --> 01:14:39,354
Barrett Ames: The way that I
think about it is threefold. And

1264
01:14:39,354 --> 01:14:43,707
it, it really ties into like,
well, why is it awesome to work

1265
01:14:43,707 --> 01:14:48,061
at BotBuilt every day. A that
for, for me as a CTO there are,

1266
01:14:48,061 --> 01:14:52,272
there are there's an endless
number of problems right. Like

1267
01:14:52,272 --> 01:14:56,340
and for me that's really
exciting. For some people that's

1268
01:14:56,340 --> 01:15:00,623
overwhelming. Super fun. But
like, you know, I'll solve this

1269
01:15:00,623 --> 01:15:04,976
problem there will be another
one. Yeah. And it's because for

1270
01:15:04,976 --> 01:15:09,544
the last 70 or 80 years, there's
been basically no innovation in

1271
01:15:09,544 --> 01:15:13,541
construction. So you're 100%
right. There's a ton of low

1272
01:15:13,541 --> 01:15:17,395
hanging fruit. Where where
strides have been made that

1273
01:15:17,395 --> 01:15:21,678
they're typically made in high
end commercial where when you

1274
01:15:21,678 --> 01:15:25,746
fail, failure means losses in
the hundreds of millions of

1275
01:15:25,746 --> 01:15:30,171
dollars. We haven't seen a lot
of innovation in homes. Because

1276
01:15:30,171 --> 01:15:34,382
if your home goes up in, you
know, ten, $15,000 in price or

1277
01:15:34,382 --> 01:15:38,522
whatever your mortgage goes up
by, you know, ten, 15 to 50

1278
01:15:38,522 --> 01:15:42,875
bucks a month or something like
that. Right? And so you don't

1279
01:15:42,875 --> 01:15:47,443
really like the kind of American
consumer has been boiled like a

1280
01:15:47,443 --> 01:15:51,797
frog, as the prices keep going
up, because there's not really

1281
01:15:51,797 --> 01:15:56,293
anything pushing it down. So the
result is that the US is short

1282
01:15:56,293 --> 01:16:00,433
homes, about tons, four and a
half to 5 million homes. And

1283
01:16:00,433 --> 01:16:04,787
we're not building them fast
enough to catch up. That problem

1284
01:16:04,787 --> 01:16:06,000
is getting worse.

1285
01:16:06,000 --> 01:16:09,638
Audrow Nash: I wonder, I wonder
how you work with, baby boomers

1286
01:16:09,638 --> 01:16:13,103
moving into retirement. Yeah.
For everything. And if they'll

1287
01:16:13,103 --> 01:16:16,799
move into, like, condos and free
up their homes or pass away and

1288
01:16:16,799 --> 01:16:20,263
free up their homes or things
like this. But yeah, there's a

1289
01:16:20,263 --> 01:16:23,613
huge home shortage and so many
people are just moving into

1290
01:16:23,613 --> 01:16:27,020
apartments. Right. Which is
really interesting. And like if

1291
01:16:27,020 --> 01:16:30,658
you live in a many big cities,
it's just absolutely prohibitive

1292
01:16:30,658 --> 01:16:34,238
to buy a home, right, for this
kind of thing. So you're taking

1293
01:16:34,238 --> 01:16:37,761
a bite out of that huge problem?
Well, I guess the four and a

1294
01:16:37,761 --> 01:16:41,283
half to four or 4 to 5 million.
Yeah. Is that does that count

1295
01:16:41,283 --> 01:16:44,806
people that are in apartments or
is that just how do they get

1296
01:16:44,806 --> 01:16:47,000
that number? I don't know if you
know.

1297
01:16:47,000 --> 01:16:50,975
Barrett Ames: Yeah, I there's
like 3 or 4 different people

1298
01:16:50,975 --> 01:16:54,882
that that work on this problem.
We know Stephen Kim who's

1299
01:16:54,882 --> 01:16:58,788
economist at Evercore, if I
remember correctly, he's just

1300
01:16:58,788 --> 01:17:02,832
down the road here in Raleigh.
Cool. And there's a bunch of

1301
01:17:02,832 --> 01:17:06,875
ways that they get at this
number. And I don't remember any

1302
01:17:06,875 --> 01:17:09,000
of them off the top of my head.

1303
01:17:09,000 --> 01:17:11,000
Audrow Nash: So that's okay.

1304
01:17:11,000 --> 01:17:12,000
Barrett Ames: But the.

1305
01:17:12,000 --> 01:17:13,000
Audrow Nash: Cool it's a huge
growth.

1306
01:17:13,000 --> 01:17:16,814
Barrett Ames: Rate. The cool
thing is that he he actually

1307
01:17:16,814 --> 01:17:20,963
releases a report every year on
the state of housing, and just

1308
01:17:20,963 --> 01:17:25,178
looks at, oh, that's also like
latent demand, all sorts of, you

1309
01:17:25,178 --> 01:17:29,260
know, various technical aspects.
He's doing it from, like, an

1310
01:17:29,260 --> 01:17:31,000
investor standpoint. Yeah.

1311
01:17:31,000 --> 01:17:34,398
Audrow Nash: And so with this
huge shortage of homes that

1312
01:17:34,398 --> 01:17:37,856
pushes up the price of all the
other homes, making it even

1313
01:17:37,856 --> 01:17:41,373
harder for people to buy homes,
right then, so you guys are

1314
01:17:41,373 --> 01:17:45,188
trying to increase supply? Yep.
Which will make it so that costs

1315
01:17:45,188 --> 01:17:49,004
come down. More people can be in
homes, right? I really like the

1316
01:17:49,004 --> 01:17:52,819
mission. I think like I've been
a homeowner now for almost three

1317
01:17:52,819 --> 01:17:56,158
years and it is glorious. I
absolutely love it. Like the

1318
01:17:56,158 --> 01:17:59,794
best thing is sitting in the
backyard, right. And everything.

1319
01:17:59,794 --> 01:18:02,000
Yeah. And but, it's a yeah. Go
ahead.

1320
01:18:02,000 --> 01:18:05,643
Barrett Ames: It's actually
pretty interesting because, I

1321
01:18:05,643 --> 01:18:09,477
bring this up because not a lot
of people believe this. More

1322
01:18:09,477 --> 01:18:13,376
supply will drop the price down.
And, and and there there's a

1323
01:18:13,376 --> 01:18:17,019
number of people who don't
believe this, but, Austin is a

1324
01:18:17,019 --> 01:18:19,000
great example of a city that I.

1325
01:18:19,000 --> 01:18:20,000
Audrow Nash: Was going to say if
you didn't.

1326
01:18:20,000 --> 01:18:25,314
Barrett Ames: Yeah, that that
has massively increased its

1327
01:18:25,314 --> 01:18:30,627
housing supply and has seen,
rents and mortgages drop, by

1328
01:18:30,627 --> 01:18:31,000
15%.

1329
01:18:31,000 --> 01:18:34,436
Audrow Nash: Totally. Yeah. We
bought a house right around that

1330
01:18:34,436 --> 01:18:37,818
time, and we're in San Antonio
and it went through San Antonio

1331
01:18:37,818 --> 01:18:40,873
is like an hour away from
Austin. We saw the same thing.

1332
01:18:40,873 --> 01:18:44,091
And then it's just amazing. It's
like market forces working

1333
01:18:44,091 --> 01:18:47,473
because you see the prices go
rise up because they can't build

1334
01:18:47,473 --> 01:18:50,418
as fast as the demand is
growing, because everyone has

1335
01:18:50,418 --> 01:18:53,582
the internet and everyone
decided to move there all at the

1336
01:18:53,582 --> 01:18:56,964
same time. Right. And then, they
build and build and build and

1337
01:18:56,964 --> 01:19:00,345
then years later, it's like the
demand or the, the, the demand

1338
01:19:00,345 --> 01:19:03,727
matches the supply. And then you
see prices come right back in

1339
01:19:03,727 --> 01:19:04,000
line.

1340
01:19:04,000 --> 01:19:05,000
Barrett Ames: And it was just
remarkable.

1341
01:19:05,000 --> 01:19:07,754
Audrow Nash: Yeah. Like I oh
wonderful I absolutely love

1342
01:19:07,754 --> 01:19:08,000
that.

1343
01:19:08,000 --> 01:19:11,643
Barrett Ames: Yeah. So so
there's, there's, you know big

1344
01:19:11,643 --> 01:19:15,416
technical problems. There's a
big societal problem. Right.

1345
01:19:15,416 --> 01:19:19,254
Like if we don't solve the
housing problem there's going to

1346
01:19:19,254 --> 01:19:23,287
be a lot of unrest. Right. Like
people that the American dream

1347
01:19:23,287 --> 01:19:27,321
kind of really, really affects
how a lot of people think about

1348
01:19:27,321 --> 01:19:30,769
whether or not they're
successful in life. Right. And

1349
01:19:30,769 --> 01:19:34,542
so you could have this big
societal unrest if, if you have

1350
01:19:34,542 --> 01:19:38,054
a whole generation that can't
get into get into homes,

1351
01:19:38,054 --> 01:19:41,958
totally. And then the the last
bit is it's a massive market,

1352
01:19:41,958 --> 01:19:46,056
right? Like it's $1.7 trillion
in the US. And I want to drive a

1353
01:19:46,056 --> 01:19:50,024
fancy car and have a nice house.
So you, like, you put all of

1354
01:19:50,024 --> 01:19:51,000
those together?

1355
01:19:51,000 --> 01:19:52,000
Audrow Nash: I think it's forces
aligned.

1356
01:19:52,000 --> 01:19:56,235
Barrett Ames: I think it's a
great place to build a robotics

1357
01:19:56,235 --> 01:20:00,824
company. And there's so much, so
much to do. I would love to have

1358
01:20:00,824 --> 01:20:04,000
more friends come and build and
construction.

1359
01:20:04,000 --> 01:20:07,249
Audrow Nash: Hell, yeah. Yeah,
that would be awesome. Oh. One

1360
01:20:07,249 --> 01:20:10,391
thought with the you mentioned
unrest from lack of housing.

1361
01:20:10,391 --> 01:20:13,479
Yeah. One other thing is that if
people don't have houses,

1362
01:20:13,479 --> 01:20:16,621
houses, or like space, it's
harder to have kids because you

1363
01:20:16,621 --> 01:20:19,763
guys are all on top of each
other. And, like, we need a big

1364
01:20:19,763 --> 01:20:22,746
replacement generation coming
up. Yeah, for this kind of

1365
01:20:22,746 --> 01:20:25,994
thing. Yeah. To just keep life
going and make it so we have a

1366
01:20:25,994 --> 01:20:28,923
good, strong consumption
economy, right? Yeah. 30 years

1367
01:20:28,923 --> 01:20:32,172
from now. Yeah. Or this kind of
thing. Yeah. Because a lot of

1368
01:20:32,172 --> 01:20:35,260
advanced societies, you see,
they just move, industrialize

1369
01:20:35,260 --> 01:20:38,562
and then they stop having kids
and the birthrate plummets like

1370
01:20:38,562 --> 01:20:40,000
a lot of Europe. Right? So.

1371
01:20:40,000 --> 01:20:43,667
Barrett Ames: Right. Right.
Yeah. Yeah. When we when we

1372
01:20:43,667 --> 01:20:47,600
moved to Durham, we, we had, a
tiny little two bedroom, two

1373
01:20:47,600 --> 01:20:51,533
bath, got four kids. Now, I now
have a four bed, four bath.

1374
01:20:51,533 --> 01:20:55,000
Definitely need lots of space
for kids. Can confirm.

1375
01:20:55,000 --> 01:20:59,295
Audrow Nash: Oh, totally can
confirm. Yeah, yeah. That's

1376
01:20:59,295 --> 01:21:03,820
great. Let's see. So tell me
about North Carolina because I

1377
01:21:03,820 --> 01:21:08,652
don't know too many. Like. So I
lived in Charlotte for a while.

1378
01:21:08,652 --> 01:21:13,407
I really loved it. But tell me a
bit about Charlotte or, North

1379
01:21:13,407 --> 01:21:18,392
Carolina as a place for you guys
as a robotics company. Yeah. And

1380
01:21:18,392 --> 01:21:21,000
being in the southeast in
general.

1381
01:21:21,000 --> 01:21:25,345
Barrett Ames: Yeah. The as a
robotics company, where we're

1382
01:21:25,345 --> 01:21:29,615
located is actually, I think
deeply underappreciated. The

1383
01:21:29,615 --> 01:21:33,585
they call it the Research
Triangle. Right. Basically,

1384
01:21:33,585 --> 01:21:37,706
Durham is Chapel Hill and
Raleigh, in that triangle, we

1385
01:21:37,706 --> 01:21:42,125
actually have the second highest
number of computer science

1386
01:21:42,125 --> 01:21:46,620
pitches. We're only be by the
Bay area. Oh. And like, people

1387
01:21:46,620 --> 01:21:51,040
are like, no, that can't be
true. That is that. But RedHat,

1388
01:21:51,040 --> 01:21:55,834
IBM, Cisco, Lenovo, Epic Games,
there's all that's a whole bunch

1389
01:21:55,834 --> 01:22:00,179
of smaller companies, that have
spun out of that. And then

1390
01:22:00,179 --> 01:22:04,000
there's, there's a bunch of
biotech. And so we see.

1391
01:22:04,000 --> 01:22:05,810
Audrow Nash: Oh, yeah, that's
what the area is known for,

1392
01:22:05,810 --> 01:22:06,000
right.

1393
01:22:06,000 --> 01:22:10,450
Barrett Ames: For sure. Right.
But we also see a lot of

1394
01:22:10,450 --> 01:22:15,466
surgical robots because of that
as well. And then you add Duke

1395
01:22:15,466 --> 01:22:20,563
and NC state. I have wonderful,
engineering programs. I went to

1396
01:22:20,563 --> 01:22:25,498
Duke for my PhD, so, like, I
guess I'll begrudgingly say uncW

1397
01:22:25,498 --> 01:22:30,191
is also a good school. We
actually just made our first UMC

1398
01:22:30,191 --> 01:22:34,641
hire, and he is absolutely
crushing it. So. Hell, yeah.

1399
01:22:34,641 --> 01:22:39,252
It's it's it's great to, to be
in kind of this, this area

1400
01:22:39,252 --> 01:22:43,864
where, I can have, a big house,
lots of kids, and start a

1401
01:22:43,864 --> 01:22:48,233
robotics company. And, you know,
like, that's all very

1402
01:22:48,233 --> 01:22:52,845
affordable, right? Like it's
not, I'm not I'm not getting

1403
01:22:52,845 --> 01:22:56,000
crushed by rent in San
Francisco. Yeah.

1404
01:22:56,000 --> 01:22:59,502
Audrow Nash: Yeah. It's crazy.
My my house when we bought in,

1405
01:22:59,502 --> 01:23:02,601
like, the housing peak when
everyone was doing it, the

1406
01:23:02,601 --> 01:23:06,160
mortgage on my house is the same
as my studio apartment in San

1407
01:23:06,160 --> 01:23:09,432
Francisco. Yeah, just before
moving here. Right. And that

1408
01:23:09,432 --> 01:23:12,876
price has increased since that
was, like, Covid discount for

1409
01:23:12,876 --> 01:23:15,000
this kind of thing. Yeah. So
bonkers.

1410
01:23:15,000 --> 01:23:18,782
Barrett Ames: Right. And then
the the southeast as a whole.

1411
01:23:18,782 --> 01:23:22,500
That's right. Like the Sunbelt,
is growing and everybody's

1412
01:23:22,500 --> 01:23:26,218
building houses down here. So
it's a wonderful place for a

1413
01:23:26,218 --> 01:23:29,872
construction robotics company.
And then there's just this

1414
01:23:29,872 --> 01:23:34,038
really weird happenstance, where
Durham, where we're located, has

1415
01:23:34,038 --> 01:23:37,821
become kind of a construction
tech hub. There are two other

1416
01:23:37,821 --> 01:23:41,410
two other construction tech
companies, both doing really

1417
01:23:41,410 --> 01:23:45,064
cool things. Planted, is down
the road. They're building,

1418
01:23:45,064 --> 01:23:49,167
plywood sheets, from grass. And
the cool thing there is that it?

1419
01:23:49,167 --> 01:23:50,000
It's, a very.

1420
01:23:50,000 --> 01:23:51,000
Audrow Nash: Don't know what
that means at all.

1421
01:23:51,000 --> 01:23:55,298
Barrett Ames: It's like a very
particular type of grass, but

1422
01:23:55,298 --> 01:23:59,955
what it what it boils down to is
it's not as, susceptible to mold

1423
01:23:59,955 --> 01:24:04,181
growth, that it's a it's a
couple of space guys that rolled

1424
01:24:04,181 --> 01:24:08,265
out of space and decided to
build up, a create a greener,

1425
01:24:08,265 --> 01:24:12,205
safer, cheaper, building
material. And then hammock is,

1426
01:24:12,205 --> 01:24:16,360
is also down the road and
they're doing like, they want to

1427
01:24:16,360 --> 01:24:20,443
make it. So that way you can
design your, your dream home

1428
01:24:20,443 --> 01:24:24,813
from your house. Right. Like,
it's just, building out kind of

1429
01:24:24,813 --> 01:24:29,254
the video game aspect of, like,
designing your house. So cool.

1430
01:24:29,254 --> 01:24:33,481
The three of us, you know, we
it's it's been really cool to

1431
01:24:33,481 --> 01:24:37,636
just be surrounded by other
people who are. We're thinking

1432
01:24:37,636 --> 01:24:40,000
about how to change
construction.

1433
01:24:40,000 --> 01:24:44,491
Audrow Nash: Hell, yeah. And
also makers and doing the good

1434
01:24:44,491 --> 01:24:49,135
problems and everything. Yeah.
That's great. Yep. Hell, yeah.

1435
01:24:49,135 --> 01:24:53,779
What do you. So if you had to
start over or, say, advice to a

1436
01:24:53,779 --> 01:24:58,118
young founder, what do you
think? Like what would be some

1437
01:24:58,118 --> 01:25:02,000
really good things to
communicate to a younger you.

1438
01:25:02,000 --> 01:25:06,486
Barrett Ames: Yeah. Yeah.
Younger me or earlier. Yeah.

1439
01:25:06,486 --> 01:25:11,637
Yeah, yeah, yeah. I think, like,
we came to the, teleoperation

1440
01:25:11,637 --> 01:25:16,455
driven development, part way
iteratively. Yeah. And I like

1441
01:25:16,455 --> 01:25:21,688
that would be the biggest thing
is, is start with that as your,

1442
01:25:21,688 --> 01:25:26,756
your formative paradigm. And
then, honestly, like, just don't

1443
01:25:26,756 --> 01:25:31,740
talk to SAS investors. You're
just wasting your time, right?

1444
01:25:31,740 --> 01:25:36,559
Like, a lot of them. A lot of
them will totally. Just like

1445
01:25:36,559 --> 01:25:41,294
they'll say they're cool and
they're like, oh yeah, we do

1446
01:25:41,294 --> 01:25:46,528
hard tech, right? Like that's a
cool thing to say. But like, if

1447
01:25:46,528 --> 01:25:51,097
they haven't made robotics
investment, like, just don't

1448
01:25:51,097 --> 01:25:55,998
expect them to, right. And then,
I would, I would also like

1449
01:25:55,998 --> 01:26:01,065
update. KVK right. So like we
have a lot of, we have a lot of

1450
01:26:01,065 --> 01:26:06,216
strategic that have helped us.
What is CV. Oh sorry. What is a

1451
01:26:06,216 --> 01:26:11,200
corporate venture capital? Oh
yeah. Totally. And at least in

1452
01:26:11,200 --> 01:26:15,520
construction, it's been it's
been massive for really

1453
01:26:15,520 --> 01:26:20,837
understanding more of the nuance
of, of all the, the particulars

1454
01:26:20,837 --> 01:26:22,000
in this field.

1455
01:26:22,000 --> 01:26:24,685
Audrow Nash: Totally. And you're
finding good partners and

1456
01:26:24,685 --> 01:26:27,000
they're opening up a lot of
doors, I expect. Yeah.

1457
01:26:27,000 --> 01:26:31,377
Barrett Ames: Yeah. So that
those are kind of the, the big

1458
01:26:31,377 --> 01:26:36,057
areas that I would think about,
it's, and, and the last thing,

1459
01:26:36,057 --> 01:26:40,509
the last thing I'll say, founder
market fit is ridiculously

1460
01:26:40,509 --> 01:26:45,113
important. And maybe, maybe it's
just, under underappreciated

1461
01:26:45,113 --> 01:26:49,717
from my perspective. But I don't
hear a lot of people talking

1462
01:26:49,717 --> 01:26:51,000
about it anymore.

1463
01:26:51,000 --> 01:26:52,629
Audrow Nash: What is it? Yeah.
What do you mean, cylinder

1464
01:26:52,629 --> 01:26:53,000
market? Yeah.

1465
01:26:53,000 --> 01:26:55,000
Barrett Ames: So, like, if it's
like.

1466
01:26:55,000 --> 01:26:56,000
Audrow Nash: Finding markets,
it.

1467
01:26:56,000 --> 01:27:00,692
Barrett Ames: No no no no no.
Like like, you as a founder, do

1468
01:27:00,692 --> 01:27:05,000
you fit with the market that
you're trying to sell into?

1469
01:27:05,000 --> 01:27:07,844
Audrow Nash: And so I, I don't
hear this about ever basically.

1470
01:27:07,844 --> 01:27:10,000
Yeah. So that that's a great
point. Okay. Yeah.

1471
01:27:10,000 --> 01:27:14,870
Barrett Ames: Tell me more.
Yeah. So like the, the idea but

1472
01:27:14,870 --> 01:27:19,575
one, one way that has been
described by outsiders is that

1473
01:27:19,575 --> 01:27:24,776
we're white collar workers, with
a blue collar attitude. Right.

1474
01:27:24,776 --> 01:27:29,729
And, and so that lets us engage
in construction. Right. Like

1475
01:27:29,729 --> 01:27:34,599
this is probably the longest
I've gone without cursing, for

1476
01:27:34,599 --> 01:27:39,222
several days. Right. And you
didn't have to do that. All

1477
01:27:39,222 --> 01:27:44,175
right, I know, I know, but,
like, I don't know, the kids, my

1478
01:27:44,175 --> 01:27:45,000
watch. So.

1479
01:27:45,000 --> 01:27:47,000
Audrow Nash: Yeah. So funny.

1480
01:27:47,000 --> 01:27:50,740
Barrett Ames: But, like, you
know, there's there's

1481
01:27:50,740 --> 01:27:55,526
superficial stuff like that, but
there's also like, like, do you

1482
01:27:55,526 --> 01:27:59,341
share values with your
customers? Right. And if you

1483
01:27:59,341 --> 01:28:04,127
don't, maybe you don't want them
as customers. Right? Because it

1484
01:28:04,127 --> 01:28:08,241
can be really hard for you to
have any kind of customer

1485
01:28:08,241 --> 01:28:12,803
empathy. If you're not like if
there's not some kind of value

1486
01:28:12,803 --> 01:28:14,000
alignment there.

1487
01:28:14,000 --> 01:28:17,698
Audrow Nash: How would you know
about that? I guess looking at

1488
01:28:17,698 --> 01:28:21,097
your values. But I do think it's
also very easy to delude

1489
01:28:21,097 --> 01:28:24,497
yourself into something that
seems to be lucrative, where

1490
01:28:24,497 --> 01:28:28,016
you're like, oh, I could be
interested in that, or I'm sure

1491
01:28:28,016 --> 01:28:31,237
I'm interested in this
altruistic purpose. Even though

1492
01:28:31,237 --> 01:28:34,696
it doesn't align with who you
are and what you're actually

1493
01:28:34,696 --> 01:28:38,155
interested in. Like, how would
you find that? Because that

1494
01:28:38,155 --> 01:28:41,078
seems like, a good
introspective. Yeah. Challenge

1495
01:28:41,078 --> 01:28:44,000
that, I think a lot of people
will struggle with.

1496
01:28:44,000 --> 01:28:50,843
Barrett Ames: Yeah, that's a
good question. I mean, I it's

1497
01:28:50,843 --> 01:28:57,803
just what I gravitate towards,
right. And so, like, I enjoy

1498
01:28:57,803 --> 01:29:05,000
hanging out with construction
people. So, that's kind of how.

1499
01:29:05,000 --> 01:29:06,905
Audrow Nash: How I the simple
answer that I think is kind of

1500
01:29:06,905 --> 01:29:07,000
it.

1501
01:29:07,000 --> 01:29:12,154
Barrett Ames: Right. Yeah. Yeah.
Like, you know, it's it's just,

1502
01:29:12,154 --> 01:29:17,309
Yeah. There, there, there's just
something about the, the market

1503
01:29:17,309 --> 01:29:21,577
that you want to go into that
you find interesting or

1504
01:29:21,577 --> 01:29:26,651
fascinating, because at the end
of the day, like, it's going to

1505
01:29:26,651 --> 01:29:31,644
be deeply frustrating, right?
Like, totally there there are so

1506
01:29:31,644 --> 01:29:36,718
many things that are going to go
wrong and so many people that,

1507
01:29:36,718 --> 01:29:41,792
that, you know, that, that I'll
butt heads with. Brant, our CEO

1508
01:29:41,792 --> 01:29:47,027
is amazing at being friends with
everyone. And so like, for, as a

1509
01:29:47,027 --> 01:29:52,020
founder, like, you just have to,
like, what's going to get you

1510
01:29:52,020 --> 01:29:55,000
through whatever that
frustration is.

1511
01:29:55,000 --> 01:29:58,384
Audrow Nash: Okay. I like that a
lot. I have not heard about

1512
01:29:58,384 --> 01:30:01,542
founder Market Fit. I think
that's a really good idea to

1513
01:30:01,542 --> 01:30:04,983
pick customer or pick some group
to work with that you really

1514
01:30:04,983 --> 01:30:08,423
like anyways. Because yeah, I
say you, you hate the values of

1515
01:30:08,423 --> 01:30:11,751
one group like you find it's
silly or something, right. And

1516
01:30:11,751 --> 01:30:15,247
go and try to work with them.
And now it's like your startup's

1517
01:30:15,247 --> 01:30:18,857
going to take ten years. And so
now you're interfacing with this

1518
01:30:18,857 --> 01:30:22,128
group the whole time. It's a
really easy way to get burned

1519
01:30:22,128 --> 01:30:25,625
out, right. And not do a good
job. Whereas if you like, really

1520
01:30:25,625 --> 01:30:28,952
like the values of them, then
you're going to enjoy hanging

1521
01:30:28,952 --> 01:30:32,505
out with them. You're going to
enjoy trying to understand their

1522
01:30:32,505 --> 01:30:35,777
problems. And, you guys will
probably both be more open to

1523
01:30:35,777 --> 01:30:38,935
working with each other. Like
they'll try your thing and

1524
01:30:38,935 --> 01:30:42,150
they'll let you in on good
feedback and not sabotage your

1525
01:30:42,150 --> 01:30:45,477
device or something like this,
which might occur. Right? If

1526
01:30:45,477 --> 01:30:47,000
they really dislike. Right?

1527
01:30:47,000 --> 01:30:50,000
Barrett Ames: Yeah, 100%.

1528
01:30:50,000 --> 01:30:54,048
Audrow Nash: Hell yeah. So
looking forward, I know that

1529
01:30:54,048 --> 01:30:58,390
it's hard. And you pointed to
three things that look really

1530
01:30:58,390 --> 01:31:02,658
exciting. In the future, a
botbuilt, but like, say, 2 to 4

1531
01:31:02,658 --> 01:31:07,000
years from now, where do you
expect you guys will be? Yeah.

1532
01:31:07,000 --> 01:31:11,892
Barrett Ames: We'll be on site
100% like. All right. Oh, yeah.

1533
01:31:11,892 --> 01:31:16,546
I, I think for us to capture
ever more value and to provide

1534
01:31:16,546 --> 01:31:21,280
ever more value, we have to get
out of the warehouse. And we

1535
01:31:21,280 --> 01:31:26,329
like very specifically chose to
begin in the warehouse because I

1536
01:31:26,329 --> 01:31:31,458
knew just enough about lumber to
know that, like, it was going to

1537
01:31:31,458 --> 01:31:36,349
be painful to learn how to do
it. And I wanted to do that in a

1538
01:31:36,349 --> 01:31:41,083
controlled environment. But to
really make an impact, to, to

1539
01:31:41,083 --> 01:31:45,580
fundamentally change, how
construction is, we need to get

1540
01:31:45,580 --> 01:31:47,000
onto the job site.

1541
01:31:47,000 --> 01:31:50,816
Audrow Nash: That's awesome.
Hell, yeah. Any thoughts with,

1542
01:31:50,816 --> 01:31:54,502
like, size of the company,
number of houses deployed? And

1543
01:31:54,502 --> 01:31:58,124
these can be like, because
obviously none of us know the

1544
01:31:58,124 --> 01:32:01,940
future. But like, your current
rate of growth, growth seems

1545
01:32:01,940 --> 01:32:05,756
amazing. And I suppose if a
fraction of that is capped over

1546
01:32:05,756 --> 01:32:09,507
the years, it will be quite
impressive where you guys get.

1547
01:32:09,507 --> 01:32:13,000
But, looking out in the future,
any ideas around that?

1548
01:32:13,000 --> 01:32:18,489
Barrett Ames: Yeah. So right now
we have, three robot teams, that

1549
01:32:18,489 --> 01:32:23,978
are producing all of our houses.
The robot team is just a pair of

1550
01:32:23,978 --> 01:32:28,369
industrial arms that work
together. I think in three

1551
01:32:28,369 --> 01:32:32,000
years, we'll have call it 150 to
200 teams.

1552
01:32:32,000 --> 01:32:34,000
Audrow Nash: Holy cow, that'd be
amazing.

1553
01:32:34,000 --> 01:32:38,276
Barrett Ames: And we'll we'll
have deployed with a whole bunch

1554
01:32:38,276 --> 01:32:42,483
of franchise partners, right?
Like, we can't we can't own the

1555
01:32:42,483 --> 01:32:46,000
US just by ourselves.
Particularly in construction,

1556
01:32:46,000 --> 01:32:50,345
where, so much of the sales is
like, well, my dad knew his dad,

1557
01:32:50,345 --> 01:32:54,138
right? But it's a very
relationship based business. But

1558
01:32:54,138 --> 01:32:58,552
we can make them more efficient,
and let them focus on what they

1559
01:32:58,552 --> 01:33:02,828
like doing, which is sell me.
And so, we're actually doing our

1560
01:33:02,828 --> 01:33:06,000
first deployment, this this
summer to Florida.

1561
01:33:06,000 --> 01:33:08,000
Audrow Nash: Oh, so out of North
Carolina, you're saying? Yeah.

1562
01:33:08,000 --> 01:33:13,000
Barrett Ames: So we'll have a
bunch of teams set up, down in

1563
01:33:13,000 --> 01:33:18,167
Florida. And that'll that'll be
kind of, the beginning of this

1564
01:33:18,167 --> 01:33:20,000
franchise opportunity.

1565
01:33:20,000 --> 01:33:23,268
Audrow Nash: Hell, yeah. And the
start of, like, remote scaling,

1566
01:33:23,268 --> 01:33:26,434
which is its own hard problem.
Right. And that will be amazing

1567
01:33:26,434 --> 01:33:28,987
because. Yeah, once you nail
that, then you can go

1568
01:33:28,987 --> 01:33:31,694
everywhere, right? Come to
Texas, go everywhere else.

1569
01:33:31,694 --> 01:33:32,000
Right?

1570
01:33:32,000 --> 01:33:33,000
Barrett Ames: Right.

1571
01:33:32,000 --> 01:33:37,128
Audrow Nash: Hell, yeah. Okay.
Wrapping up, what do you hope

1572
01:33:37,128 --> 01:33:42,000
our watchers and listeners take
away from this interview?

1573
01:33:42,000 --> 01:33:47,304
Barrett Ames: You know, I, I
think there's two things. One is

1574
01:33:47,304 --> 01:33:52,870
that, the future of construction
is automated. And second, that,

1575
01:33:52,870 --> 01:33:58,000
working on robots that work in
construction, it is a really

1576
01:33:58,000 --> 01:34:03,043
interesting problem, that can
help a lot of people, right?

1577
01:34:03,043 --> 01:34:08,435
Like you can you can do really
good things for people and work

1578
01:34:08,435 --> 01:34:12,000
on really interesting technical
problems.

1579
01:34:12,000 --> 01:34:14,337
Audrow Nash: Yeah. I think it's
a super exciting space. All

1580
01:34:14,337 --> 01:34:16,000
right. Hell yeah. Thank you
Barrett. Yeah.

1581
01:34:16,000 --> 01:34:19,000
Barrett Ames: Thank you. Great
to be here.

1582
01:34:19,000 --> 01:34:21,000
Audrow Nash: All right. Bye
everyone.
