1
00:00:00,000 --> 00:00:03,646
Audrow Nash: Hi, everyone,
Audrow here, In this episode, I

2
00:00:03,646 --> 00:00:06,914
share with you several
interviews that I recorded at

3
00:00:06,914 --> 00:00:10,937
ROSCon in 2024, in Denmark. For
those of you who don't know, ROS

4
00:00:10,937 --> 00:00:14,709
is a global robotics conference
hosted by Open Robotics that

5
00:00:14,709 --> 00:00:18,354
brings together a lot of the
robot operating system or ROS

6
00:00:18,354 --> 00:00:22,377
community. It's super fun and a
great way to meet a bunch of the

7
00:00:22,377 --> 00:00:26,274
people building the future. The
first interview is with Marcus

8
00:00:26,274 --> 00:00:30,109
Scheunemann, who is the head of
autonomy at Dexory. Dexory is

9
00:00:30,109 --> 00:00:33,691
based in the UK and is one of
the most promising robotics

10
00:00:33,691 --> 00:00:37,149
startups I've talked to in a
while. They just had their

11
00:00:37,149 --> 00:00:40,543
series B and are scaling
rapidly. The most interesting

12
00:00:40,543 --> 00:00:44,440
part to me is that they put a
huge amount of effort into their

13
00:00:44,440 --> 00:00:48,400
testing infrastructure, and that
seems to be one of the reasons

14
00:00:48,400 --> 00:00:52,109
they're able to scale so fast.
The second interview is with

15
00:00:52,109 --> 00:00:56,069
Christine Fraser, who's the CEO
and founder of Asimovo. Asimovo

16
00:00:56,069 --> 00:01:00,091
is making it easier for teams to
make robots by making it easier

17
00:01:00,091 --> 00:01:03,926
to have repeatable developer
setups. They build everything on

18
00:01:03,926 --> 00:01:07,760
Kubernetes, which gives them a
lot of advantages, but I think

19
00:01:07,760 --> 00:01:11,783
we don't often get in robotics.
You may want to use it yourself.

20
00:01:11,783 --> 00:01:15,743
Finally, the third interview is
with Angelo Corsaro, who is the

21
00:01:15,743 --> 00:01:19,137
CEO and CTO of ZettaScale
Technologies, and it's about

22
00:01:19,137 --> 00:01:22,909
Zenoh. If you were at Rasgon,
you'll know that Zenoh was the

23
00:01:22,909 --> 00:01:26,869
big topic for the conference. If
you don't know ROS 2 uses Data

24
00:01:26,869 --> 00:01:30,703
Distribution Service or DDS as
its middleware. But DDS wasn't

25
00:01:30,703 --> 00:01:34,726
designed for robot applications.
And there are some rough edges,

26
00:01:34,726 --> 00:01:38,371
like there may be a lot of
initial configuration to get it

27
00:01:38,371 --> 00:01:42,143
working, and it doesn't scale
great for multi-robot systems.

28
00:01:42,143 --> 00:01:46,040
Zenoh works to address the rough
edges of DDS, and will be the

29
00:01:46,040 --> 00:01:49,937
default middleware for a future
ROS 2 release. You can already

30
00:01:49,937 --> 00:01:53,457
use it if you want to build from
source. This was really

31
00:01:53,457 --> 00:01:57,291
exciting, and it was great to
hear about Zenoh firsthand from

32
00:01:57,291 --> 00:02:01,189
Angelo. I also recorded another
three interviews that couldn't

33
00:02:01,189 --> 00:02:04,646
fit into this episode, so
they'll be published on X and

34
00:02:04,646 --> 00:02:08,606
YouTube in the next weeks. Grab
a snack, get comfortable, and I

35
00:02:08,606 --> 00:02:12,000
hope you enjoy. Alright, would
you introduce yourself?

36
00:02:12,000 --> 00:02:15,600
Marcus Scheunemann: Yes. Of
course. I'm Marcus Cinnamon, and

37
00:02:15,600 --> 00:02:18,960
I'm the head of autonomy in
Dexory. And yeah there since

38
00:02:18,960 --> 00:02:22,260
three years. We are in the
logistics space and building

39
00:02:22,260 --> 00:02:26,040
this kind of robot since 2 years
now. So this guy is actually a

40
00:02:26,040 --> 00:02:29,580
year old. we kind of released it
when it was, it's, the New

41
00:02:29,580 --> 00:02:33,000
Orleans, New Orleans ROSCon. And
yeah, that's what it is.

42
00:02:33,000 --> 00:02:36,417
Audrow Nash: So tell me a bit
about the company. How large are

43
00:02:36,417 --> 00:02:39,559
you guys? What stage of funding
are you at? Like, all the

44
00:02:39,559 --> 00:02:40,000
details.

45
00:02:40,000 --> 00:02:44,034
Marcus Scheunemann: What's the
size? Like, two years ago, we

46
00:02:44,034 --> 00:02:48,269
didn't had a robot, and we had
an idea we wanted to go into the

47
00:02:48,269 --> 00:02:52,504
logistics space. We wanted to do
revolutionize. Disrupt how you

48
00:02:52,504 --> 00:02:56,134
say it. There's the logistics
sector and the inventory

49
00:02:56,134 --> 00:02:59,899
management. So no robot. Two
years ago. Last year, March

50
00:02:59,899 --> 00:03:04,202
2023. In March, we had our first
deployment. So, well, there's a

51
00:03:04,202 --> 00:03:08,168
first robot. We had one robot.
We were 20 people. Something

52
00:03:08,168 --> 00:03:12,000
like this. Now, this time
around, we are 200 people. Yes.

53
00:03:12,000 --> 00:03:14,000
Audrow Nash: That is so
exciting. More than.

54
00:03:14,000 --> 00:03:17,851
Marcus Scheunemann: 90 robots.
Yeah. Building three robots per

55
00:03:17,851 --> 00:03:21,267
week of this size and and
deployed in three continents.

56
00:03:21,267 --> 00:03:24,000
Eight countries. So, yeah, it's
been a ride.

57
00:03:24,000 --> 00:03:27,286
Audrow Nash: And you just got
your series B funding, if I

58
00:03:27,286 --> 00:03:30,860
understand correctly. That's so
exciting because that's a very

59
00:03:30,860 --> 00:03:34,318
hard one for robotics companies
I hear because you start you

60
00:03:34,318 --> 00:03:37,431
have to be coming into the
scaling phase for them, for

61
00:03:37,431 --> 00:03:40,890
investors to want to do a series
B And you are there, you're

62
00:03:40,890 --> 00:03:44,348
starting to scale and you've
come very quick. Two years, you

63
00:03:44,348 --> 00:03:47,000
say from no robots to robot.
Yeah. No, I mean.

64
00:03:47,000 --> 00:03:50,320
Marcus Scheunemann: It's a it's
a constant fight that challenges

65
00:03:50,320 --> 00:03:53,639
everybody who went to the scale
up knows that. Right. But that's

66
00:03:53,639 --> 00:03:56,803
lots of fun starts because I
loved working on robots. I loved

67
00:03:56,803 --> 00:03:59,967
working on prototypes. Yeah, I
know actually the whole vision

68
00:03:59,967 --> 00:04:03,131
that all they said. And how do
we have to design code and how

69
00:04:03,131 --> 00:04:06,296
do we have test the robot to
bring it to scale and production

70
00:04:06,296 --> 00:04:09,200
izing it? Now I can really
unfolded and actually make it

71
00:04:09,200 --> 00:04:12,416
happen. So for me personally,
it's a challenge. I love it, but

72
00:04:12,416 --> 00:04:15,736
it is. But it is. Right. And you
are absolutely correct. Serious

73
00:04:15,736 --> 00:04:18,744
B is hard because currently
investors are, cautious, let's

74
00:04:18,744 --> 00:04:21,908
put it that way. They want to
see some multipliers. They want

75
00:04:21,908 --> 00:04:24,813
to see some revenue. Yes. And
creating revenue fast is a

76
00:04:24,813 --> 00:04:27,925
challenge. So you have to get
out with the robots very, very

77
00:04:27,925 --> 00:04:30,000
early, create revenue and then
multiply.

78
00:04:30,000 --> 00:04:33,611
Audrow Nash: Definitely. And
because the create revenue fast

79
00:04:33,611 --> 00:04:37,282
is kind of the key part to me
because many companies, if they

80
00:04:37,282 --> 00:04:40,954
are creating revenue, but slowly
it might not make sense as a

81
00:04:40,954 --> 00:04:44,384
venture capital company because
it's not going to get the

82
00:04:44,384 --> 00:04:47,995
returns they want on a timeline
they want. And it's exciting

83
00:04:47,995 --> 00:04:51,606
that you guys are so quickly
creating value. So like so much

84
00:04:51,606 --> 00:04:54,917
that you're scaling, it's
unbelievable to go to so many

85
00:04:54,917 --> 00:04:56,000
robots so quickly.

86
00:04:56,000 --> 00:04:59,189
Marcus Scheunemann: And I thank
you for that. I mean, it's

87
00:04:59,189 --> 00:05:02,324
really appreciate it actually
get a little bit goosebumps

88
00:05:02,324 --> 00:05:05,568
there then then thinking about.
So sometimes we look at our

89
00:05:05,568 --> 00:05:08,922
venture, our journey and it's
still a bit unreal. Yeah. But I

90
00:05:08,922 --> 00:05:11,891
tell you two things. They have
like, the pace is fast,

91
00:05:11,891 --> 00:05:15,465
absolutely essential. So we need
to be fast in the space because,

92
00:05:15,465 --> 00:05:18,710
yes, we found a solution to a
problem that exists, but I, I

93
00:05:18,710 --> 00:05:21,954
must say someone else is also
working on it. So the pace is

94
00:05:21,954 --> 00:05:25,473
essential. So we need the money,
we need the venture capital and

95
00:05:25,473 --> 00:05:28,662
we have to create obviously
there's, there's this traction

96
00:05:28,662 --> 00:05:31,852
then. Yeah, they needs the
revenue. And the second part is

97
00:05:31,852 --> 00:05:35,206
and that's probably interesting
for the listeners, I was very

98
00:05:35,206 --> 00:05:38,395
skeptical to do that, to be
fair. So I learned a lot, from

99
00:05:38,395 --> 00:05:39,000
my founder.

100
00:05:39,000 --> 00:05:40,000
Audrow Nash: The scale quickly.

101
00:05:40,000 --> 00:05:43,245
Marcus Scheunemann: Want to move
to the customer fast, let's put

102
00:05:43,245 --> 00:05:46,439
it that way. So there's a robot
you have and you have a vision.

103
00:05:46,439 --> 00:05:49,279
How accurate and how nice it
should do everything. So we

104
00:05:49,279 --> 00:05:52,575
didn't compromise on safety. For
matter of fact, we always tested

105
00:05:52,575 --> 00:05:55,566
the safety inside. Oh that's
fine. But I thought like, man,

106
00:05:55,566 --> 00:05:58,558
we can do a bit more on the
performance side. But I learned

107
00:05:58,558 --> 00:06:01,701
then from, from from my, from my
founders. They were like, no,

108
00:06:01,701 --> 00:06:04,794
let's push it out to a customer
and learn more. And I like to

109
00:06:04,794 --> 00:06:07,887
learning more about it. But I
was also like, is it not? There

110
00:06:07,887 --> 00:06:10,930
is another risk, right? But as
long as you keep it safe, you

111
00:06:10,930 --> 00:06:14,175
want to be as fast as possible.
It's a customer. It's a learning

112
00:06:14,175 --> 00:06:16,000
we created, made a double as a
pace.

113
00:06:16,000 --> 00:06:18,757
Audrow Nash: I wasn't saying why
bet. Yeah, yeah. Because if

114
00:06:18,757 --> 00:06:21,376
you're not, you're like you're
iterating, you're learning

115
00:06:21,376 --> 00:06:23,903
lessons. You're actually
figuring out what the customer

116
00:06:23,903 --> 00:06:26,522
wants and how to charge them,
and what kind of things are

117
00:06:26,522 --> 00:06:28,911
sticking points for your
technology that you have to

118
00:06:28,911 --> 00:06:31,759
focus on exactly. That they're
getting out there early makes a

119
00:06:31,759 --> 00:06:33,000
lot of sense, and you have.

120
00:06:33,000 --> 00:06:35,857
Marcus Scheunemann: Assumptions
that are wrong. I only as an

121
00:06:35,857 --> 00:06:38,619
engineer, you see maybe the
problem a little bit different

122
00:06:38,619 --> 00:06:41,333
then. Then your customer sees
the problem. So you get the

123
00:06:41,333 --> 00:06:44,048
learning in and you adapt and
you work on the right stuff

124
00:06:44,048 --> 00:06:46,857
early on. So that's that's the
important part. And I think,

125
00:06:46,857 --> 00:06:49,476
yeah, it might sound arrogant,
but I think this one, we

126
00:06:49,476 --> 00:06:52,000
actually did very, very well.
That's why we, so fast.

127
00:06:52,000 --> 00:06:55,714
Audrow Nash: Companies do this
this year. I feel like getting

128
00:06:55,714 --> 00:06:59,611
to series B is a huge milestone
for robotics companies, and most

129
00:06:59,611 --> 00:07:03,082
fail before it. Yes. I'm so,
like a little arrogant perf.

130
00:07:03,082 --> 00:07:07,040
Okay, well, you deserve it. That
that really kind of it. Yeah. So

131
00:07:07,040 --> 00:07:10,876
okay. So the company it sounds
very promising what you guys are

132
00:07:10,876 --> 00:07:14,651
doing. Tell me a bit about the
problem you're solving and what

133
00:07:14,651 --> 00:07:18,000
you, robot like the approach
that your robot is taking.

134
00:07:18,000 --> 00:07:21,492
Marcus Scheunemann: Okay, so the
problem space is, loss

135
00:07:21,492 --> 00:07:24,984
inventory. Pretty much. So,
like, the customer wants to

136
00:07:24,984 --> 00:07:26,000
track, ideally..

137
00:07:26,000 --> 00:07:28,958
Audrow Nash: How large of a
problem is this? Because I've

138
00:07:28,958 --> 00:07:31,863
heard that it's like half of
inventory or something goes

139
00:07:31,863 --> 00:07:35,080
missing or something like crazy
numbers. Yeah. How significant

140
00:07:35,080 --> 00:07:37,000
of a problem is lost inventory.
Well.

141
00:07:37,000 --> 00:07:40,978
Marcus Scheunemann: I'm not a
sales person here. Right. So I,

142
00:07:40,978 --> 00:07:44,891
I don't know exactly what we are
saying normally, but when I

143
00:07:44,891 --> 00:07:48,413
speak to the customer in the
floor space, there's like

144
00:07:48,413 --> 00:07:51,935
they've depending on the
customer if they have service

145
00:07:51,935 --> 00:07:56,109
max or depending on one value of
an item. Right. Like one pallet

146
00:07:56,109 --> 00:07:59,826
can be millions worth if you
just place it into the wrong

147
00:07:59,826 --> 00:08:03,739
aisle in the warehouse that is
1000 100,000mi², square meter

148
00:08:03,739 --> 00:08:04,000
big.

149
00:08:04,000 --> 00:08:05,000
Audrow Nash: It's good for that
line forever.

150
00:08:05,000 --> 00:08:08,470
Marcus Scheunemann: It's it's
yours forever. Yeah. So I was at

151
00:08:08,470 --> 00:08:12,051
the customer side that he found
the box, which was ten years old

152
00:08:12,051 --> 00:08:15,129
because that doubt our our
results, they were like, no,

153
00:08:15,129 --> 00:08:18,319
that can't be the case. This is,
this is this can't exist

154
00:08:18,319 --> 00:08:21,620
anymore. But we found it and
they're like, wow, that's all.

155
00:08:21,620 --> 00:08:25,034
So we, so it just gets lost.
There are two ways, you know, to

156
00:08:25,034 --> 00:08:28,672
to bring value. It's either high
good items that are getting lost

157
00:08:28,672 --> 00:08:32,029
and, but it's just like they're
not there on time anymore or

158
00:08:32,029 --> 00:08:35,499
just in time in general. Right?
If you have to, if you have to

159
00:08:35,499 --> 00:08:38,800
stock a trek, truck, which is
going to the, to the facility

160
00:08:38,800 --> 00:08:42,102
for production, and you don't
find everything just in time,

161
00:08:42,102 --> 00:08:45,460
then you have a problem. You
have a lot of costs. So to bear

162
00:08:45,460 --> 00:08:48,650
for, for the person you you
don't deliver the right goods

163
00:08:48,650 --> 00:08:51,000
too. So these are the two main
points now.

164
00:08:51,000 --> 00:08:54,830
Audrow Nash: Okay. And so to
manage inventory your robot has

165
00:08:54,830 --> 00:08:58,596
this large telescoping set of
sensors. Yeah. So it can scan

166
00:08:58,596 --> 00:09:02,617
all the aisles that have stacks
and stacks and stacks of items.

167
00:09:02,617 --> 00:09:06,638
Yeah. Right on pallets okay. So
that's a general approach. Tell

168
00:09:06,638 --> 00:09:09,000
me a little bit about the robot
okay.

169
00:09:09,000 --> 00:09:12,612
Marcus Scheunemann: Yeah.
Exactly. So we have this we have

170
00:09:12,612 --> 00:09:16,473
this tower. We wanted to go this
a tower to be fast. Right. So

171
00:09:16,473 --> 00:09:20,271
what we can create. And here
again, we learned going early to

172
00:09:20,271 --> 00:09:23,260
the customer. So our first
customers had smaller

173
00:09:23,260 --> 00:09:26,000
warehouses. I'm talking like
20,000, square.

174
00:09:26,000 --> 00:09:27,000
Audrow Nash: And it just seems
enormous.

175
00:09:27,000 --> 00:09:30,889
Marcus Scheunemann: Yeah, yeah.
And now I have a result. It is

176
00:09:30,889 --> 00:09:34,589
enormous. I mean, like, wow, how
can it sets? Oh, yeah. But

177
00:09:34,589 --> 00:09:38,353
nowadays it's ten times bigger
where we are going. Wow. And,

178
00:09:38,353 --> 00:09:41,991
we, we anticipated large of
ours. We just we are not sure,

179
00:09:41,991 --> 00:09:46,005
but this one is like, you have a
lot of seats. We can be service

180
00:09:46,005 --> 00:09:50,019
a tower going like this, a full
scanning standing face, the long

181
00:09:50,019 --> 00:09:53,908
erect, there's one half a meter
per second. You can scan up to

182
00:09:53,908 --> 00:09:57,985
15,000 locations per hour, which
means, like most warehouses have

183
00:09:57,985 --> 00:10:01,497
like 100 or 200,000. So we
manage, in less than a day to

184
00:10:01,497 --> 00:10:05,072
scan, which is the largest
warehouses, which is, which is

185
00:10:05,072 --> 00:10:09,149
creates exactly the value we are
talking about. Yeah. It's a real

186
00:10:09,149 --> 00:10:12,662
time almost of what what we
consider in warehouses, real

187
00:10:12,662 --> 00:10:16,362
time, digital twin of their
assets. It makes them much more

188
00:10:16,362 --> 00:10:20,000
productive. So that's why we
went for the frequency there.

189
00:10:20,000 --> 00:10:22,000
Audrow Nash: For the speed and
everything like this.

190
00:10:22,000 --> 00:10:25,441
Marcus Scheunemann: Also speed.
But there's also an element. And

191
00:10:25,441 --> 00:10:28,720
that's why it kind of emerged of
another thing as well. So we

192
00:10:28,720 --> 00:10:31,677
came from retail. We had a
really cute my favorite mini

193
00:10:31,677 --> 00:10:32,000
robot.

194
00:10:32,000 --> 00:10:34,118
Audrow Nash: So like we heard
about store doors or something

195
00:10:34,118 --> 00:10:35,000
like that, except you go.

196
00:10:35,000 --> 00:10:38,339
Marcus Scheunemann: To the store
electric stores and the and

197
00:10:38,339 --> 00:10:41,844
checking, stock and inventory,
pretty much all the store. There

198
00:10:41,844 --> 00:10:45,405
are a few, a few companies doing
that. Yeah. And, this robot had

199
00:10:45,405 --> 00:10:48,577
this similar scanning face as
you see there. But you must

200
00:10:48,577 --> 00:10:52,027
imagine that this robot kind of
is one of the sections you see

201
00:10:52,027 --> 00:10:55,199
in the store. Yeah. So this
robot here, this example, has

202
00:10:55,199 --> 00:10:58,816
six sections and you can imagine
the kind of six limbs. And so we

203
00:10:58,816 --> 00:11:02,321
could use old business logic of
our robot and just reuse it for

204
00:11:02,321 --> 00:11:05,827
this robot scale it up to six.
And that made us incredibly fast

205
00:11:05,827 --> 00:11:08,998
in the beginning. So this
knowledge, you could just scale

206
00:11:08,998 --> 00:11:10,000
up this. Oh, yeah.

207
00:11:10,000 --> 00:11:13,000
Audrow Nash: And are you using,
ROS, the robot operating system?

208
00:11:13,000 --> 00:11:16,844
Marcus Scheunemann: Yeah, yeah,
yeah, it's a sponsor, of course.

209
00:11:16,844 --> 00:11:20,748
And, we use a lot of open source
software. We contribute back to,

210
00:11:20,748 --> 00:11:24,292
to, to us. So, like, it's very
important for us to, to have

211
00:11:24,292 --> 00:11:28,136
some knowledge of the community
to move fast, see where they are

212
00:11:28,136 --> 00:11:31,800
going and then get back. So what
we are using is Ros two iron

213
00:11:31,800 --> 00:11:35,044
currently. And so cool. We want
to switch like already

214
00:11:35,044 --> 00:11:38,588
yesterday. But sometimes there's
crunch time. So yeah. Yeah

215
00:11:38,588 --> 00:11:42,312
obviously because the end of end
of support coming via and and

216
00:11:42,312 --> 00:11:45,735
the always on the, on the latest
release and yeah we will

217
00:11:45,735 --> 00:11:48,979
actually talk tomorrow also
about that why why this is

218
00:11:48,979 --> 00:11:50,000
important for us.

219
00:11:50,000 --> 00:11:53,944
Audrow Nash: Oh yeah I like that
because it seems like a lot of

220
00:11:53,944 --> 00:11:57,701
companies, they will be like
many companies are still on Ros

221
00:11:57,701 --> 00:12:01,770
one. Yeah. And like some the Ros
boss for humble version but it's

222
00:12:01,770 --> 00:12:05,526
it's like so a lot of companies
they'll pick one and they'll

223
00:12:05,526 --> 00:12:08,970
stay on it versus keep
upgrading. And I guess it's kind

224
00:12:08,970 --> 00:12:13,039
of like you have a little bit of
pain every time you're upgrading

225
00:12:13,039 --> 00:12:16,482
to the new distribution. But
then, you over time you're

226
00:12:16,482 --> 00:12:19,926
eating that pain slowly as
opposed to having it, all at

227
00:12:19,926 --> 00:12:23,306
once when you switch from Ros
one to Ros two. Yeah. Or

228
00:12:23,306 --> 00:12:27,000
something like this, you know
that you guys are doing that.

229
00:12:27,000 --> 00:12:30,287
Marcus Scheunemann: No. That's
perfect. I mean, there's a

230
00:12:30,287 --> 00:12:33,631
there's a motivation element you
have to have like so your

231
00:12:33,631 --> 00:12:37,322
company must embrace robotics in
their core. Yeah. And when they

232
00:12:37,322 --> 00:12:40,666
do that and you know, you can
give them easily the idea of

233
00:12:40,666 --> 00:12:43,895
maintainability and even more
feature ready is is is our

234
00:12:43,895 --> 00:12:47,182
central bit. So we have to
update. So we have to give the

235
00:12:47,182 --> 00:12:50,700
customer the best stuff that is
there because they may see it

236
00:12:50,700 --> 00:12:54,390
somewhere else. We have robotics
as a service. That's an initial

237
00:12:54,390 --> 00:12:58,023
idea okay. We we are not selling
the robot. We are just putting

238
00:12:58,023 --> 00:13:01,367
the robot there and we are
selling the customer data. And,

239
00:13:01,367 --> 00:13:04,712
to be always on the on the
forefront of the technology, we

240
00:13:04,712 --> 00:13:08,287
are on the forefront and with
updating which was to. But I bet

241
00:13:08,287 --> 00:13:11,458
I give you that like as a
company is one was one is the

242
00:13:11,458 --> 00:13:14,803
switch. The switch is as hard.
Yes. And I think one of the

243
00:13:14,803 --> 00:13:18,090
biggest problems is that you
need to invest a lot of time

244
00:13:18,090 --> 00:13:21,030
into the right testing
infrastructure to understand

245
00:13:21,030 --> 00:13:24,721
when you do the switch, that you
can quantitatively say that you

246
00:13:24,721 --> 00:13:27,950
don't break anything your
customer said before. And this

247
00:13:27,950 --> 00:13:31,410
is horribly expensive. And we
are doing this. So like we are

248
00:13:31,410 --> 00:13:34,927
investing a lot in the I never
call it testing when I talk to

249
00:13:34,927 --> 00:13:38,156
anybody, I call it like,
maintainability, insurance that

250
00:13:38,156 --> 00:13:41,328
you can it's easier to sell
internally, because that's,

251
00:13:41,328 --> 00:13:43,000
that's that's a money that's.

252
00:13:43,000 --> 00:13:45,000
Audrow Nash: Driven wants to pay
for testing.

253
00:13:45,000 --> 00:13:47,000
Marcus Scheunemann: It's not
literally not tested.

254
00:13:47,000 --> 00:13:48,655
Audrow Nash: And so
maintainability or something

255
00:13:48,655 --> 00:13:49,000
like that.

256
00:13:49,000 --> 00:13:52,084
Marcus Scheunemann: And future
has been it's already creating

257
00:13:52,084 --> 00:13:55,169
more features easily. Time to
market. Yeah. Reducing any time

258
00:13:55,169 --> 00:13:58,000
to market for each feature. This
is my yeah this is how.

259
00:13:58,000 --> 00:14:01,557
Audrow Nash: I communicate it I
love it I think that's a really

260
00:14:01,557 --> 00:14:04,832
smart approach. So I with that
what advice do you have for

261
00:14:04,832 --> 00:14:08,277
someone who wants more testing
at their organization and this

262
00:14:08,277 --> 00:14:11,044
kind of thing? I changing the
words a little bit.

263
00:14:11,044 --> 00:14:14,149
Maintainability seems like a
good one, but like imagine

264
00:14:14,149 --> 00:14:17,594
someone in a robotics company
where they want to increase the

265
00:14:17,594 --> 00:14:21,095
testing. But it's harder to sell
management on it because they

266
00:14:21,095 --> 00:14:24,426
want new features or something
like this. Like how does how

267
00:14:24,426 --> 00:14:27,758
would you frame it or what are
your thoughts on making that

268
00:14:27,758 --> 00:14:29,000
more likely to happen?

269
00:14:29,000 --> 00:14:32,625
Marcus Scheunemann: It is it's
the time to market. This is

270
00:14:32,625 --> 00:14:36,438
still one of the key qubits,
because if you if you can ensure

271
00:14:36,438 --> 00:14:39,938
that a new feature is not
breaking the old stuff it with

272
00:14:39,938 --> 00:14:44,000
customers, and especially if you
update your whole fleet, what we

273
00:14:44,000 --> 00:14:47,438
are doing as a robot, as a
service company, then you to

274
00:14:47,438 --> 00:14:51,125
decrease as time you need to
testing resources. So and that

275
00:14:51,125 --> 00:14:54,938
is that is kind of the hinder
the metric I'm using mainly to,

276
00:14:54,938 --> 00:14:58,875
to communicate it. Other advise
him I mean it's hard. So like I

277
00:14:58,875 --> 00:15:02,250
think the biggest challenge
every robotics person in a

278
00:15:02,250 --> 00:15:06,250
company is facing is that we are
still in the very early ages of

279
00:15:06,250 --> 00:15:09,813
robotics very much. And the
complexity that, that is that

280
00:15:09,813 --> 00:15:12,875
it's difference in software,
like handling, whole

281
00:15:12,875 --> 00:15:16,688
electromechanical system with
software, it's usually not that

282
00:15:16,688 --> 00:15:20,375
central in, in companies. And
you have to make somehow sure

283
00:15:20,375 --> 00:15:24,000
that this is a central part
where you can drive decisions.

284
00:15:24,000 --> 00:15:26,307
Audrow Nash: I don't fully
understand. What do you mean the

285
00:15:26,307 --> 00:15:28,771
electromechanical like actually
testing on a physical robot? Is

286
00:15:28,771 --> 00:15:31,000
that what you mean, or what do
you mean, that's for sure.

287
00:15:31,000 --> 00:15:33,714
Marcus Scheunemann: There's
also. No, no, I'm leaving the

288
00:15:33,714 --> 00:15:36,381
testing part here. It's an all
about like embracing that

289
00:15:36,381 --> 00:15:39,000
robotics is hard a things. That
is rarely the case. So,

290
00:15:39,000 --> 00:15:42,321
Audrow Nash: What do they do?
They try to abstract it and stay

291
00:15:42,321 --> 00:15:45,000
away from the hardware. Or what
did people do now?

292
00:15:45,000 --> 00:15:49,829
Marcus Scheunemann: Yeah. No. So
it's also I would try to get

293
00:15:49,829 --> 00:15:54,500
this, when you have a software
driven approach, you are not

294
00:15:54,500 --> 00:15:59,646
tackling the problems at a core.
So like, when you say testing in

295
00:15:59,646 --> 00:16:04,000
software, I read, if you can
automate this, you create.

296
00:16:04,000 --> 00:16:06,000
Audrow Nash: You you write it in
simulation. Exactly.

297
00:16:06,000 --> 00:16:09,064
Marcus Scheunemann: You SDF
specific processes you go

298
00:16:09,064 --> 00:16:12,532
through like you have a QA,
maybe, and they're just do like,

299
00:16:12,532 --> 00:16:16,000
oh, I use numbers from 0 to 100.
And then I check it. Right.

300
00:16:16,000 --> 00:16:17,000
Audrow Nash: Like normally the
it's more for testing.

301
00:16:17,000 --> 00:16:20,202
Marcus Scheunemann: Executes it
with more complicated. But in

302
00:16:20,202 --> 00:16:23,352
the end it just says like, oh, I
take crazy numbers and then

303
00:16:23,352 --> 00:16:26,554
let's see what's happening. This
robotics is crazy, right? So

304
00:16:26,554 --> 00:16:29,914
there can be say I environmental
factors like so your sensor all

305
00:16:29,914 --> 00:16:32,854
the time gets a bit untidy for
some reason or like needs

306
00:16:32,854 --> 00:16:35,898
cleaning and then the error. So
the magnitude of the error

307
00:16:35,898 --> 00:16:39,048
changes. So you have to take all
these bits into account and

308
00:16:39,048 --> 00:16:42,303
create constant learnings and
monitoring to be to be adaptable

309
00:16:42,303 --> 00:16:45,453
and to create time for that.
People have to embrace it. This

310
00:16:45,453 --> 00:16:48,445
is complex way, and I think a
lot of the methods done for

311
00:16:48,445 --> 00:16:51,700
companies now are very software
focused. So a lot of templates

312
00:16:51,700 --> 00:16:54,954
we are using, it's like, oh, we
do it in software this way. It

313
00:16:54,954 --> 00:16:58,052
should be done in robotics this
way. And we kind of have to

314
00:16:58,052 --> 00:17:00,991
reinvent the wheel. Yes. And
there needs to be something

315
00:17:00,991 --> 00:17:04,351
different. And so the advice is
like for, for companies, I would

316
00:17:04,351 --> 00:17:07,448
say you have to embrace the core
complexity. So you have to

317
00:17:07,448 --> 00:17:10,546
listen to your robotics people
because they usually do know

318
00:17:10,546 --> 00:17:13,695
where the pain points are. And
if you just not if you ignore

319
00:17:13,695 --> 00:17:16,740
that and think like, well, just
test it with some numbers,

320
00:17:16,740 --> 00:17:18,000
that's not how it works.

321
00:17:18,000 --> 00:17:22,053
Audrow Nash: Okay. Where do you
guys think you're headed in the

322
00:17:22,053 --> 00:17:26,041
next, say, 2 to 5 years? I know
you're scaling like crazy, but

323
00:17:26,041 --> 00:17:29,000
what do you think projecting out
2 to 5 years?

324
00:17:29,000 --> 00:17:34,192
Marcus Scheunemann: Yeah. So,
yeah. So pace is fast, of utmost

325
00:17:34,192 --> 00:17:39,133
importance. To, to to create the
right revenue, to get more

326
00:17:39,133 --> 00:17:44,241
funding, etc.. So, currently we
have five times the robots in

327
00:17:44,241 --> 00:17:46,000
the last year. Right.

328
00:17:46,000 --> 00:17:47,000
Audrow Nash: And while.

329
00:17:47,000 --> 00:17:49,872
Marcus Scheunemann: It is, it is
and you have the right

330
00:17:49,872 --> 00:17:53,005
infrastructure in place to, to
scale this further. So pretty

331
00:17:53,005 --> 00:17:56,243
much personally I think sky's
the limit. I never thought about

332
00:17:56,243 --> 00:17:59,533
anything longer than a month ago
with months, the months ahead.

333
00:17:59,533 --> 00:18:02,770
Yes, exactly. Because that's the
nature of the scale up. Yeah,

334
00:18:02,770 --> 00:18:05,956
yeah, yeah. I guess our next
rescan, I would assume we have a

335
00:18:05,956 --> 00:18:07,000
thousand robots out.

336
00:18:07,000 --> 00:18:11,184
Audrow Nash: Wow. Yeah. I mean
that's a pay one year. Yeah.

337
00:18:11,184 --> 00:18:15,582
Yeah. Because you five times in
the last year. So another five

338
00:18:15,582 --> 00:18:17,000
times. That's crazy.

339
00:18:17,000 --> 00:18:21,052
Marcus Scheunemann: That's very
ambitious to be fair. But yeah

340
00:18:21,052 --> 00:18:25,170
we have already we have already
100 out we want to produce six,

341
00:18:25,170 --> 00:18:27,000
six a week. Yeah. Let's see.

342
00:18:27,000 --> 00:18:29,500
Audrow Nash: Okay. Yeah. The
manufacturing might be the

343
00:18:29,500 --> 00:18:30,000
difficulty.

344
00:18:30,000 --> 00:18:32,070
Marcus Scheunemann:
Manufacturing is exactly like

345
00:18:32,070 --> 00:18:34,521
there's always a new challenge
for you, where you scaling,

346
00:18:34,521 --> 00:18:36,000
where you have to further
optimize.

347
00:18:36,000 --> 00:18:38,280
Audrow Nash: Because you keep
hitting bottlenecks. Basically,

348
00:18:38,280 --> 00:18:40,000
you keep moving from bottleneck
to bottleneck.

349
00:18:40,000 --> 00:18:42,942
Marcus Scheunemann: And you
learn exactly. It's the same.

350
00:18:42,942 --> 00:18:45,832
It's the same point again.
Right? So you learn while you

351
00:18:45,832 --> 00:18:48,877
are doing it. So we don't have,
we are not there for 4 or 5

352
00:18:48,877 --> 00:18:52,026
years yet. You don't have the
full picture of maintainability

353
00:18:52,026 --> 00:18:55,071
and reliability of, of of a
decade. You have to save it for

354
00:18:55,071 --> 00:18:58,116
two years, pretty much. So
there's still a lot to learn. So

355
00:18:58,116 --> 00:19:01,265
yes, we probably have, yeah.
That's that's correct. Me there.

356
00:19:01,265 --> 00:19:04,000
Maybe maybe it's a 600. Well
it's built or something.

357
00:19:04,000 --> 00:19:08,932
Audrow Nash: And still it's so
exciting. It's a lot of robots.

358
00:19:08,932 --> 00:19:11,000
It is. And so quickly too.

359
00:19:11,000 --> 00:19:13,259
Marcus Scheunemann: Yeah. Thank
you for that. I think so too.

360
00:19:13,259 --> 00:19:15,407
It's that right. It's that's
that's a roller coaster to be

361
00:19:15,407 --> 00:19:16,000
fair. It's fast.

362
00:19:16,000 --> 00:19:18,000
Audrow Nash: Oh yeah. Well great
talking with you.

363
00:19:18,000 --> 00:19:21,250
Marcus Scheunemann: Thank you so
much and have a great one. Enjoy

364
00:19:21,250 --> 00:19:22,000
the conference.

365
00:19:22,000 --> 00:19:25,410
Audrow Nash: Yes, definitely.
Here's my conversation with

366
00:19:25,410 --> 00:19:29,000
Christine Fraser from Asimovo.
Would you introduce yourself?

367
00:19:29,000 --> 00:19:34,462
Christine Fraser: I'm Christine
Fraser, I'm the CEO and founder

368
00:19:34,462 --> 00:19:39,318
of Asimovo And Asimovo is
RoboDevOps platform. So it's a

369
00:19:39,318 --> 00:19:44,000
development platform for
developing and testing robots

370
00:19:44,000 --> 00:19:47,957
Audrow Nash: Hell yeah , so tell
me a bit about like, what do you

371
00:19:47,957 --> 00:19:51,000
guys offer I suppose with your
developer platform.

372
00:19:51,000 --> 00:19:54,659
Christine Fraser: So here
basically what we find was that

373
00:19:54,659 --> 00:19:58,511
if you're developing complex
robotics behaviors, are working

374
00:19:58,511 --> 00:20:01,977
on complex robotics tasks,
normally you're not working

375
00:20:01,977 --> 00:20:05,957
alone, you're working as a team
or, most autonomous behaviors.

376
00:20:05,957 --> 00:20:09,809
And I wanting to kind of also
collaborate, for example, with

377
00:20:09,809 --> 00:20:13,853
machine learning experts or data
scientists or AI tools or even

378
00:20:13,853 --> 00:20:17,705
test engineers. So what you want
is to be working on all the

379
00:20:17,705 --> 00:20:21,685
development and robotic tasks,
but doing it within a team. And

380
00:20:21,685 --> 00:20:25,729
what we found was that the tools
weren't really compatible with

381
00:20:25,729 --> 00:20:29,709
doing that easily. And also as
you do more and more simulation

382
00:20:29,709 --> 00:20:33,753
for building complex behaviors,
you need more compute power and

383
00:20:33,753 --> 00:20:36,000
your laptop runs out at some
point.

384
00:20:36,000 --> 00:20:37,000
Audrow Nash: Yeah, very much.

385
00:20:37,000 --> 00:20:40,411
Christine Fraser: These were the
two big things that we

386
00:20:40,411 --> 00:20:44,380
identified early on. If it needs
to be a lot more collaboration,

387
00:20:44,380 --> 00:20:48,039
less developing only in silos.
And you need to basically to

388
00:20:48,039 --> 00:20:51,698
connect whatever compute you
want. And that's kind of where

389
00:20:51,698 --> 00:20:55,233
we start. And so we developed,
the biggest sandbox we can

390
00:20:55,233 --> 00:20:58,953
create where you can have a
digital source of truth for your

391
00:20:58,953 --> 00:21:02,860
whole project. Everyone can work
on that and populated with the

392
00:21:02,860 --> 00:21:06,581
same assets and that you're all
developing. And then you can

393
00:21:06,581 --> 00:21:09,372
choose to work in your
integrated development

394
00:21:09,372 --> 00:21:13,031
environment, either on your
local machines or in the cloud,

395
00:21:13,031 --> 00:21:16,814
or we can connect to on prem
servers or super compute as need

396
00:21:16,814 --> 00:21:17,000
it.

397
00:21:17,000 --> 00:21:20,647
Audrow Nash: Wow! Okay, so that
seems like quite a lot that you

398
00:21:20,647 --> 00:21:23,658
guys are doing like many
different permutations. And

399
00:21:23,658 --> 00:21:27,247
permutations of things that you
may be doing for customers has

400
00:21:27,247 --> 00:21:30,779
that you've kind of iterated to
all that, whereas some people

401
00:21:30,779 --> 00:21:34,021
want on prem devices, some
people. So how does that like

402
00:21:34,021 --> 00:21:37,553
tell me a bit more about that
whole journey of getting to all

403
00:21:37,553 --> 00:21:41,142
these different solutions, and
then also how you have a common

404
00:21:41,142 --> 00:21:44,558
part with all that, because it
sounds a little bit like you

405
00:21:44,558 --> 00:21:48,205
could be spreading yourself then
supporting all these different

406
00:21:48,205 --> 00:21:50,000
things. But tell me about that.

407
00:21:50,000 --> 00:21:53,651
Christine Fraser: It took us
quite a long time to build our

408
00:21:53,651 --> 00:21:57,487
everybody. Let me put it up. And
this wasn't my quick app that

409
00:21:57,487 --> 00:22:00,767
you just create. Yeah. So we're
talking about serious

410
00:22:00,767 --> 00:22:04,294
infrastructure that we're
building. So we, use Kubernetes

411
00:22:04,294 --> 00:22:08,068
so we can connect that to any
computer while we, use a lot of

412
00:22:08,068 --> 00:22:11,533
containerized isolation. But
often in a project, you're,

413
00:22:11,533 --> 00:22:15,122
you're looking at something
that's very compact. So you're

414
00:22:15,122 --> 00:22:18,526
talking about like the total
toolkit can quite often be

415
00:22:18,526 --> 00:22:22,424
between 10 and 20GB. So and then
you're collecting robotic data

416
00:22:22,424 --> 00:22:25,951
that's large files that are,
that aren't used and easy to

417
00:22:25,951 --> 00:22:29,416
upload into Google Drive or
whatever. So what we find is

418
00:22:29,416 --> 00:22:32,757
that actually the tools that you
need for robotics are

419
00:22:32,757 --> 00:22:36,284
professional level tools, and
they don't really exist. So

420
00:22:36,284 --> 00:22:39,812
rather than reinvent the wheel
or develop lots of complex

421
00:22:39,812 --> 00:22:43,400
behaviors, we've focused on
developing the tool kit needed

422
00:22:43,400 --> 00:22:46,742
to do the job. And how can we
professionalize robotics

423
00:22:46,742 --> 00:22:50,331
development. And if you can
enable teams to work together,

424
00:22:50,331 --> 00:22:54,105
then projects can be run faster
and projects are products are

425
00:22:54,105 --> 00:22:57,818
robots that are developed by
multidisciplinary teams will be

426
00:22:57,818 --> 00:23:01,592
stronger as a result, rather
than developed by one PhD person

427
00:23:01,592 --> 00:23:05,119
working only on their machine
and a little silo trying to

428
00:23:05,119 --> 00:23:08,708
connect that all to more of a
digital source of truth that

429
00:23:08,708 --> 00:23:12,112
many can collaborate on will
give it a higher chance of

430
00:23:12,112 --> 00:23:15,639
success, and also a higher
chance of surviving way beyond

431
00:23:15,639 --> 00:23:17,000
the development stage.

432
00:23:17,000 --> 00:23:20,032
Audrow Nash: Definitely, yeah.
Because especially if you can

433
00:23:20,032 --> 00:23:23,064
get people working on a system
together, it's like you don't

434
00:23:23,064 --> 00:23:25,894
have as much difficulty and
setup if you don't have much

435
00:23:25,894 --> 00:23:29,077
difficulty setting up testing,
like everyone doesn't have their

436
00:23:29,077 --> 00:23:32,261
own Docker file or a different
approach to get the whole system

437
00:23:32,261 --> 00:23:35,495
set up. So I can develop and you
just manage it for them and for

438
00:23:35,495 --> 00:23:36,000
that they.

439
00:23:36,000 --> 00:23:38,850
Christine Fraser: Didn't take
all of that way. But what you

440
00:23:38,850 --> 00:23:41,845
find is that lots of robotics
engineers say, well, I've got it

441
00:23:41,845 --> 00:23:44,647
set up exactly as I want it.
I've got this driver and that

442
00:23:44,647 --> 00:23:47,691
driver and this driver, and this
is exactly what I want. So you

443
00:23:47,691 --> 00:23:50,444
can still work like that. You
can still have all of that.

444
00:23:50,444 --> 00:23:53,246
Okay. But we enable you then to
have ten other people also

445
00:23:53,246 --> 00:23:56,000
working on exactly the same set
up that you've got. Yeah.

446
00:23:56,000 --> 00:23:59,753
Audrow Nash: So how what was the
abstraction you made for that or

447
00:23:59,753 --> 00:24:03,333
like how how do you enable that
kind of workflow. Because that

448
00:24:03,333 --> 00:24:06,624
seems like very different
things. So like I have a Docker

449
00:24:06,624 --> 00:24:09,915
container. I'm like forwarding
my I don't know why X11 or

450
00:24:09,915 --> 00:24:13,610
whatever it is. Things I can do
visualization. How do you how do

451
00:24:13,610 --> 00:24:17,132
you support diverse workflows
where you have it on the device

452
00:24:17,132 --> 00:24:20,654
and you have it in the cloud and
on the browser or however it

453
00:24:20,654 --> 00:24:21,000
works?

454
00:24:21,000 --> 00:24:24,880
Christine Fraser: Yeah. So we we
kind of see the projects of

455
00:24:24,880 --> 00:24:28,825
okay, so the number one, what
are the tools. Am I using. Then

456
00:24:28,825 --> 00:24:32,446
what am I actually developing.
So what's our asset we're

457
00:24:32,446 --> 00:24:36,649
actually creating. And then what
do I want to do with it okay. So

458
00:24:36,649 --> 00:24:40,400
first of all the toolkit but
it's all just files. It's all

459
00:24:40,400 --> 00:24:44,151
just you know so yes. So off off
the shelf we've got is it

460
00:24:44,151 --> 00:24:48,095
different possible versions and
roles. Distributions. You got

461
00:24:48,095 --> 00:24:52,169
things like slam and navigation
and standard. So all the things

462
00:24:52,169 --> 00:24:56,373
that each 8% of all ROS projects
probably use the same basic 80%.

463
00:24:56,373 --> 00:25:00,382
Yeah. And then anything which is
more specific, you can either

464
00:25:00,382 --> 00:25:04,327
add it to your base image if you
want everyone to have it, or

465
00:25:04,327 --> 00:25:08,207
those are just simple, easy
modules that can be added. So we

466
00:25:08,207 --> 00:25:12,281
then turn them into components
which are like assets. So in the

467
00:25:12,281 --> 00:25:16,355
same way your test environment
can be an asset or your URDF can

468
00:25:16,355 --> 00:25:19,847
be an asset. Little drivers and
components can also be

469
00:25:19,847 --> 00:25:23,597
individual assets. So then you
can say, okay, whenever I'm

470
00:25:23,597 --> 00:25:27,154
setting up the project this is
my toolkit. And then I'm

471
00:25:27,154 --> 00:25:31,099
populating it with all of these
assets which has capabilities

472
00:25:31,099 --> 00:25:35,302
and things to work on. Okay. And
that's all set up in an easy mix

473
00:25:35,302 --> 00:25:39,376
and match way because we see the
future of kind of the software

474
00:25:39,376 --> 00:25:42,997
side of robotics being more mix
and match. And so if you

475
00:25:42,997 --> 00:25:47,007
encourage people to set up that
way, it's also easy for people

476
00:25:47,007 --> 00:25:51,145
to understand why this is in the
project or not. And it's you're

477
00:25:51,145 --> 00:25:54,379
bringing it in because you want
it to do a certain

478
00:25:54,379 --> 00:25:58,000
functionality, or you want your
robot to have a certain.

479
00:25:58,000 --> 00:25:59,000
Audrow Nash: Yeah it’s
purposeful.

480
00:25:59,000 --> 00:26:01,092
Christine Fraser: Very
purposeful. And then it's also

481
00:26:01,092 --> 00:26:03,500
easier for them to when you're
setting up the next project to

482
00:26:03,500 --> 00:26:05,829
the next project to the next
project, people understand why

483
00:26:05,829 --> 00:26:08,000
things are there. It's not just
hidden in one big file.

484
00:26:08,000 --> 00:26:10,000
Audrow Nash: Yeah. For sure.

485
00:26:10,000 --> 00:26:13,377
Christine Fraser: One person
with a PhD brain, and all of a

486
00:26:13,377 --> 00:26:16,467
sudden it becomes something
that's more accessible and

487
00:26:16,467 --> 00:26:19,844
people understand what the
different things can do and then

488
00:26:19,844 --> 00:26:23,278
why different projects are set
up in different ways. So it's

489
00:26:23,278 --> 00:26:26,655
also about skill dissemination.
And then everyone can learn

490
00:26:26,655 --> 00:26:30,089
together. And then as you're
publishing your project. So for

491
00:26:30,089 --> 00:26:33,580
example we have an open source
project, one of the ones which

492
00:26:33,580 --> 00:26:36,556
might be showing in the
background. And it was three

493
00:26:36,556 --> 00:26:39,704
different ways of doing
navigation, all built on top of

494
00:26:39,704 --> 00:26:42,967
ROS Oh, so it was to the field
navigation up and down the

495
00:26:42,967 --> 00:26:46,000
arrows, navigation and precision
docking, navigation.

496
00:26:46,000 --> 00:26:47,000
Angelo Corsaro: Cool.

497
00:26:47,000 --> 00:26:49,994
Christine Fraser: That started
off as one big file, and we

498
00:26:49,994 --> 00:26:53,195
split it all up so that the test
environment was separate from

499
00:26:53,195 --> 00:26:56,034
the rover robot, which was
separate, which was then all

500
00:26:56,034 --> 00:26:58,976
these different navigation
packages as all published in a

501
00:26:58,976 --> 00:27:02,125
way that people can go and demo
it and trial it and then say,

502
00:27:02,125 --> 00:27:05,378
well, I don't need all the other
stuff, but I do need precision

503
00:27:05,378 --> 00:27:08,475
docking. So I'm going to take
that part and I'm going to put

504
00:27:08,475 --> 00:27:11,211
it into my next project. And
we've also got two other

505
00:27:11,211 --> 00:27:14,102
research projects saying
actually, that test environment

506
00:27:14,102 --> 00:27:17,302
is super useful because it's got
a field and it's got this. So

507
00:27:17,302 --> 00:27:20,400
we're just going to take that
tension environment and use it

508
00:27:20,400 --> 00:27:22,000
as an input to our next budget.

509
00:27:22,000 --> 00:27:24,892
Audrow Nash: Yeah, so much more
composable robotics projects.

510
00:27:24,892 --> 00:27:27,642
Yeah, that's very cool. Are you
leaning on Kubernetes very

511
00:27:27,642 --> 00:27:30,724
heavily for this because so it's
like you pop up one Kubernetes I

512
00:27:30,724 --> 00:27:33,000
don't know what they're called
like instance or.

513
00:27:33,000 --> 00:27:36,000
Christine Fraser: Yeah, that's
an instance again. And so.

514
00:27:36,000 --> 00:27:37,000
Audrow Nash: For for navigation.

515
00:27:37,000 --> 00:27:40,960
Christine Fraser: Yeah, and we
used authorization and we use

516
00:27:40,960 --> 00:27:45,053
them and Kubernetes. But what
Kubernetes allows us to do is be

517
00:27:45,053 --> 00:27:48,881
cloud agnostic or compute
agnostic actually. So we use the

518
00:27:48,881 --> 00:27:52,842
same way as if we wanted to
switch from Amazon to Azure. For

519
00:27:52,842 --> 00:27:56,472
example, we could also switch
from Amazon to your local

520
00:27:56,472 --> 00:27:57,000
machine.

521
00:27:57,000 --> 00:28:00,111
Audrow Nash: Yes. Oh that's how
you do it with all this. So you

522
00:28:00,111 --> 00:28:01,000
spin up a cluster.

523
00:28:01,000 --> 00:28:04,034
Christine Fraser: We've been
deployed there. And then you

524
00:28:04,034 --> 00:28:06,908
decide, where do I want to run
that last year. So that

525
00:28:06,908 --> 00:28:09,462
integrated development
environment and then what

526
00:28:09,462 --> 00:28:12,389
compute power do I want to
create on time. And your own

527
00:28:12,389 --> 00:28:15,636
laptop is one of the options.
It's just a drop down. Yeah. My

528
00:28:15,636 --> 00:28:18,829
laptop or I want to go to on
prem servers if we've connected

529
00:28:18,829 --> 00:28:20,000
it or I want to go to.

530
00:28:20,000 --> 00:28:23,442
Audrow Nash: Oh clever. Yeah.
Okay. I think when we were is

531
00:28:23,442 --> 00:28:27,000
this a relatively new switch to
switch to Kubernetes or like.

532
00:28:27,000 --> 00:28:30,913
Christine Fraser: No, no, it was
always it was always there. But

533
00:28:30,913 --> 00:28:34,825
I think the biggest learning we
got from ROSCon last year, okay,

534
00:28:34,825 --> 00:28:38,616
was being able to work locally
was a must be must. Yeah. So if

535
00:28:38,616 --> 00:28:41,000
you force people to only use the
cloud.

536
00:28:41,000 --> 00:28:42,000
Audrow Nash: They don't want it.

537
00:28:42,000 --> 00:28:45,825
Christine Fraser: They don't
want it for two reasons, two

538
00:28:45,825 --> 00:28:49,918
very clear reasons. Number one,
it's extremely expensive. Why

539
00:28:49,918 --> 00:28:53,944
when most of the things are over
I don't know. Yeah a lot of

540
00:28:53,944 --> 00:28:57,702
compute costs are unnecessary.
And number two my toolkit

541
00:28:57,702 --> 00:29:01,863
doesn't work. I can't use ARVs.
Yeah, and my gazebo. I've only

542
00:29:01,863 --> 00:29:05,755
got a gazebo viewer. I don't
have my full gazebo. And they

543
00:29:05,755 --> 00:29:09,579
all work brilliantly on my local
machine. I want my local

544
00:29:09,579 --> 00:29:13,471
machine. So this ability to
empower the local, but connect

545
00:29:13,471 --> 00:29:16,894
to the cloud for both
collaboration and compute. Is

546
00:29:16,894 --> 00:29:20,584
the magic circle that we've seen
as being the must have

547
00:29:20,584 --> 00:29:23,000
functionality for robotics
adoption.

548
00:29:23,000 --> 00:29:28,020
Audrow Nash: Yes. Okay. I like
that quite a lot. And then so.

549
00:29:28,020 --> 00:29:33,369
That seems very cool. So what. I
guess you have clusters running.

550
00:29:33,369 --> 00:29:37,978
Clusters have all of the
different parts, you might add.

551
00:29:37,978 --> 00:29:43,162
You might have not two you have
normal ROS You might add gazebo

552
00:29:43,162 --> 00:29:48,017
in. And then from this cluster
you can run it on your local

553
00:29:48,017 --> 00:29:53,120
machine. Can you also run some
of it on your local machine and

554
00:29:53,120 --> 00:29:56,000
some of it in the cloud, like if
I.

555
00:29:56,000 --> 00:30:00,000
Christine Fraser: Am yes or no
like that you you spin up.

556
00:30:00,000 --> 00:30:01,000
Audrow Nash: Yeah.

557
00:30:01,000 --> 00:30:03,588
Christine Fraser: But I can
basically work on it in the

558
00:30:03,588 --> 00:30:05,000
morning and at 10:00 I'm safe.

559
00:30:05,000 --> 00:30:06,000
Angelo Corsaro: Yeah.

560
00:30:06,000 --> 00:30:08,000
Christine Fraser: Close reopen.

561
00:30:08,000 --> 00:30:10,490
Audrow Nash: Oh yeah. Because
the cloud, the cluster is still

562
00:30:10,490 --> 00:30:12,000
running for the rest and then
reopen.

563
00:30:12,000 --> 00:30:16,322
Christine Fraser: But connect it
then to the compute to then run

564
00:30:16,322 --> 00:30:19,766
the simulation cycles. For
example. So, so the, the

565
00:30:19,766 --> 00:30:23,751
repository is always in the
cloud. Then I'm taking that and

566
00:30:23,751 --> 00:30:27,803
saying okay, this is what I'm
working on now. Yeah. And when

567
00:30:27,803 --> 00:30:31,990
you close your simulation, as I
said, you want to make changes

568
00:30:31,990 --> 00:30:35,974
and save to your project. And
then I go back in and open it

569
00:30:35,974 --> 00:30:38,000
again and then I can continue.

570
00:30:38,000 --> 00:30:42,063
Audrow Nash: So with this, if I,
as I'm a developer working on

571
00:30:42,063 --> 00:30:45,929
this, I have my project in
there. How do I relate my, like,

572
00:30:45,929 --> 00:30:49,860
development state to, like, my
GitHub or my git project that

573
00:30:49,860 --> 00:30:53,464
I'm working on? It seems like
it's two separate version

574
00:30:53,464 --> 00:30:57,134
things, because the cluster can
have one version and the

575
00:30:57,134 --> 00:31:01,000
repository can have one version,
or they coupled where the.

576
00:31:01,000 --> 00:31:02,000
Christine Fraser: Trees are
coupled.

577
00:31:02,000 --> 00:31:04,471
Audrow Nash: Okay okay. So
you're git config file. Let's

578
00:31:04,471 --> 00:31:05,000
set up okay.

579
00:31:05,000 --> 00:31:08,796
Christine Fraser: So your your
git repository is connected to

580
00:31:08,796 --> 00:31:12,156
your Somerville project
repository. Yep. So I wouldn't

581
00:31:12,156 --> 00:31:15,889
get you to save your code to
Somerville and then but to get.

582
00:31:15,889 --> 00:31:19,000
No I would dare you to save your
code. But to get.

583
00:31:19,000 --> 00:31:20,000
Audrow Nash: And then and then
Somerville and.

584
00:31:20,000 --> 00:31:23,437
Christine Fraser: Then it will
say, oh, there's a, there's a

585
00:31:23,437 --> 00:31:26,759
latest version that's been
uploaded to the git repository.

586
00:31:26,759 --> 00:31:30,368
Do you want to do that? Which is
fine for code. But if I make a

587
00:31:30,368 --> 00:31:33,862
change to my gazebo simulation
environment, that's not and it

588
00:31:33,862 --> 00:31:37,470
will that is and I Somerville or
I've collected a big gross bag

589
00:31:37,470 --> 00:31:40,735
of data from the real robot and
I've uploaded it into the

590
00:31:40,735 --> 00:31:44,000
project that's also part of your
Asimov repository. Yeah.

591
00:31:44,000 --> 00:31:46,949
Audrow Nash: You have like
physical assets or you had some

592
00:31:46,949 --> 00:31:49,949
names for things. Yeah. So yeah,
you have your ROS bag, you

593
00:31:49,949 --> 00:31:52,746
upload that to the project.
Everyone can have access to

594
00:31:52,746 --> 00:31:53,000
that.

595
00:31:53,000 --> 00:31:55,000
Christine Fraser: Everyone's got
access to it. So you and I do.

596
00:31:55,000 --> 00:31:56,000
Audrow Nash: Quite a lot of
things.

597
00:31:56,000 --> 00:31:59,172
Christine Fraser: Yeah. So you
might not not be the one who

598
00:31:59,172 --> 00:32:02,505
wants to. You might then bring
in a test engineer or you might

599
00:32:02,505 --> 00:32:05,570
bring in a data scientist, or
you might, who then is then

600
00:32:05,570 --> 00:32:08,957
taking what you're creating in
simulation and working on it and

601
00:32:08,957 --> 00:32:11,806
giving you feedback, for
example. And you're going to

602
00:32:11,806 --> 00:32:15,032
also do exactly the same way
working between your simulation

603
00:32:15,032 --> 00:32:16,000
or simulated data.

604
00:32:16,000 --> 00:32:17,000
Angelo Corsaro: Yeah.

605
00:32:17,000 --> 00:32:20,188
Christine Fraser: And then
running the same on your

606
00:32:20,188 --> 00:32:24,125
hardware and creating physical
data. Yeah. And this also can be

607
00:32:24,125 --> 00:32:27,563
switch back and forth. So
there's also that came up. So

608
00:32:27,563 --> 00:32:28,000
that's.

609
00:32:28,000 --> 00:32:31,404
Audrow Nash: Very cool. Yeah.
You have a lot of flexibility

610
00:32:31,404 --> 00:32:34,635
because you're built on
Kubernetes. Yeah. Okay. With all

611
00:32:34,635 --> 00:32:38,327
of this so do you have. So if I,
if I have my git repository and

612
00:32:38,327 --> 00:32:41,731
I have my robotics project and
that. Well, do you have good

613
00:32:41,731 --> 00:32:45,192
support for like GitHub actions
for this kind of thing where

614
00:32:45,192 --> 00:32:48,308
it's like I spin up an
environment somehow and I don't

615
00:32:48,308 --> 00:32:52,000
know, maybe you have the action
trigger or something in your, in

616
00:32:52,000 --> 00:32:55,000
some cluster somewhere. That or
if how does it work?

617
00:32:55,000 --> 00:32:59,120
Christine Fraser: It's not all
fully in there yet. There's not

618
00:32:59,120 --> 00:33:02,908
all this all fully in that
bucket. The basics that are in

619
00:33:02,908 --> 00:33:07,028
there. But, it's got a lot of
nice side features that we don't

620
00:33:07,028 --> 00:33:11,016
have fully integrated. But your
robot repository is updated.

621
00:33:11,016 --> 00:33:14,804
Then that update focus will will
follow through into that

622
00:33:14,804 --> 00:33:16,000
Somerville update.

623
00:33:16,000 --> 00:33:19,053
Audrow Nash: Do you mean at all?
Eventually you'll have it

624
00:33:19,053 --> 00:33:22,316
because everything well, like
when Asimovo eventually has that

625
00:33:22,316 --> 00:33:23,000
feature, you.

626
00:33:23,000 --> 00:33:26,558
Christine Fraser: Know, so so it
coordinates it at rep level.

627
00:33:26,558 --> 00:33:30,117
Yeah. No, but sometimes you'll
want to do something yourself.

628
00:33:30,117 --> 00:33:33,442
And not all of those would be
immediately available in at

629
00:33:33,442 --> 00:33:36,883
Somerville, depending on how
you've built it and how you're

630
00:33:36,883 --> 00:33:40,442
working on it. But most people
actually doing the development

631
00:33:40,442 --> 00:33:44,000
in the platform, it's just then
where do you save it back to?

632
00:33:44,000 --> 00:33:46,000
Audrow Nash: Yeah, yeah, it's
on.

633
00:33:46,000 --> 00:33:48,682
Christine Fraser: And you do
that already with git. You've

634
00:33:48,682 --> 00:33:51,457
got a code based repository and
you're kind of branching it.

635
00:33:51,457 --> 00:33:54,000
Yeah, I'm working on it and then
merging it back. Yeah.

636
00:33:54,000 --> 00:33:56,000
Audrow Nash: Your remote is a
separate thing from your where.

637
00:33:56,000 --> 00:33:59,730
Christine Fraser: You're working
off the same principle. So

638
00:33:59,730 --> 00:34:03,333
because all of that is best
practice. Annoying. Yeah. But

639
00:34:03,333 --> 00:34:06,810
around also simulation
environments or cool and results

640
00:34:06,810 --> 00:34:07,000
or.

641
00:34:07,000 --> 00:34:11,267
Audrow Nash: Yes. So where where
do you think you guys are. Like

642
00:34:11,267 --> 00:34:15,267
where are you headed in the next
few years? And like I guess

643
00:34:15,267 --> 00:34:19,333
actually before that, how are
how are you doing as a company?

644
00:34:19,333 --> 00:34:23,000
How is adoption. Yeah. How is
funding all these things.

645
00:34:23,000 --> 00:34:27,471
Christine Fraser: So like our
early evangelist group is not

646
00:34:27,471 --> 00:34:32,245
what we thought it was going to
be like. I was it. So our early

647
00:34:32,245 --> 00:34:36,640
evangelist group is actually
academia. Okay. So anyone who

648
00:34:36,640 --> 00:34:41,263
teaches robotics says, I don't
need all your fancy simulation

649
00:34:41,263 --> 00:34:45,961
runners because I just want to
teach 50 students and have them

650
00:34:45,961 --> 00:34:50,660
set up really, really easily so
I can take exactly what you've

651
00:34:50,660 --> 00:34:55,510
already built, set up my course,
have ten students, 50 students,

652
00:34:55,510 --> 00:35:00,056
high funded students. Access it
and work off it immediately.

653
00:35:00,056 --> 00:35:04,452
Perfect. I can do my large
research projects which involve

654
00:35:04,452 --> 00:35:09,226
SMEs and companies and, and my
researchers, and we can all work

655
00:35:09,226 --> 00:35:14,000
off, a large complex project and
we can all now access version.

656
00:35:14,000 --> 00:35:15,000
Audrow Nash: Yeah. So I have the
latest version.

657
00:35:15,000 --> 00:35:18,743
Christine Fraser: So for both
research and also then

658
00:35:18,743 --> 00:35:23,205
publishing of that research also
then increases the impact. So

659
00:35:23,205 --> 00:35:27,668
there's quite a lot of pauses
And there like you don't need to

660
00:35:27,668 --> 00:35:31,986
keep building anything else.
What. You've got enough for us.

661
00:35:31,986 --> 00:35:36,161
So that's really nice. The
second emerging group, which is

662
00:35:36,161 --> 00:35:39,975
also not where we thought we
would start, is hardware

663
00:35:39,975 --> 00:35:44,366
companies. Let's say I, I have a
robot. Hardware is hard. I'm

664
00:35:44,366 --> 00:35:48,900
really good at my robot, but I
don't want to develop my 100 use

665
00:35:48,900 --> 00:35:53,363
cases for it. Okay. Well, I want
to do is showcase it in a way

666
00:35:53,363 --> 00:35:57,969
that is easy for other people to
try it before they buy it. It's

667
00:35:57,969 --> 00:36:02,000
easy for them to understand what
my hardware is good at.

668
00:36:02,000 --> 00:36:04,000
Audrow Nash: They get a
simulation basically for free.

669
00:36:04,000 --> 00:36:07,884
Christine Fraser: And it's also
a digital like, no, all of the

670
00:36:07,884 --> 00:36:11,830
people who buy it want to then
build do a digital build to then

671
00:36:11,830 --> 00:36:15,275
apply it to something. Yeah,
they can then have a whole

672
00:36:15,275 --> 00:36:19,158
digital part of their product,
that service they can use it to

673
00:36:19,158 --> 00:36:22,916
for sales demos. They can send
somebody a remote control. We

674
00:36:22,916 --> 00:36:26,863
can try it. That's it. By just
logging into our platform. Yeah.

675
00:36:26,863 --> 00:36:30,621
And so that's another really
interesting group. Really nice.

676
00:36:30,621 --> 00:36:34,379
And then another interesting
group are people with problems.

677
00:36:34,379 --> 00:36:38,075
So they're like everyone. So
this is people who aren't even

678
00:36:38,075 --> 00:36:42,084
robotics companies. I'm talking
not robotics companies. Yeah. So

679
00:36:42,084 --> 00:36:45,967
these are more interesting. So
this is for example, apple tree

680
00:36:45,967 --> 00:36:49,600
farmers in the Netherlands, 150
of them. Oh, they've got a

681
00:36:49,600 --> 00:36:53,421
digital twin of the apple pre
and post pruning. If they could

682
00:36:53,421 --> 00:36:56,866
showcase that to all of the
robotics, SMEs and academic

683
00:36:56,866 --> 00:37:00,624
people out there who are looking
for interesting problems to

684
00:37:00,624 --> 00:37:01,000
solve.

685
00:37:01,000 --> 00:37:02,000
Audrow Nash: Oh, that's so cool.

686
00:37:02,000 --> 00:37:05,908
Christine Fraser: And then I've
got Boskalis, a multibillion

687
00:37:05,908 --> 00:37:09,295
dollar dredging company that
just want a really good

688
00:37:09,295 --> 00:37:13,398
underwater simulator. And we can
connect them with the top five

689
00:37:13,398 --> 00:37:16,785
universities within Europe. So
these are also really

690
00:37:16,785 --> 00:37:20,889
interesting. And so ultimately
we will become we are a solution

691
00:37:20,889 --> 00:37:24,927
for a company. We're solving a
company's problem. We are a way

692
00:37:24,927 --> 00:37:29,031
that a company can get a robot
to market fast, because they can

693
00:37:29,031 --> 00:37:33,069
get ten people working on it in
a very efficient way. They can

694
00:37:33,069 --> 00:37:36,000
get. And people working on
complex behaviors.

695
00:37:36,000 --> 00:37:38,154
Audrow Nash: You have an
environment you can distribute,

696
00:37:38,154 --> 00:37:39,000
however, and wherever.

697
00:37:39,000 --> 00:37:43,603
Christine Fraser: Hardware
companies don't have to develop

698
00:37:43,603 --> 00:37:48,603
all the software skills. So then
you also can begin to create a

699
00:37:48,603 --> 00:37:53,048
bit of a, if, well, a digital
supply chains are the apps

700
00:37:53,048 --> 00:37:57,730
behind. Buying the robot is very
cool. So people who can be

701
00:37:57,730 --> 00:38:02,333
really good at one thing but
don't want to provide a whole

702
00:38:02,333 --> 00:38:07,016
solution. They can showcase
examples of their functionality

703
00:38:07,016 --> 00:38:11,937
and then people can come in and
give them consultancy work to,

704
00:38:11,937 --> 00:38:14,000
to do the next best thing.

705
00:38:14,000 --> 00:38:15,000
Audrow Nash: Yeah.

706
00:38:15,000 --> 00:38:18,903
Christine Fraser: Yeah. So but,
but this kind of ecosystem play

707
00:38:18,903 --> 00:38:22,434
or the beginnings of a kind of
development marketplace is

708
00:38:22,434 --> 00:38:26,398
becoming really interesting. But
it's early days and it's one of

709
00:38:26,398 --> 00:38:29,000
those takes time to set up. It
takes time.

710
00:38:29,000 --> 00:38:32,471
Audrow Nash: Figure out the
business model around it. Yeah.

711
00:38:32,471 --> 00:38:35,824
How are you in terms of funding?
Like what? What round of

712
00:38:35,824 --> 00:38:38,000
funding? I assume you have
investors.

713
00:38:38,000 --> 00:38:42,360
Christine Fraser: So we're a
spin out of, kind of private

714
00:38:42,360 --> 00:38:46,950
research organization. So we
have a kind of convertible loan

715
00:38:46,950 --> 00:38:51,769
agreement with them, for, up to,
seed funding stage. So there's

716
00:38:51,769 --> 00:38:55,976
kind of all of last year we got
started, and this year,

717
00:38:55,976 --> 00:39:00,259
basically, with just getting
enough market traction with

718
00:39:00,259 --> 00:39:04,619
initial customers now. So our
SAS platform went live last

719
00:39:04,619 --> 00:39:08,979
month. We've signed up, the
Dutch universities can buy it

720
00:39:08,979 --> 00:39:13,187
through a central purchasing
location, which was a big,

721
00:39:13,187 --> 00:39:17,853
hurdle. And because I know all
their students could just do a

722
00:39:17,853 --> 00:39:19,000
single sign on.

723
00:39:19,000 --> 00:39:20,000
Audrow Nash: Super easy.

724
00:39:20,000 --> 00:39:21,000
Christine Fraser: So then it's
easy.

725
00:39:21,000 --> 00:39:22,000
Audrow Nash: That's great.

726
00:39:21,000 --> 00:39:24,401
Christine Fraser: So then the
first universities are using it,

727
00:39:24,401 --> 00:39:27,582
and we are, part of at least 4
or 5 research projects with

728
00:39:27,582 --> 00:39:30,544
various partners already. Yeah.
On our first couple of

729
00:39:30,544 --> 00:39:34,000
commercial customers and all of
that is happening this quarter.

730
00:39:34,000 --> 00:39:35,000
Audrow Nash: That's exciting!

731
00:39:35,000 --> 00:39:38,125
Christine Fraser: Then opening a
seed round once that market

732
00:39:38,125 --> 00:39:40,000
traction is tangible. And our.
Yeah.

733
00:39:40,000 --> 00:39:43,411
Audrow Nash: Cool. Where are you
guys based out of? The

734
00:39:43,411 --> 00:39:47,070
Netherlands. Okay. Makes sense
to the Netherlands examples.

735
00:39:47,070 --> 00:39:48,000
Yeah. Oh, yeah.

736
00:39:48,000 --> 00:39:50,035
Christine Fraser: Although my
accent is not Dutch, so that

737
00:39:50,035 --> 00:39:52,000
always confusing. Where are you
from? I'm from Scotland.

738
00:39:52,000 --> 00:39:55,073
Audrow Nash: Scotland? How did
you end up in the Netherlands to

739
00:39:55,073 --> 00:39:56,000
start this company?

740
00:39:56,000 --> 00:39:57,000
Christine Fraser: A tall,
attractive man.

741
00:39:57,000 --> 00:40:01,000
Audrow Nash: Hahaha yeah, that's
so funny.

742
00:40:01,000 --> 00:40:03,313
Christine Fraser: Mean that was.
I've been living there for nine

743
00:40:03,313 --> 00:40:04,000
years. Netherlands.

744
00:40:04,000 --> 00:40:07,429
Audrow Nash: Oh yeah. What do
you think? For the like? Where

745
00:40:07,429 --> 00:40:10,000
do you hope to be in a year from
now and for?

746
00:40:10,000 --> 00:40:13,498
Christine Fraser: For us, what
we are doing with the seed

747
00:40:13,498 --> 00:40:17,365
round, as we will be asking for
basic my funding for a year and

748
00:40:17,365 --> 00:40:21,292
a half to two years with a view
that that will take us from what

749
00:40:21,292 --> 00:40:25,036
I consider now to be our MVP.
It's more than that for some of

750
00:40:25,036 --> 00:40:27,000
the customer base. But yeah, in.

751
00:40:27,000 --> 00:40:29,000
Audrow Nash: My my demo there,
like we have enough.

752
00:40:29,000 --> 00:40:32,449
Christine Fraser: It's an MVP
towards more of a company based

753
00:40:32,449 --> 00:40:35,841
solution now. And I believe we
need a year and a half to two

754
00:40:35,841 --> 00:40:39,233
years to fully understand the
sales cycle, the buying cycle,

755
00:40:39,233 --> 00:40:42,173
which is long, and the
stickiness in academia moving

756
00:40:42,173 --> 00:40:45,735
the second. No, no, just oh, and
even within a company, you try

757
00:40:45,735 --> 00:40:49,014
it, you get individuals to do
it, but then you move into a

758
00:40:49,014 --> 00:40:52,293
kind of setting up one project.
Yeah. And then that proves

759
00:40:52,293 --> 00:40:55,855
itself. And then you do more or
you might start with a research

760
00:40:55,855 --> 00:40:59,247
project, which then turns into,
okay, we want to listen, but

761
00:40:59,247 --> 00:41:01,000
once people are in, they're in.

762
00:41:01,000 --> 00:41:04,000
Audrow Nash: Yeah. Kind of like,
yeah.

763
00:41:04,000 --> 00:41:06,682
Christine Fraser: It's helped so
much. Does that cycle. And

764
00:41:06,682 --> 00:41:09,500
robotics is one of these things
which is moving super fast but

765
00:41:09,500 --> 00:41:10,000
super slow.

766
00:41:10,000 --> 00:41:13,190
Audrow Nash: I know she's a good
way to put it, but super slow.

767
00:41:13,190 --> 00:41:14,000
Yeah. Very much.

768
00:41:14,000 --> 00:41:17,630
Christine Fraser: And especially
the mobile robotics or

769
00:41:17,630 --> 00:41:21,920
autonomous behaviors. It's hard.
It's hard. It's hard, it's hard.

770
00:41:21,920 --> 00:41:25,813
So those robots don't don't
produce themselves overnight in

771
00:41:25,813 --> 00:41:29,971
five minutes. So these companies
are all struggling financially

772
00:41:29,971 --> 00:41:34,261
or to make money know. So we can
help because we can make it more

773
00:41:34,261 --> 00:41:38,419
effective, more efficient, a lot
less. We work a lot less time,

774
00:41:38,419 --> 00:41:42,313
but they also need to be able to
buy into in their decision

775
00:41:42,313 --> 00:41:46,537
making process. So yeah, so like
for us, we would like to become

776
00:41:46,537 --> 00:41:50,430
the beginning of of a real
marketplace where academic makes

777
00:41:50,430 --> 00:41:54,654
industry. Yeah. And specifically
SMEs finding really nice niche.

778
00:41:54,654 --> 00:41:58,746
Let's assume, small and medium
sized enterprises. Sorry. It's,

779
00:41:58,746 --> 00:42:00,000
it's like European,

780
00:42:00,000 --> 00:42:03,000
Audrow Nash: I don't know,
acronym. Anyway. Yeah. Okay.

781
00:42:03,000 --> 00:42:06,052
Christine Fraser: But also
because most of those are

782
00:42:06,052 --> 00:42:09,866
presently. So anything under 500
people. Yeah. Which is basically

783
00:42:09,866 --> 00:42:13,270
almost all of them mobile
robotics companies. Right. Well,

784
00:42:13,270 --> 00:42:16,850
today and then we can grow with
them as they grow. And what I

785
00:42:16,850 --> 00:42:20,077
would love is for an ecosystem
where there's 100 really

786
00:42:20,077 --> 00:42:23,188
successful companies all
collaborating and working on

787
00:42:23,188 --> 00:42:26,415
different pieces rather than
industrial robotics, which

788
00:42:26,415 --> 00:42:28,000
really have got ten Giants.

789
00:42:28,000 --> 00:42:29,000
Audrow Nash: Yeah.

790
00:42:29,000 --> 00:42:32,000
Christine Fraser: There's really
ten giants which own that space.

791
00:42:32,000 --> 00:42:33,000
Angelo Corsaro: Yeah.

792
00:42:33,000 --> 00:42:36,358
Christine Fraser: And I would
really like autonomous behavior

793
00:42:36,358 --> 00:42:39,385
building and the kind of tier
two and tier three mobile

794
00:42:39,385 --> 00:42:42,853
robotics applications have got a
bigger potential for impact in

795
00:42:42,853 --> 00:42:45,000
society. I bet to get to market
faster.

796
00:42:45,000 --> 00:42:46,000
Audrow Nash: Yes.

797
00:42:45,000 --> 00:42:48,588
Christine Fraser: And that's a
big part of our kind of vision

798
00:42:48,588 --> 00:42:52,059
is to enable the tier two and
tier three more applications.

799
00:42:52,059 --> 00:42:55,706
But people can't afford to put 5
million into the development.

800
00:42:55,706 --> 00:42:59,294
Yeah for sure. Get it to market.
So if we can find a way that

801
00:42:59,294 --> 00:43:02,941
they can all get to market much.
Yeah. And then they all grow,

802
00:43:02,941 --> 00:43:05,000
then that is a whole benefit.
Yeah.

803
00:43:05,000 --> 00:43:08,389
Audrow Nash: Okay. Hell yeah.
Yeah. So where can people go to

804
00:43:08,389 --> 00:43:09,000
learn more?

805
00:43:09,000 --> 00:43:12,398
Christine Fraser: If you go to
our website, Asimovo.com, you

806
00:43:12,398 --> 00:43:15,966
can sign or you can just we've
got case studies on there. We've

807
00:43:15,966 --> 00:43:18,968
also got develop our case
studies, which is more code

808
00:43:18,968 --> 00:43:22,366
based. Or you can jump into the
platform and set yourself up

809
00:43:22,366 --> 00:43:25,878
with a free account and all of
these ones which are showing in

810
00:43:25,878 --> 00:43:28,993
the background. You can
basically get most of those for

811
00:43:28,993 --> 00:43:32,335
for free on the platform, jump
in there, demo it so all the

812
00:43:32,335 --> 00:43:35,959
documentation and run your own.
It takes a couple of minutes for

813
00:43:35,959 --> 00:43:38,848
it to be spun up on the
hardware. Yeah. Kubernetes.

814
00:43:38,848 --> 00:43:42,473
Yeah. But once you're you're in
there you can then play with it.

815
00:43:42,473 --> 00:43:45,701
And anything you like you can
copy. And that can be as an

816
00:43:45,701 --> 00:43:49,212
individual asset or as a whole
project. Cool. So, follow us on

817
00:43:49,212 --> 00:43:52,837
LinkedIn to keep up to date with
who we're hiring and what we're

818
00:43:52,837 --> 00:43:55,952
doing. Sign up for our
newsletter. Also, all the things

819
00:43:55,952 --> 00:43:59,181
and all of those kind of things,
but there's a lot on the

820
00:43:59,181 --> 00:44:02,352
website. But for this type of
project, with this type of

821
00:44:02,352 --> 00:44:05,920
customer, try it for yourself.
But if you're a company, or even

822
00:44:05,920 --> 00:44:09,432
an individual who wants to know
more, we've got webinars which

823
00:44:09,432 --> 00:44:13,000
take you through more of a click
but buy click and teaches you.

824
00:44:13,000 --> 00:44:14,000
Audrow Nash: Yeah.

825
00:44:14,000 --> 00:44:17,664
Christine Fraser: And we also
have got more specific,

826
00:44:17,664 --> 00:44:21,949
onboarding for companies because
there's certain functionality

827
00:44:21,949 --> 00:44:26,166
you don't see as an individual.
Yeah, as an institute or as a

828
00:44:26,166 --> 00:44:29,000
company. And you need a kind of
demo for.

829
00:44:29,000 --> 00:44:31,000
Audrow Nash: Right. Hell yeah.
Thank you very much.

830
00:44:31,000 --> 00:44:32,000
Christine Fraser: Thank you.

831
00:44:32,000 --> 00:44:35,353
Audrow Nash: And our last
interview, my conversation with

832
00:44:35,353 --> 00:44:38,882
Angelo Corsaro from ZettaScale
Technologies. Hi would do you

833
00:44:38,882 --> 00:44:40,000
introduce yourself?

834
00:44:40,000 --> 00:44:43,508
Angelo Corsaro: Sure. Hello. So
I'm Angelo Angelo Corsaro

835
00:44:43,508 --> 00:44:47,323
ZettaScale CEO and CTO and, you
know enjoying as usual the ROS

836
00:44:47,323 --> 00:44:48,000
conference.

837
00:44:48,000 --> 00:44:51,343
Audrow Nash: Hell yeah. So tell
me what you guys make at

838
00:44:51,343 --> 00:44:52,000
ZettaScale.

839
00:44:52,000 --> 00:44:54,813
Angelo Corsaro: So at
ZettaScale, I mean, we

840
00:44:54,813 --> 00:44:58,394
differentiate ourselves.
Protocol experts. So we started

841
00:44:58,394 --> 00:45:02,358
working. Actually, I was one of
the first one that started DDS

842
00:45:02,358 --> 00:45:05,939
standardization back in the
days. Right. And then around

843
00:45:05,939 --> 00:45:09,519
2015, I decided actually to
leave the DDS working group,

844
00:45:09,519 --> 00:45:13,611
resign as a co-chair. I started
working on a new protocol Zenoh.

845
00:45:13,611 --> 00:45:17,384
The question might be why did I
decide to do that? So first

846
00:45:17,384 --> 00:45:21,284
there were multiple reasons.
Okay. So and those were based on

847
00:45:21,284 --> 00:45:25,248
the experience that we had on,
smart cities. Okay. And to some

848
00:45:25,248 --> 00:45:28,957
application in aerospace and
defense that were showing two

849
00:45:28,957 --> 00:45:32,857
things on one side, some of the
limitations of DDS in getting

850
00:45:32,857 --> 00:45:36,757
outside of the, let's say, sweet
spot, the kind of system for

851
00:45:36,757 --> 00:45:40,529
which had been designed, I’ll
get there in a moment. And on

852
00:45:40,529 --> 00:45:44,430
the other side, the fact that
while, you know, traditionally,

853
00:45:44,430 --> 00:45:48,202
there were people working on
embedded system, which weren't

854
00:45:48,202 --> 00:45:51,655
always connected, and then
people working on what they

855
00:45:51,655 --> 00:45:55,619
would call edge system, which
were connected but closed. Okay.

856
00:45:55,619 --> 00:45:59,583
And then, you know, there was
the cloud, but all of these were

857
00:45:59,583 --> 00:46:03,419
loosely connected, connected,
and we were transitioning to a

858
00:46:03,419 --> 00:46:07,064
world in which system, more and
more were spanning from a

859
00:46:07,064 --> 00:46:10,772
microcontroller up to a data
center. Think about the robot

860
00:46:10,772 --> 00:46:14,737
today. Yeah, right. I mean, in a
robot, even like the one that

861
00:46:14,737 --> 00:46:18,637
lives somewhere over there, you
have microcontroller, we have

862
00:46:18,637 --> 00:46:22,601
general purpose processor, but
then this data has to flow into

863
00:46:22,601 --> 00:46:26,693
it. And there was no protocol. A
couple of great. Okay. That was

864
00:46:26,693 --> 00:46:30,018
creating lots of, you know,
struggling over customer

865
00:46:30,018 --> 00:46:32,000
frustration. For me it was not.

866
00:46:32,000 --> 00:46:34,000
Audrow Nash: Like it doesn't
fit.

867
00:46:34,000 --> 00:46:37,306
Angelo Corsaro: It was ugly. I
mean you had to integrate

868
00:46:37,306 --> 00:46:40,316
multiple things and even
esthetically right. Easily

869
00:46:40,316 --> 00:46:43,799
beautiful and sensor esthetics
for me it's quite important.

870
00:46:43,799 --> 00:46:47,340
Yeah. And I think in tactical
matters something has to be as

871
00:46:47,340 --> 00:46:51,000
simple as possible. Definitely.
Absolutely beautiful. Elegant.

872
00:46:51,000 --> 00:46:53,985
Audrow Nash: For sure. Yeah. And
if it's simple to you can have a

873
00:46:53,985 --> 00:46:56,786
simple understanding of it and
then you can do a lot with it.

874
00:46:56,786 --> 00:46:59,495
If you have a lot of complexity,
you get bogged down in the

875
00:46:59,495 --> 00:47:00,000
complexity.

876
00:47:00,000 --> 00:47:02,850
Angelo Corsaro: And again, you
know, it's very hard to do

877
00:47:02,850 --> 00:47:05,950
things that are simple because
it assumes that you have deeply

878
00:47:05,950 --> 00:47:09,200
understood the problem. Okay. So
it's much easier to come up with

879
00:47:09,200 --> 00:47:11,000
solutions that are very
complicated.

880
00:47:11,000 --> 00:47:15,000
Audrow Nash: Oh yes. Yeah. Super
spiky. Lots of things.

881
00:47:15,000 --> 00:47:18,532
Angelo Corsaro: That's sometimes
that's an indication that you

882
00:47:18,532 --> 00:47:21,949
have not really distilled the
problem. Yeah. And you have no

883
00:47:21,949 --> 00:47:25,595
real really captured the key of
the problem. And so getting back

884
00:47:25,595 --> 00:47:29,241
to to the reason that pushed us
to do that. So I wanted really a

885
00:47:29,241 --> 00:47:32,544
protocol that could work. And I
went and there was another

886
00:47:32,544 --> 00:47:33,000
problem.

887
00:47:33,000 --> 00:47:36,000
Audrow Nash: An end to end like
microcontroller to data center.

888
00:47:36,000 --> 00:47:38,816
Angelo Corsaro: Yes. And the
other problem was this. So in

889
00:47:38,816 --> 00:47:41,777
order to do this right, there
were several challenges because

890
00:47:41,777 --> 00:47:44,689
it means you need to be able to
work on a low power network,

891
00:47:44,689 --> 00:47:46,000
which are very constraints.

892
00:47:46,000 --> 00:47:47,000
Audrow Nash: Microcontrollers.

893
00:47:47,000 --> 00:47:50,186
Angelo Corsaro: Okay. Right. On
microcontroller two that are

894
00:47:50,186 --> 00:47:53,425
constraints in computing and
CPU. But then low power networks

895
00:47:53,425 --> 00:47:56,717
are constrained in bandwidth.
You need to run efficiently over

896
00:47:56,717 --> 00:47:59,000
the internet, right? You need to
be secure.

897
00:47:59,000 --> 00:48:03,000
Audrow Nash: Okay. Yeah. So it's
it's a huge spectrum.

898
00:48:03,000 --> 00:48:06,081
Angelo Corsaro: It's not easy
because, you know, there is

899
00:48:06,081 --> 00:48:09,378
something very simple. Okay.
Which sequence number resolution

900
00:48:09,378 --> 00:48:12,676
do we use most people don't even
think about sequence number.

901
00:48:12,676 --> 00:48:13,000
Right.

902
00:48:13,000 --> 00:48:16,000
Audrow Nash: Those for you
indexing basically.

903
00:48:16,000 --> 00:48:18,965
Angelo Corsaro: Right.
Essentially decides how many

904
00:48:18,965 --> 00:48:22,744
messages you can have in flight,
while not being able to detect a

905
00:48:22,744 --> 00:48:26,174
rollover. So if you're running
on a constrained system, you

906
00:48:26,174 --> 00:48:29,779
want this number to be fairly
small. Yeah, because that limits

907
00:48:29,779 --> 00:48:32,919
the wide overhead. If you're
running, let's say over a

908
00:48:32,919 --> 00:48:36,581
satellite link, you want to have
lots of parking on fly. So you

909
00:48:36,581 --> 00:48:39,895
want to have a fairly big
resolution for sequence number.

910
00:48:39,895 --> 00:48:43,384
Yeah. Typical protocol. And the
fixed resolution that didn't

911
00:48:43,384 --> 00:48:46,756
work for us. So that's something
that you can configure in

912
00:48:46,756 --> 00:48:50,244
trouble. Every session can use a
different sequence. Now for

913
00:48:50,244 --> 00:48:54,023
instance okay. There are lots of
protocol engineering that we did

914
00:48:54,023 --> 00:48:57,047
as such to make sure that we
could read efficiently,

915
00:48:57,047 --> 00:49:00,302
communicate end to end. Yeah,
that's one point. But that

916
00:49:00,302 --> 00:49:03,674
wasn't enough. The other point
is, as we have this kind of

917
00:49:03,674 --> 00:49:07,395
systems and especially you think
about robots, and as soon as we

918
00:49:07,395 --> 00:49:10,593
start adding things that are
battery powered, there's a

919
00:49:10,593 --> 00:49:14,023
question of where do we store
the data? Because pushing the

920
00:49:14,023 --> 00:49:17,279
data to the cloud requires
connectivity, requires better

921
00:49:17,279 --> 00:49:20,884
communication, consumes not so
bad. Wouldn't it be great if we

922
00:49:20,884 --> 00:49:24,488
could keep data stored locally
or as close as possible, and in

923
00:49:24,488 --> 00:49:27,628
query from anywhere, without
knowing where the data is

924
00:49:27,628 --> 00:49:30,767
stored? That's the other
problem. That's solved, okay?

925
00:49:30,767 --> 00:49:34,081
And that's like, you know,
usually people discovered zero

926
00:49:34,081 --> 00:49:37,802
and that's you. I say the tip of
the iceberg is, oh, super cool.

927
00:49:37,802 --> 00:49:41,116
The a pub sub that can span and
twin. And then they, they

928
00:49:41,116 --> 00:49:44,779
discover the rest of Xeno which
is the query. So the query, the

929
00:49:44,779 --> 00:49:48,267
distributed query. And that's
where the real power, you know

930
00:49:48,267 --> 00:49:51,000
comes in. The realization comes
wow okay. Yeah.

931
00:49:51,000 --> 00:49:53,609
Audrow Nash: What's I haven't
yet discover that I guess. All

932
00:49:53,609 --> 00:49:56,000
right. So the distributed query
tell me how to do that.

933
00:49:56,000 --> 00:49:59,763
Angelo Corsaro: Right. So the
concept it's actually fairly

934
00:49:59,763 --> 00:50:03,527
trivial, but the implications
are quite interesting. So we

935
00:50:03,527 --> 00:50:07,290
have this concept of equitable.
What is equitable. It's an

936
00:50:07,290 --> 00:50:10,924
oracle that it's around the
network. And it simply tells

937
00:50:10,924 --> 00:50:14,882
that it provides data for, you
know, a given data center or a

938
00:50:14,882 --> 00:50:19,034
key expression. So you copy it.
Okay. Nicky, expression is a way

939
00:50:19,034 --> 00:50:23,122
of representing a data set. And
then, how does it provide data?

940
00:50:23,122 --> 00:50:26,691
We shouldn't care. But then
we'll get that. And so then

941
00:50:26,691 --> 00:50:30,519
there is this get operations
end, which is how you actually

942
00:50:30,519 --> 00:50:34,412
submit a query. When you do I
get you provide a selector. We

943
00:50:34,412 --> 00:50:38,176
call it in this selector. You
specify the data set in some

944
00:50:38,176 --> 00:50:42,069
predicates. And at that point
right, imagine us the data set

945
00:50:42,069 --> 00:50:45,897
that you're specifying. It's
really, you know, logic like a

946
00:50:45,897 --> 00:50:47,000
mathematical set.

947
00:50:47,000 --> 00:50:48,000
Audrow Nash: Yeah.

948
00:50:48,000 --> 00:50:50,000
Angelo Corsaro: And in the case
what zero does.

949
00:50:50,000 --> 00:50:51,000
Audrow Nash: It narrow it down.

950
00:50:51,000 --> 00:50:54,328
Angelo Corsaro: It looks at the
set you're looking for. It does

951
00:50:54,328 --> 00:50:57,709
a set cover. Yeah. With respect
to the, the variable that are on

952
00:50:57,709 --> 00:51:00,931
the network and then decides
what is the optimal set of query

953
00:51:00,931 --> 00:51:04,311
boat to which we need to run the
query in order to get your data

954
00:51:04,311 --> 00:51:07,481
in. The consequence is that you
don't need to know where the

955
00:51:07,481 --> 00:51:10,650
data is. Yeah. Depending on
where you are on the network, we

956
00:51:10,650 --> 00:51:13,661
might rule that maybe pretty
magical. Yeah. Two different

957
00:51:13,661 --> 00:51:17,095
credible. Yeah. So that supports
natively mobility load balancing

958
00:51:17,095 --> 00:51:20,000
and fault tolerance without you
even thinking about it.

959
00:51:20,000 --> 00:51:21,000
Audrow Nash: Amazing.

960
00:51:21,000 --> 00:51:25,095
Angelo Corsaro: And then if you
think about a database right.

961
00:51:25,095 --> 00:51:29,189
The combination of a subscribe
in an equitable is the reading

962
00:51:29,189 --> 00:51:33,015
and writing side of database.
And that's how we integrate

963
00:51:33,015 --> 00:51:36,707
databases and like, you know,
Amazon S3 or Rox DB or in

964
00:51:36,707 --> 00:51:40,601
flexible or whatever. And a
computation is just equitable.

965
00:51:40,601 --> 00:51:44,695
And now if you think again for a
moment, you know, map reduce

966
00:51:44,695 --> 00:51:48,320
okay, what is MapReduce. You
want to get a computation

967
00:51:48,320 --> 00:51:52,415
trigger on multiple node and
then you want to consolidate the

968
00:51:52,415 --> 00:51:56,241
result. Yeah. But that you can
achieve that with credible

969
00:51:56,241 --> 00:52:00,403
because you can have multiple
queryable essentially responding

970
00:52:00,403 --> 00:52:04,229
the same query, operating a
different subset of the data.

971
00:52:04,229 --> 00:52:08,324
And then when the query score is
computed, we consolidate the

972
00:52:08,324 --> 00:52:12,687
data back order. Oh. So all of a
sudden in one protocol, yes, you

973
00:52:12,687 --> 00:52:16,312
can deal with that. Then most
pubsub that addressed so

974
00:52:16,312 --> 00:52:20,272
storages in computation in a
completely transparent manner.

975
00:52:20,272 --> 00:52:23,964
So you do not need to know where
the data is, where the

976
00:52:23,964 --> 00:52:28,260
computation is. Yeah. End up by
as a byproduct to member routing

977
00:52:28,260 --> 00:52:32,422
algorithm where you depending on
where you are on the network,

978
00:52:32,422 --> 00:52:36,651
you're always going to pick the
optimal place either to get the

979
00:52:36,651 --> 00:52:39,000
data or to trigger the
computation.

980
00:52:39,000 --> 00:52:43,308
Audrow Nash: Yeah, that's so
cool. Thinking of it like a

981
00:52:43,308 --> 00:52:47,538
distributed database is a very
cool way. it has all the

982
00:52:47,538 --> 00:52:50,000
computation and it roots it all.

983
00:52:50,000 --> 00:52:53,008
Angelo Corsaro: Correct. And we
also have alignment. So imagine

984
00:52:53,008 --> 00:52:55,492
that the network gets
partitioned. Yeah. If you have

985
00:52:55,492 --> 00:52:58,404
different storage then whenever
the network heals the storage

986
00:52:58,404 --> 00:53:01,413
are realigned automatically. So
we have a pretty cool idea. And

987
00:53:01,413 --> 00:53:04,039
so that's the bottom of the
iceberg that usually people

988
00:53:04,039 --> 00:53:06,857
discover that later once they
realize the full power. It is

989
00:53:06,857 --> 00:53:07,000
it.

990
00:53:07,000 --> 00:53:10,859
Audrow Nash: Wow. Yeah yeah
yeah, I see it I see it okay.

991
00:53:10,859 --> 00:53:14,786
With this. How does it work?
Like, I know, I know, it's it

992
00:53:14,786 --> 00:53:18,646
sounds like it's very complex.
You have this very amazing

993
00:53:18,646 --> 00:53:20,000
networking protocol.

994
00:53:20,000 --> 00:53:22,000
Angelo Corsaro: Yeah. Go ahead.
How does it.

995
00:53:22,000 --> 00:53:25,561
Audrow Nash: Like at a high
level. How can I understand

996
00:53:25,561 --> 00:53:29,122
what's going on that leads to
this, really nice routing

997
00:53:29,122 --> 00:53:31,000
behavior and synchronization.

998
00:53:31,000 --> 00:53:35,432
Angelo Corsaro: I mean, the
underlying protocol is actually

999
00:53:35,432 --> 00:53:40,089
quite simple. And I think what
allowed us to make it so simple

1000
00:53:40,089 --> 00:53:44,897
is that we really narrow to the
key primitives. Okay. And in the

1001
00:53:44,897 --> 00:53:49,479
end to what we do. Right? We
just route key especially on the

1002
00:53:49,479 --> 00:53:53,836
network. Right. So and there are
some technical, let's say

1003
00:53:53,836 --> 00:53:58,568
complexities on how to make this
efficient, but that, you know,

1004
00:53:58,568 --> 00:54:03,000
the key concept are actually
pretty straightforward, right?

1005
00:54:03,000 --> 00:54:06,556
Audrow Nash: So when you say key
expression, what does that mean

1006
00:54:06,556 --> 00:54:07,000
exactly?

1007
00:54:07,000 --> 00:54:10,852
Angelo Corsaro: Imagine a ROS
topic. Right. That's the basic

1008
00:54:10,852 --> 00:54:14,704
form of the key expression.
Yeah. Now imagine that. Right. I

1009
00:54:14,704 --> 00:54:18,427
have multiple robots I could
say. But one there's less but

1010
00:54:18,427 --> 00:54:22,472
one's less same developed. Yeah.
So that's a key expression for

1011
00:54:22,472 --> 00:54:26,516
us. It's a full key. There are
no wildcards that represent just

1012
00:54:26,516 --> 00:54:30,432
one data item okay. If I have
multiple robots I could say but

1013
00:54:30,432 --> 00:54:33,000
star slash cm devo. So imagine
that now.

1014
00:54:33,000 --> 00:54:34,905
Audrow Nash: Also you can have
wildcard car any queries. And

1015
00:54:34,905 --> 00:54:35,000
so.

1016
00:54:35,000 --> 00:54:36,000
Angelo Corsaro: Whenever you.

1017
00:54:36,000 --> 00:54:38,234
Audrow Nash: Have one and you
could probably do some filtering

1018
00:54:38,234 --> 00:54:40,000
con. Right. So. Oh that's super
super cool. Okay.

1019
00:54:40,000 --> 00:54:42,819
Angelo Corsaro: So because this
issue when I talk about key

1020
00:54:42,819 --> 00:54:45,638
expression you realize because
of wildcard, all of a sudden

1021
00:54:45,638 --> 00:54:48,553
you're representing a set of
keys as opposed to a single key.

1022
00:54:48,553 --> 00:54:51,276
Yeah. And you can do that on
subscription. Right. Because

1023
00:54:51,276 --> 00:54:54,000
maybe I want to receive there's
some table from multiple.

1024
00:54:54,000 --> 00:54:55,000
Audrow Nash: Robo that's so
powerful. Yeah.

1025
00:54:55,000 --> 00:54:59,776
Angelo Corsaro: And in query
because imagine right. I want to

1026
00:54:59,776 --> 00:55:04,786
know the state of all my fate of
robots. Yeah. You could do get.

1027
00:55:04,786 --> 00:55:09,797
But isn't it get less robot star
slash battery level. That would

1028
00:55:09,797 --> 00:55:14,416
trigger a distributed query.
Yeah. And get the data to you.

1029
00:55:14,416 --> 00:55:17,000
Now, what is that data? Well,
it.

1030
00:55:17,000 --> 00:55:18,000
Audrow Nash: Could be in a bunch
of nodes.

1031
00:55:18,000 --> 00:55:21,242
Angelo Corsaro: Might be on your
robots. Yeah. Might be stored in

1032
00:55:21,242 --> 00:55:24,384
a cloud. And you are closer to
the cloud because imagine I'm by

1033
00:55:24,384 --> 00:55:27,476
the robots and I do the query
maybe the fastest way to resolve

1034
00:55:27,476 --> 00:55:30,668
it is straight on the robot. If
you are, you know, in California

1035
00:55:30,668 --> 00:55:33,362
and the robot is here and the
data is also stored on a

1036
00:55:33,362 --> 00:55:36,603
storage, you know, on the Google
Cloud. Yeah. The closest storage

1037
00:55:36,603 --> 00:55:38,000
will be on the Google Cloud.

1038
00:55:38,000 --> 00:55:39,000
Audrow Nash: Yeah.

1039
00:55:39,000 --> 00:55:41,552
Angelo Corsaro: Do you need to
care? No. And that's great

1040
00:55:41,552 --> 00:55:44,284
because you don't care whether
they suck at Google to get the

1041
00:55:44,284 --> 00:55:45,000
data, don't you.

1042
00:55:45,000 --> 00:55:49,270
Audrow Nash: Yeah, I guess I
don't fully understand how it's

1043
00:55:49,270 --> 00:55:53,327
working from there. So that
seems wonderful. Yeah. And so

1044
00:55:53,327 --> 00:55:57,527
you get to not care about where
the servers. Yeah. Okay. So

1045
00:55:57,527 --> 00:56:01,868
continuing I guess, with the
explanation of how it's working,

1046
00:56:01,868 --> 00:56:05,000
like, I guess conceptually, what
we're okay.

1047
00:56:05,000 --> 00:56:10,181
Angelo Corsaro: So conceptually
with respect to how we figure

1048
00:56:10,181 --> 00:56:15,363
out where about to do exactly
that. Okay. So essentially in a

1049
00:56:15,363 --> 00:56:20,544
Zenoh system, you have to think
that, we reconstruct a graph.

1050
00:56:20,544 --> 00:56:25,556
Okay. And over this graph,
imagine that we have annotations

1051
00:56:25,556 --> 00:56:27,000
like, like going.

1052
00:56:27,000 --> 00:56:29,000
Audrow Nash: Everywhere meta
data at each node.

1053
00:56:29,000 --> 00:56:31,955
Angelo Corsaro: So let's make an
analogy with when you're driving

1054
00:56:31,955 --> 00:56:34,682
okay. So when you're driving in
this, one of the the ways in

1055
00:56:34,682 --> 00:56:37,000
which we make this query very
efficient okay. Yeah.

1056
00:56:37,000 --> 00:56:39,000
Audrow Nash: This is exactly
what I want to try it on.

1057
00:56:39,000 --> 00:56:43,100
Angelo Corsaro: So suppose that
you are going from San Francisco

1058
00:56:43,100 --> 00:56:46,560
to Santa Monica. Okay. So where
do you look as a first

1059
00:56:46,560 --> 00:56:50,340
indication, you look like L.A.
potentially. Right. Yeah. So

1060
00:56:50,340 --> 00:56:54,377
right. And one pretty close. And
then once you get to L.A., you

1061
00:56:54,377 --> 00:56:58,157
look the road to get towards
Santa Monica. Yep. And the way

1062
00:56:58,157 --> 00:57:02,129
to make this efficient. Right.
You don't get all the detail to

1063
00:57:02,129 --> 00:57:05,589
in your mind right of that.
Yeah. You refine it to the

1064
00:57:05,589 --> 00:57:09,177
closest you get to your
destination. This is exactly how

1065
00:57:09,177 --> 00:57:13,021
Zenoh works. Okay. So imagine
that now you have a graph that

1066
00:57:13,021 --> 00:57:16,801
represents Zenoh nodes. Those
could be infrastructural node

1067
00:57:16,801 --> 00:57:20,645
like routers or application.
Yeah. And would you construct a

1068
00:57:20,645 --> 00:57:24,297
graph in which, like in the
analogy that I made with the,

1069
00:57:24,297 --> 00:57:28,013
you know, road indication where
the generalization of this

1070
00:57:28,013 --> 00:57:31,793
coding information. So like in
the case of the, of the, you

1071
00:57:31,793 --> 00:57:35,573
know, both a star cmdvl or both,
if you want to get the the

1072
00:57:35,573 --> 00:57:39,353
battery status of a specific
robot okay. Then every hope we

1073
00:57:39,353 --> 00:57:43,133
will know how to route your
request. And this gets refined.

1074
00:57:43,133 --> 00:57:46,465
And the closer we get to the
target, the more detail

1075
00:57:46,465 --> 00:57:50,117
information we have on how to
route it. I know this still

1076
00:57:50,117 --> 00:57:53,000
sounds vague, but that's really
how it works.

1077
00:57:53,000 --> 00:57:54,000
Audrow Nash: That's what you're
doing and.

1078
00:57:54,000 --> 00:57:57,823
Angelo Corsaro: This is how we
manage to really consolidate

1079
00:57:57,823 --> 00:58:01,647
discovery. You know, every
people often say, oh, we're from

1080
00:58:01,647 --> 00:58:05,600
DDS discovery. If we make an
analogy in DDS you have everyone

1081
00:58:05,600 --> 00:58:09,294
has to know everything about
everyone else. Yes. And then

1082
00:58:09,294 --> 00:58:13,441
just kind of escape. No. Right.
In Zenoh, the point is that, you

1083
00:58:13,441 --> 00:58:17,199
don't need to know anything
about everyone else. Maybe you

1084
00:58:17,199 --> 00:58:20,958
just need to know that if you
need anything from both one,

1085
00:58:20,958 --> 00:58:25,041
that's that way. And but two is
that way. That's the intuition.

1086
00:58:25,041 --> 00:58:28,670
Yeah. And then in terms of
mathematics, because there is

1087
00:58:28,670 --> 00:58:32,687
some mathematics because it's
all set theory. It's set theory.

1088
00:58:32,687 --> 00:58:36,576
Set theory, generalization.
There are also some arguing that

1089
00:58:36,576 --> 00:58:40,399
we use in terms of equivalence
between regular. So there is

1090
00:58:40,399 --> 00:58:44,352
some complex math behind. We are
then the result for the user

1091
00:58:44,352 --> 00:58:45,000
it's easy.

1092
00:58:45,000 --> 00:58:48,240
Audrow Nash: Yeah. So cool.
Yeah. That you just you have a

1093
00:58:48,240 --> 00:58:51,872
is it like it's almost heuristic
based. Go that way. First you're

1094
00:58:51,872 --> 00:58:55,000
going to go to LA first and then
you go to Santa Monica.

1095
00:58:55,000 --> 00:58:58,994
Angelo Corsaro: So you could use
a resource. But in our case it's

1096
00:58:58,994 --> 00:59:02,866
not touristic is generalizing. I
mean so in a way you can think

1097
00:59:02,866 --> 00:59:06,799
about it as every time, the more
you get further away from, from

1098
00:59:06,799 --> 00:59:10,056
the source of information,
right, the more you have a

1099
00:59:10,056 --> 00:59:13,866
generic description of the data
set that the reason behind it.

1100
00:59:13,866 --> 00:59:17,553
Yeah. Okay. And you also get it
a little. Yeah. Once in your

1101
00:59:17,553 --> 00:59:21,363
application that way. Right. And
if I, if all the information,

1102
00:59:21,363 --> 00:59:25,173
the detail information about one
application gets to me, I get

1103
00:59:25,173 --> 00:59:28,553
overwhelmed with data. If I
generalize this information

1104
00:59:28,553 --> 00:59:32,302
light and I get it distilled.
Birchfield. Yeah, right. I know

1105
00:59:32,302 --> 00:59:35,559
this sounds absurd, but
literally this is what we do.

1106
00:59:35,559 --> 00:59:39,000
Okay. So cool. And that's the
secret of the probability.

1107
00:59:39,000 --> 00:59:41,928
Audrow Nash: Yeah. And that's
why it's esthetically pleasing,

1108
00:59:41,928 --> 00:59:44,904
because you have this general
approach that lets you get these

1109
00:59:44,904 --> 00:59:47,640
very powerful results, and it
handles a lot of the, like,

1110
00:59:47,640 --> 00:59:50,376
tricky cases where I can
synchronize your data and things

1111
00:59:50,376 --> 00:59:51,000
like elegant.

1112
00:59:51,000 --> 00:59:53,000
Angelo Corsaro: And the user
doesn't need to think, right.

1113
00:59:53,000 --> 00:59:54,000
Audrow Nash: You know, a simple
interface.

1114
00:59:54,000 --> 00:59:57,556
Angelo Corsaro: Super simple
abstraction, and something else

1115
00:59:57,556 --> 01:00:01,407
that sometimes is missed. Right.
So we've been talking about, you

1116
01:00:01,407 --> 01:00:05,022
know, both one seem developed
that so usually across topic as

1117
01:00:05,022 --> 01:00:08,341
are called. Yeah. And they are
mapping to what we call Z

1118
01:00:08,341 --> 01:00:11,837
resources. So that also three.
And then you also hear I was

1119
01:00:11,837 --> 01:00:15,393
telling well we have you know
just a handful of byte of wide

1120
01:00:15,393 --> 01:00:19,007
overhead. So how does the tool
work? Because if you are using

1121
01:00:19,007 --> 01:00:22,800
string only that string is more
than a few bits. And there is no

1122
01:00:22,800 --> 01:00:26,415
the data. So in Zenoh there is
an automatic mapping of key to

1123
01:00:26,415 --> 01:00:30,089
small integers. Yeah. This is an
automatically and it's unique

1124
01:00:30,089 --> 01:00:33,881
within a session. So if you and
me are communicating I will come

1125
01:00:33,881 --> 01:00:37,496
up with a mapping that maybe I
will be using only with you or

1126
01:00:37,496 --> 01:00:40,993
maybe with everyone else. It
depends on concurrency, right.

1127
01:00:40,993 --> 01:00:44,489
Because maybe we map the same
key to different ID, which is

1128
01:00:44,489 --> 01:00:48,104
perfectly legal. And as long as
I know that it's your mapping

1129
01:00:48,104 --> 01:00:51,244
field. Yeah. And that's nice
because you get the word

1130
01:00:51,244 --> 01:00:54,919
efficiency and you get the user
experience. Yeah. Because, you

1131
01:00:54,919 --> 01:00:58,119
know, there were other
technologies in which you could

1132
01:00:58,119 --> 01:01:01,556
map integer to keys, but then
you had to as a user to keep

1133
01:01:01,556 --> 01:01:05,289
dynamic with your, which is not
scalable, you know, large scale

1134
01:01:05,289 --> 01:01:06,000
distributed.

1135
01:01:06,000 --> 01:01:08,000
Audrow Nash: So lots of errors.

1136
01:01:08,000 --> 01:01:11,051
Angelo Corsaro: Exactly. So once
again usability right. User

1137
01:01:11,051 --> 01:01:14,356
friendliness. But then the magic
behind to make it efficient. And

1138
01:01:14,356 --> 01:01:17,000
then that's taking you know that
ZetaScale approach.

1139
01:01:17,000 --> 01:01:21,823
Audrow Nash: Super cool. Yeah.
What problems explicitly is this

1140
01:01:21,823 --> 01:01:26,340
solving from DDS, because and
then also how like we're very

1141
01:01:26,340 --> 01:01:31,010
excited about this as a ROS 2
community. I would love to hear

1142
01:01:31,010 --> 01:01:33,000
a bit more about that too.

1143
01:01:33,000 --> 01:01:36,649
Angelo Corsaro: Yeah. So I think
there are a few probably a

1144
01:01:36,649 --> 01:01:40,423
couple of problems. First of
all, as of today in ROS there is

1145
01:01:40,423 --> 01:01:44,258
not a single solution to go from
the microcontroller up to the

1146
01:01:44,258 --> 01:01:45,000
data center.

1147
01:01:45,000 --> 01:01:46,000
Audrow Nash: Very real.

1148
01:01:45,000 --> 01:01:48,438
Angelo Corsaro: Why? Because we
have to use a protocol on the

1149
01:01:48,438 --> 01:01:51,876
microcontroller, which actually
is not DDS because Excel C is

1150
01:01:51,876 --> 01:01:55,315
another protocol, is a client
server. Probably. Then you have

1151
01:01:55,315 --> 01:01:58,584
DDS inside the robot and DDS
cannot cross the internet. So

1152
01:01:58,584 --> 01:02:02,191
you have to use something else.
And that's why you know there is

1153
01:02:02,191 --> 01:02:05,460
the Amputee Bridge. So that's
already nice because we will

1154
01:02:05,460 --> 01:02:08,898
have one protocol all you I use
end to end and that's correct

1155
01:02:08,898 --> 01:02:12,224
okay. So that's one problem. I
think it's very important in

1156
01:02:12,224 --> 01:02:14,985
terms of performance
streamlining monitoring your

1157
01:02:14,985 --> 01:02:16,000
system. All right.

1158
01:02:16,000 --> 01:02:21,000
Audrow Nash: Yes, definitely.

1159
01:02:21,000 --> 01:02:24,000
Angelo Corsaro: Yes. What I call
the digital Frankenstein.

1160
01:02:24,000 --> 01:02:27,024
Audrow Nash: Yeah. If it's like
sitting together, these are all

1161
01:02:27,024 --> 01:02:30,000
Frankenstein. Yeah. Right. Yeah.
Good way of putting it. Yeah.

1162
01:02:30,000 --> 01:02:33,735
Angelo Corsaro: And so that's
one problem. Second problem is

1163
01:02:33,735 --> 01:02:37,097
the discovery. Right. So we've
shown and this has been

1164
01:02:37,097 --> 01:02:41,082
reproduced by user. Even if you
take a simple turtle bot and you

1165
01:02:41,082 --> 01:02:44,630
make comparison with white
shark, we reduce off the shelf

1166
01:02:44,630 --> 01:02:46,000
discovery by over 90%.

1167
01:02:46,000 --> 01:02:48,000
Audrow Nash: Wow. That's huge.
That's amazing.

1168
01:02:48,000 --> 01:02:50,000
Angelo Corsaro: And again,
that's because the protocol

1169
01:02:50,000 --> 01:02:51,000
works very differently, as.

1170
01:02:51,000 --> 01:02:53,917
Audrow Nash: If it's
fundamentally different. Whereas

1171
01:02:53,917 --> 01:02:57,000
DDS has to know. Everybody has
to know about everything.

1172
01:02:57,000 --> 01:02:58,941
Angelo Corsaro: And again, you
tell me everything about

1173
01:02:58,941 --> 01:03:01,235
yourself. Me, I don't care. Just
tell me that you're weighing up.

1174
01:03:01,235 --> 01:03:03,000
And I know that if I need
something, I have to go.

1175
01:03:03,000 --> 01:03:04,000
Audrow Nash: Yes.

1176
01:03:04,000 --> 01:03:08,447
Angelo Corsaro: Yeah. That's I
what? It's like, you know? Cool,

1177
01:03:08,447 --> 01:03:10,000
man. Yeah. Cool. Okay.

1178
01:03:10,000 --> 01:03:12,000
Audrow Nash: So,

1179
01:03:12,000 --> 01:03:15,726
Angelo Corsaro: That's the
discovery point. We discussed

1180
01:03:15,726 --> 01:03:19,518
the connectivity. And then there
is another point. So the

1181
01:03:19,518 --> 01:03:23,576
quotable are being used today to
efficiently map services and

1182
01:03:23,576 --> 01:03:27,568
actions. So even today during
the presentation from Chris on

1183
01:03:27,568 --> 01:03:31,493
the to state, state of the House
is known that when you use

1184
01:03:31,493 --> 01:03:35,618
different device vendors, there
is no interoperability or they

1185
01:03:35,618 --> 01:03:39,810
have problem with services. Why?
Because of how this works. DDS

1186
01:03:39,810 --> 01:03:44,068
doesn't natively support remote
procedure. Call is built on top.

1187
01:03:44,068 --> 01:03:48,259
But again for how this works and
how they do this and specified

1188
01:03:48,259 --> 01:03:51,985
there are some gray area that
make it very hard light to

1189
01:03:51,985 --> 01:03:56,110
append or set the same behavior
in Zenoh queryable or built in

1190
01:03:56,110 --> 01:04:00,301
Queryable is a generalization of
RPC, okay, and that is used to

1191
01:04:00,301 --> 01:04:04,359
very efficiently map services,
increase scalability, increase

1192
01:04:04,359 --> 01:04:08,551
the terms. Do you think that's
that's quite important. And then

1193
01:04:08,551 --> 01:04:12,875
something else that is not there
yet. But I mean at least I would

1194
01:04:12,875 --> 01:04:16,601
really like to see it soon is
true. Zero copy in ROS. So

1195
01:04:16,601 --> 01:04:17,000
Zenoh.

1196
01:04:17,000 --> 01:04:18,000
Audrow Nash: As I was.

1197
01:04:17,000 --> 01:04:19,000
Angelo Corsaro: Saying, was
built in support or.

1198
01:04:19,000 --> 01:04:20,000
Audrow Nash: Okay.

1199
01:04:20,000 --> 01:04:21,000
Angelo Corsaro: Yeah, it's in
us. No.

1200
01:04:21,000 --> 01:04:23,000
Audrow Nash: Across processes.
Oh, okay. That would be a maybe.

1201
01:04:23,000 --> 01:04:26,429
Angelo Corsaro: Yes. Xen was
built in support for shared

1202
01:04:26,429 --> 01:04:30,408
memory. Yep. With a 1.0. We also
have now the ability of plugging

1203
01:04:30,408 --> 01:04:34,204
in shared memory provider. So if
you have custom hardware like

1204
01:04:34,204 --> 01:04:35,000
you know GPU.

1205
01:04:35,000 --> 01:04:37,000
Audrow Nash: Yeah yeah. And V or
something, you.

1206
01:04:37,000 --> 01:04:40,278
Angelo Corsaro: Can plug a
provider for that shared memory,

1207
01:04:40,278 --> 01:04:43,778
allocate out of that buffer. And
then if you're using Zenoh all

1208
01:04:43,778 --> 01:04:47,167
this transparent. Yep. So we
would like to have you know this

1209
01:04:47,167 --> 01:04:49,000
model expose also to ROS
program.

1210
01:04:49,000 --> 01:04:50,000
Audrow Nash: That's awesome.

1211
01:04:49,000 --> 01:04:52,200
Angelo Corsaro: Because then we
could really have true zero copy

1212
01:04:52,200 --> 01:04:54,000
zero serialization could be
garbage.

1213
01:04:54,000 --> 01:04:56,784
Audrow Nash: It would be
gorgeous. Yeah I love it. Oh,

1214
01:04:56,784 --> 01:04:59,000
yeah. You clearly loved us too,
which is I.

1215
01:04:59,000 --> 01:05:02,077
Angelo Corsaro: Do wonder if. I
mean, that's my passion. And

1216
01:05:02,077 --> 01:05:05,256
then there are some other stuff
we are doing around video. One

1217
01:05:05,256 --> 01:05:07,000
video to facilitate. Yes. Oh
wait.

1218
01:05:07,000 --> 01:05:08,000
Audrow Nash: I don't know if
that's.

1219
01:05:08,000 --> 01:05:10,000
Angelo Corsaro: Very close
somehow or two. Yeah.

1220
01:05:10,000 --> 01:05:12,000
Audrow Nash: Okay. I'd love to
hear a little bit about that.

1221
01:05:12,000 --> 01:05:14,857
Angelo Corsaro: Yeah. I mean
again it's facilitating the

1222
01:05:14,857 --> 01:05:17,561
sending of transparent
compression because we already

1223
01:05:17,561 --> 01:05:20,673
do transparent compression in
Xeno and it's just going beyond

1224
01:05:20,673 --> 01:05:22,000
and supporting that naked.

1225
01:05:22,000 --> 01:05:24,074
Audrow Nash: Yeah. Because video
has its own compression

1226
01:05:24,074 --> 01:05:25,000
algorithms and there are.

1227
01:05:25,000 --> 01:05:26,000
Angelo Corsaro: Some.

1228
01:05:25,000 --> 01:05:27,000
Audrow Nash: Peculiarities that,
you know definitely.

1229
01:05:27,000 --> 01:05:29,000
Angelo Corsaro: You will see
something compressed. So.

1230
01:05:29,000 --> 01:05:33,000
Audrow Nash: Oh I can't wait.
One, one thing I would also like

1231
01:05:33,000 --> 01:05:35,000
to hear about. So Zenoh, Zenoh.

1232
01:05:35,000 --> 01:05:37,000
Angelo Corsaro: You know, well
you could use both.

1233
01:05:37,000 --> 01:05:38,000
Audrow Nash: Okay.

1234
01:05:37,000 --> 01:05:40,789
Angelo Corsaro: So should we
talk about this. Sure. Okay. So

1235
01:05:40,789 --> 01:05:43,000
the the name of the protocol.
Yeah.

1236
01:05:43,000 --> 01:05:43,913
Audrow Nash: Where did it come
from? You were mentioning at the

1237
01:05:43,913 --> 01:05:44,000
start.

1238
01:05:44,000 --> 01:05:48,958
Angelo Corsaro: The acronym
means zero network overhead.

1239
01:05:48,958 --> 01:05:53,828
Okay. Now the H y so the H is
the O and H. So here zero

1240
01:05:53,828 --> 01:05:59,052
network overhead okay. Now in
reality the problem had to be

1241
01:05:59,052 --> 01:06:01,000
called like this just.

1242
01:06:01,000 --> 01:06:02,000
Audrow Nash: In.

1243
01:06:01,000 --> 01:06:04,716
Angelo Corsaro: Or with an N
okay. Yeah. But with an n sounds

1244
01:06:04,716 --> 01:06:08,492
kind of strange. Those people
call it Zenohn. Yeah, but that's

1245
01:06:08,492 --> 01:06:12,208
not that you would pronounce.
And the domain wasn't available

1246
01:06:12,208 --> 01:06:13,000
either. Okay?

1247
01:06:13,000 --> 01:06:16,000
Audrow Nash: It's fragmented,
like in our minds. You need it.

1248
01:06:16,000 --> 01:06:20,116
Angelo Corsaro: The question is
why Zenoh? So there are two

1249
01:06:20,116 --> 01:06:23,884
philosophers that are very close
to my heart okay? One

1250
01:06:23,884 --> 01:06:27,442
pre-Socratic. That was a
philosopher about the, the

1251
01:06:27,442 --> 01:06:31,488
paradox with things that were
getting almost close to zero

1252
01:06:31,488 --> 01:06:34,000
zero network is not possible,
right?

1253
01:06:34,000 --> 01:06:36,000
Audrow Nash: So that was it.
Yeah. Very poetic.

1254
01:06:36,000 --> 01:06:39,698
Angelo Corsaro: Right. And, I
mean, he's a he's a pretty

1255
01:06:39,698 --> 01:06:43,858
incredible philosopher, which,
you know. Oh, yeah, I like a lot

1256
01:06:43,858 --> 01:06:48,019
and know, kind of inspired this
idea of getting as close as you

1257
01:06:48,019 --> 01:06:51,783
end up. I was yeah, in the
second one, which I think it's

1258
01:06:51,783 --> 01:06:55,877
also a source of inspiration for
me is the father of stoicism.

1259
01:06:55,877 --> 01:06:59,708
Oh, okay. Oh, yeah. And which,
you know, is all about, you

1260
01:06:59,708 --> 01:07:03,802
know, the perseverance, doing
things properly right and making

1261
01:07:03,802 --> 01:07:04,000
if.

1262
01:07:04,000 --> 01:07:07,000
Audrow Nash: Other good work.
Exactly.

1263
01:07:07,000 --> 01:07:10,069
Angelo Corsaro: Because that's
what give you. Oh where is

1264
01:07:10,069 --> 01:07:13,246
essentially, you know, the
accomplishment and the happiness

1265
01:07:13,246 --> 01:07:14,000
in life. Yeah.

1266
01:07:14,000 --> 01:07:16,000
Audrow Nash: Yeah. So I think
too, it's all process.

1267
01:07:16,000 --> 01:07:19,265
Angelo Corsaro: It is. Yeah. And
again is this commitment to

1268
01:07:19,265 --> 01:07:22,639
always improve. Yep. And I think
you know that's that's what's

1269
01:07:22,639 --> 01:07:24,000
beautiful right. Oh yeah.

1270
01:07:24,000 --> 01:07:27,000
Audrow Nash: So the father of
stoicism then. Don't know.

1271
01:07:27,000 --> 01:07:30,000
Angelo Corsaro: You know and
then obviously for me.

1272
01:07:30,000 --> 01:07:31,000
Audrow Nash: It's a really a.

1273
01:07:30,000 --> 01:07:33,451
Angelo Corsaro: Marcus.
Absolutely. Yeah. But the

1274
01:07:33,451 --> 01:07:35,000
founder is isn't cool.

1275
01:07:35,000 --> 01:07:37,000
Audrow Nash: I love those old.

1276
01:07:37,000 --> 01:07:41,000
Angelo Corsaro: If you read the
meditations. Great. Right. Yeah.

1277
01:07:41,000 --> 01:07:43,408
Audrow Nash: Yeah. Amazing. And
he didn't even intend to write

1278
01:07:43,408 --> 01:07:45,000
it for anyone. Yeah. Just so
crazy. Yeah.

1279
01:07:45,000 --> 01:07:47,232
Angelo Corsaro: Yeah. If you
haven't read it, you should read

1280
01:07:47,232 --> 01:07:48,000
the meditations. Yes.

1281
01:07:48,000 --> 01:07:51,986
Audrow Nash: Yeah. Recommend it.
Yeah. So that's super cool.

1282
01:07:51,986 --> 01:07:55,973
Okay. And then I'd love to hear
a bit about the like. So you

1283
01:07:55,973 --> 01:08:00,026
guys are using rust for. Yeah,
absolutely. Rust is awesome. I

1284
01:08:00,026 --> 01:08:03,945
really like rust. It's not
something that's been common for

1285
01:08:03,945 --> 01:08:07,998
ROS 2 So we're having to add, to
support you as, like a first

1286
01:08:07,998 --> 01:08:12,118
class middleware. We are having
to add a lot of tooling around

1287
01:08:12,118 --> 01:08:15,905
rust, because rust has a very
different philosophy of how

1288
01:08:15,905 --> 01:08:20,090
package management work. Two C
plus plus Python. So I'd love to

1289
01:08:20,090 --> 01:08:24,077
hear a little bit about why
rust, why rust, and then kind of

1290
01:08:24,077 --> 01:08:27,000
how it's been going to integrate
in with ROS

1291
01:08:27,000 --> 01:08:31,067
Angelo Corsaro: Okay. Yeah. So
the first version of the zero,

1292
01:08:31,067 --> 01:08:35,000
the very first versions was
written in OCaml. Have you ever

1293
01:08:35,000 --> 01:08:36,000
heard of OCaml?

1294
01:08:36,000 --> 01:08:37,000
Audrow Nash: Yeah.

1295
01:08:36,000 --> 01:08:39,958
Angelo Corsaro: Okay. So of
course we're based in France,

1296
01:08:39,958 --> 01:08:41,000
right? So yeah.

1297
01:08:41,000 --> 01:08:44,010
Audrow Nash: I don't know too
much about OCaml. Okay. Well was

1298
01:08:44,010 --> 01:08:46,000
it kind of like TypeScript in
some sense?

1299
01:08:46,000 --> 01:08:49,409
Angelo Corsaro: So OCaml is a
variation of EML, which is one

1300
01:08:49,409 --> 01:08:52,705
of the key, you know. Yeah. And
the if you think about the

1301
01:08:52,705 --> 01:08:56,057
programing language. Yeah. On
one side you have Lisp. Yeah.

1302
01:08:56,057 --> 01:08:59,636
And it's like one family on the
other side you have ML and it's

1303
01:08:59,636 --> 01:09:03,273
another family. Okay. Okay. And
then the most popular version of

1304
01:09:03,273 --> 01:09:06,000
ML, is actually OCaml, which was
done in France.

1305
01:09:06,000 --> 01:09:08,000
Audrow Nash: And that's the ML
in it, maybe. Oh, camel.

1306
01:09:08,000 --> 01:09:12,833
Angelo Corsaro: Okay. Yeah. And
okay, so, OCaml again, why you

1307
01:09:12,833 --> 01:09:17,587
can because we care immensely
about security. Okay. So OCaml,

1308
01:09:17,587 --> 01:09:22,420
obviously it's a language for
which there are I mean, there is

1309
01:09:22,420 --> 01:09:26,629
a very sound type system,
memories managed. And by the

1310
01:09:26,629 --> 01:09:31,228
way, there are things like coq,
which, you can use to prove

1311
01:09:31,228 --> 01:09:35,749
certain algorithms. So in the
very first implementation of

1312
01:09:35,749 --> 01:09:40,504
internal with proved certain
element with Coq okay. Is France

1313
01:09:40,504 --> 01:09:45,414
okay. And that around us we have
lots of formal method guys. So

1314
01:09:45,414 --> 01:09:49,000
we had some elaboration with say
was gorgeous.

1315
01:09:49,000 --> 01:09:50,000
Audrow Nash: Yeah.

1316
01:09:50,000 --> 01:09:53,089
Angelo Corsaro: We also were a
smaller team. And you know, we

1317
01:09:53,089 --> 01:09:54,000
need productivity.

1318
01:09:54,000 --> 01:09:55,000
Audrow Nash: Yeah. So you.

1319
01:09:55,000 --> 01:09:58,668
Angelo Corsaro: Floyd, on the
network, we want to make sure

1320
01:09:58,668 --> 01:10:02,335
that, you know, we don't have,
70% of the source of attack,

1321
01:10:02,335 --> 01:10:05,816
which are related to memory
mismanagement ourselves. And

1322
01:10:05,816 --> 01:10:09,484
there were some uni here. And
really, focus on on security,

1323
01:10:09,484 --> 01:10:13,089
build on OCaml. And that could
also come up. So we will be

1324
01:10:13,089 --> 01:10:16,695
closing Zenoh also as unique
kernel as network appliances,

1325
01:10:16,695 --> 01:10:18,000
which was super cool.

1326
01:10:18,000 --> 01:10:19,000
Audrow Nash: Yeah.

1327
01:10:19,000 --> 01:10:23,351
Angelo Corsaro: Then we start to
pick up and especially we got

1328
01:10:23,351 --> 01:10:27,000
users in telecommunication,
which they told us guys.

1329
01:10:27,000 --> 01:10:28,000
Audrow Nash: Okay, right.

1330
01:10:27,000 --> 01:10:30,727
Angelo Corsaro: No channel is
nice. But you know, besides in

1331
01:10:30,727 --> 01:10:34,143
France, in Yale, in Cambridge
go. Where do you find the

1332
01:10:34,143 --> 01:10:37,000
academy program, etc. in the
financial market?

1333
01:10:37,000 --> 01:10:40,218
Audrow Nash: Oh yeah. Yeah,
they're very expensive program.

1334
01:10:40,218 --> 01:10:43,000
Oh I bet yeah. Because they're
coming from finance.

1335
01:10:43,000 --> 01:10:45,753
Angelo Corsaro: And so they said
you have to move to an end to a

1336
01:10:45,753 --> 01:10:47,000
in other programing language.

1337
01:10:47,000 --> 01:10:48,000
Audrow Nash: Because rust of
all.

1338
01:10:48,000 --> 01:10:51,629
Angelo Corsaro: The so actually
rust we had already evaluate it.

1339
01:10:51,629 --> 01:10:55,088
Yeah. Whenever we did different
underwriting. Well we did the

1340
01:10:55,088 --> 01:10:58,206
first writing of of Zenoh, but
that was before the 2018

1341
01:10:58,206 --> 01:10:59,000
edition. Okay.

1342
01:10:59,000 --> 01:11:01,140
Audrow Nash: Did it come a long
way? Because I think over the

1343
01:11:01,140 --> 01:11:03,000
last, like, you know, our six
years, it's really for.

1344
01:11:03,000 --> 01:11:06,853
Angelo Corsaro: 2018 edition, we
felt, you know, it was too much

1345
01:11:06,853 --> 01:11:10,706
of a stretch. And then you know,
when we started the open source

1346
01:11:10,706 --> 01:11:14,559
project and we did the rewrite,
it was just before 2018, but the

1347
01:11:14,559 --> 01:11:18,351
prerelease of the 2018 was there
within evaluation. And we went

1348
01:11:18,351 --> 01:11:21,000
for rust and honestly, it has
been gorgeous.

1349
01:11:21,000 --> 01:11:22,000
Audrow Nash: Rust is great.

1350
01:11:22,000 --> 01:11:24,824
Angelo Corsaro: Memory safe,
concurrency safe. So we even got

1351
01:11:24,824 --> 01:11:27,000
a guaranteed an OCaml. Much
better performance.

1352
01:11:27,000 --> 01:11:28,000
Audrow Nash: Oh so bad.

1353
01:11:28,000 --> 01:11:31,057
Angelo Corsaro: So very thick
tooling ecosystem. Yep. Right.

1354
01:11:31,057 --> 01:11:34,217
And the other reason why we are
chosen last is because you can

1355
01:11:34,217 --> 01:11:36,000
run bare metal on
microcontrollers.

1356
01:11:36,000 --> 01:11:37,000
Audrow Nash: Yes. Very true.

1357
01:11:37,000 --> 01:11:40,351
Angelo Corsaro: We have already.
If you look in our code base,

1358
01:11:40,351 --> 01:11:43,486
there are bits and pieces of
zero that are already no SDD.

1359
01:11:43,486 --> 01:11:46,946
Wow. So eventually you know you
will have the whole thing. Yeah,

1360
01:11:46,946 --> 01:11:49,000
yeah. Quorums in on city. Do it.
Yeah.

1361
01:11:49,000 --> 01:11:51,213
Audrow Nash: That's awesome. So
yeah you can support

1362
01:11:51,213 --> 01:11:53,936
microcontrollers very easily in
the same code base. And you just

1363
01:11:53,936 --> 01:11:55,000
have like the code paths.

1364
01:11:55,000 --> 01:11:58,273
Angelo Corsaro: So in today we
support with C. But eventually

1365
01:11:58,273 --> 01:12:01,385
because zero microcontroller
that is still heavily area is

1366
01:12:01,385 --> 01:12:04,820
C++ ecosystem. But we have more
and more requests about rust and

1367
01:12:04,820 --> 01:12:06,000
will be getting there.

1368
01:12:06,000 --> 01:12:09,304
Audrow Nash: Yeah yeah I've
tried rust for microcontroller

1369
01:12:09,304 --> 01:12:12,778
programing and it was like a
little difficult as a as process

1370
01:12:12,778 --> 01:12:15,000
harder than C. Well yeah. For
this guy.

1371
01:12:15,000 --> 01:12:17,634
Angelo Corsaro: They are getting
there. But whereas an

1372
01:12:17,634 --> 01:12:20,415
interesting project. Yeah I
began this credence with rust

1373
01:12:20,415 --> 01:12:23,341
has been quite terrific so far.
Yes. Obviously the additions

1374
01:12:23,341 --> 01:12:26,220
continue to to evolve. We were
probably one of the earliest

1375
01:12:26,220 --> 01:12:27,000
user of I think.

1376
01:12:27,000 --> 01:12:29,000
Audrow Nash: Oh yeah. Now has
come a long way.

1377
01:12:29,000 --> 01:12:32,478
Angelo Corsaro: That has come a
long way. And I think we have we

1378
01:12:32,478 --> 01:12:35,848
have grown with it. Okay. You
know, I don't regret the change.

1379
01:12:35,848 --> 01:12:39,000
I think the team is active in
terms of productivity price.

1380
01:12:39,000 --> 01:12:41,000
Audrow Nash: Yeah, yeah, yeah.
Rust is great. Yeah.

1381
01:12:41,000 --> 01:12:44,263
Angelo Corsaro: And then, you
know, in terms of integration in

1382
01:12:44,263 --> 01:12:47,211
ROS I mean I would say great
thanks to the open robotics

1383
01:12:47,211 --> 01:12:50,421
team. Right. Because they are
doing the heavy lifting for us.

1384
01:12:50,421 --> 01:12:53,158
I mean, they are pulling out,
Zenoh as it is. Right,

1385
01:12:53,158 --> 01:12:55,000
integrating it. So in that
respect.

1386
01:12:55,000 --> 01:12:57,224
Audrow Nash: And all the build
farm work, which I think is like

1387
01:12:57,224 --> 01:12:58,000
the bigger part. Yeah.

1388
01:12:58,000 --> 01:13:00,889
Angelo Corsaro: Of all I know
that, you know, folks like

1389
01:13:00,889 --> 01:13:03,881
Steven would like to change a
few things, and in due time,

1390
01:13:03,881 --> 01:13:06,821
they will. Yeah. But I think
it's working fine right now,

1391
01:13:06,821 --> 01:13:09,762
and, you know. Great. Thanks
again to the work. Very open

1392
01:13:09,762 --> 01:13:11,000
robotics for that. Yeah.

1393
01:13:11,000 --> 01:13:14,991
Audrow Nash: Oh, yeah. So where
do you guys where do you think

1394
01:13:14,991 --> 01:13:19,112
you'll be? Or like when you look
forward for the next, like, say

1395
01:13:19,112 --> 01:13:23,167
1 to 3 years? Yeah. Where do you
think you guys are headed? And

1396
01:13:23,167 --> 01:13:26,000
like, what do you hope for is at
a scale so.

1397
01:13:26,000 --> 01:13:29,046
Angelo Corsaro: Well, first of
all, continue to do great

1398
01:13:29,046 --> 01:13:32,146
innovation, great work in
addressing user problem because

1399
01:13:32,146 --> 01:13:35,519
I mean, I always say, you know,
to to all the settlers, that's

1400
01:13:35,519 --> 01:13:38,565
how we pull ourself. We are
here. Is that because. Yeah,

1401
01:13:38,565 --> 01:13:39,000
because.

1402
01:13:39,000 --> 01:13:42,000
Audrow Nash: Of our Atlas. Is
that correct? Yes. Love it.

1403
01:13:42,000 --> 01:13:44,000
Angelo Corsaro: Because of our
users.

1404
01:13:44,000 --> 01:13:45,000
Audrow Nash: Right.

1405
01:13:44,000 --> 01:13:47,992
Angelo Corsaro: So our goal is
to solve their problems. And you

1406
01:13:47,992 --> 01:13:51,667
know, we have grown, but we
don't want to grow like crazy.

1407
01:13:51,667 --> 01:13:54,961
We still want to grow
organically. Yes. Maintain our

1408
01:13:54,961 --> 01:13:58,383
culture, maintain, you know,
very high quality. We are

1409
01:13:58,383 --> 01:14:02,058
expanding in new domains. So we
are it's probably will get

1410
01:14:02,058 --> 01:14:05,733
public. But besides automotive,
in robotics, we are seeing

1411
01:14:05,733 --> 01:14:07,000
application medical.

1412
01:14:07,000 --> 01:14:08,000
Audrow Nash: Definitely.

1413
01:14:08,000 --> 01:14:09,000
Angelo Corsaro: A Traffic
Control management. Yeah.

1414
01:14:09,000 --> 01:14:13,000
Audrow Nash: In microcontrollers
and computer.

1415
01:14:13,000 --> 01:14:16,081
Angelo Corsaro: And by the way,
native you don't know, but we

1416
01:14:16,081 --> 01:14:18,000
already deployed in three
blockchains.

1417
01:14:18,000 --> 01:14:19,000
Audrow Nash: Wow.

1418
01:14:19,000 --> 01:14:20,000
Angelo Corsaro: Okay.

1419
01:14:19,000 --> 01:14:22,486
Audrow Nash: So cool. So that's
really cool. You're seeing it

1420
01:14:22,486 --> 01:14:23,000
in, soon.

1421
01:14:23,000 --> 01:14:26,245
Angelo Corsaro: Yeah. So I think
we'll continue to expand in

1422
01:14:26,245 --> 01:14:29,490
terms of market and scale will
continue to organically grow.

1423
01:14:29,490 --> 01:14:32,952
You know, we have now a platform
we are going for certification.

1424
01:14:32,952 --> 01:14:36,197
And yeah. So I think that now
that the scale, you know, it's

1425
01:14:36,197 --> 01:14:39,334
kind of felt like the new
innovative around the block. And

1426
01:14:39,334 --> 01:14:42,579
I hope that in three years it
will be understood that we are

1427
01:14:42,579 --> 01:14:45,770
the place to go. Right? Oh,
yeah. We get people that listen

1428
01:14:45,770 --> 01:14:48,907
to you that like to generally
solve problem and that, as I

1429
01:14:48,907 --> 01:14:52,207
always like to say, there will
be here in good and bad times,

1430
01:14:52,207 --> 01:14:55,344
but that's a big difference.
Yeah. And that's also stasis,

1431
01:14:55,344 --> 01:14:58,805
right. The big difference is you
know who you can count when our

1432
01:14:58,805 --> 01:15:02,105
time counts. And I think that
because we know we want to know

1433
01:15:02,105 --> 01:15:05,404
where is challenge. Yeah. And
you want to choose your partner

1434
01:15:05,404 --> 01:15:08,000
appropriately. And we are one of
those partners.

1435
01:15:08,000 --> 01:15:10,000
Audrow Nash: Oh yeah I love it.
All right. Thank you very.

1436
01:15:10,000 --> 01:15:18,000
Angelo Corsaro: Much. My
pleasure. Thank you very much.
