[Script Info]
Title: 
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.00,0:00:18.01,Default,,0000,0000,0000,,{\i1}36C3 preroll music{\i0}
Dialogue: 0,0:00:18.48,0:00:25.05,Default,,0000,0000,0000,,Herald: So Sasha is a doctor with a weak\Nspot for LEDs and, completely abstains
Dialogue: 0,0:00:25.05,0:00:30.54,Default,,0000,0000,0000,,from HDMI adapters these days. He wanted\Nto share with us the experience of
Dialogue: 0,0:00:30.54,0:00:36.36,Default,,0000,0000,0000,,attempting to build a delivery robots in\Nthe past 2.5 years in the Bay Area. And
Dialogue: 0,0:00:36.36,0:00:38.53,Default,,0000,0000,0000,,so, yeah, let's give him a big welcome.
Dialogue: 0,0:00:46.42,0:00:51.66,Default,,0000,0000,0000,,Sasha: Thank you, Mikael. So just a show\Nof hands who here has built robots before?
Dialogue: 0,0:00:52.53,0:00:56.05,Default,,0000,0000,0000,,Well, it's quite a few people. What about\Nautonomous robots, anybody built
Dialogue: 0,0:00:56.05,0:01:03.36,Default,,0000,0000,0000,,autonomous robots? Still quite a few\Npeople. Well, today I'm gonna be sharing
Dialogue: 0,0:01:03.36,0:01:09.15,Default,,0000,0000,0000,,with you the story of how not to build\Nautonomous robots. Over the course of the
Dialogue: 0,0:01:09.15,0:01:12.69,Default,,0000,0000,0000,,past two and a half years together with my\Nteam, we built the world's largest robotic
Dialogue: 0,0:01:12.69,0:01:17.74,Default,,0000,0000,0000,,delivery infrastructure. We went from a\Nconcept sketch to a commercially viable
Dialogue: 0,0:01:17.74,0:01:22.52,Default,,0000,0000,0000,,service running in three cities. We've had\Nlots of successes and one or two failures.
Dialogue: 0,0:01:27.20,0:01:29.95,Default,,0000,0000,0000,,So over the course of the next 45 minutes,\NI'm going to be sharing with you a couple
Dialogue: 0,0:01:29.95,0:01:33.41,Default,,0000,0000,0000,,of different stories. First of all, I'm\Ngoing to briefly introduce myself and I'm
Dialogue: 0,0:01:33.41,0:01:37.28,Default,,0000,0000,0000,,going to share the story of how we built\Nrobots, the different prototypes we had,
Dialogue: 0,0:01:37.28,0:01:40.82,Default,,0000,0000,0000,,the different iterations that we tried.\NThen I'm going to jump on manufacturing.
Dialogue: 0,0:01:40.82,0:01:44.34,Default,,0000,0000,0000,,We actually went to China and scaled up\Nour manufacturing, our production line.
Dialogue: 0,0:01:44.34,0:01:47.88,Default,,0000,0000,0000,,I'm gonna share with you the story of how\Nhe did that. And finally, I'm gonna talk
Dialogue: 0,0:01:47.88,0:01:52.32,Default,,0000,0000,0000,,about A.I. and all the magic that is\Nartificial intelligence. So we'll be able
Dialogue: 0,0:01:52.32,0:01:58.01,Default,,0000,0000,0000,,to see how we were able to crack that\Npuzzle. So without further ado, let's do
Dialogue: 0,0:01:58.01,0:02:04.38,Default,,0000,0000,0000,,the introduction. This is me. Right here.\NI like to build things. I built my first
Dialogue: 0,0:02:04.38,0:02:08.16,Default,,0000,0000,0000,,website when I was eleven and I built my\Nfirst business when I was 13. I was the
Dialogue: 0,0:02:08.16,0:02:11.95,Default,,0000,0000,0000,,iPhone repair business that I was running\Nin my bedroom. I've been really, really
Dialogue: 0,0:02:11.95,0:02:16.17,Default,,0000,0000,0000,,passionate about building things and over\Nthe course of many years I built a couple
Dialogue: 0,0:02:16.17,0:02:19.98,Default,,0000,0000,0000,,of different startups. One of them was a\Nfood delivery platform. We ended up
Dialogue: 0,0:02:19.98,0:02:24.45,Default,,0000,0000,0000,,running three different cities and doing\Nhundreds of deliveries a day. By the time
Dialogue: 0,0:02:24.45,0:02:29.59,Default,,0000,0000,0000,,I was 19. So I got to experience startups\Npretty early on. I've been really enjoying
Dialogue: 0,0:02:29.59,0:02:33.10,Default,,0000,0000,0000,,that time. After this food delivery\Nstartup failed I went to some
Dialogue: 0,0:02:33.10,0:02:37.73,Default,,0000,0000,0000,,cryptocurrency startups and then went to\Nwork for big corporations. And that's
Dialogue: 0,0:02:37.73,0:02:44.90,Default,,0000,0000,0000,,actually very boring. I dorned my office\Nwith some supplementary graphics. After a
Dialogue: 0,0:02:44.90,0:02:47.79,Default,,0000,0000,0000,,while, I got a little bit bored of this\Ncorporate life. It wasn't really for me.
Dialogue: 0,0:02:47.79,0:02:51.67,Default,,0000,0000,0000,,So I decided to get a one way ticket to\NSan Francisco. So I ended up in San
Dialogue: 0,0:02:51.67,0:02:56.74,Default,,0000,0000,0000,,Francisco staying on a friend's couch, not\Nreally knowing anybody. And I was really
Dialogue: 0,0:02:56.74,0:03:00.59,Default,,0000,0000,0000,,fortunate to be introduced to an\Nincredible group of people. And over the
Dialogue: 0,0:03:00.59,0:03:07.34,Default,,0000,0000,0000,,course of about two and a half years, we\Nstarted to take a concepts. A scatch that
Dialogue: 0,0:03:07.34,0:03:10.97,Default,,0000,0000,0000,,we had and we built up a robot. At first\Nit was something that barely even worked.
Dialogue: 0,0:03:10.97,0:03:14.38,Default,,0000,0000,0000,,But then we gradually got to something\Nthat worked a little bit better and better
Dialogue: 0,0:03:14.38,0:03:18.95,Default,,0000,0000,0000,,and better. After a while, we actually\Nmanaged to build a whole fleet of robots.
Dialogue: 0,0:03:18.95,0:03:23.68,Default,,0000,0000,0000,,I think at the peak we had 150 robots. So\Nit was a really, really cool experience.
Dialogue: 0,0:03:23.68,0:03:27.13,Default,,0000,0000,0000,,And during that time, I got to meet the\Nlieutenant governor of California, how to
Dialogue: 0,0:03:27.13,0:03:31.21,Default,,0000,0000,0000,,figure out how to do manufacturing in\NChina and most importantly, work with an
Dialogue: 0,0:03:31.21,0:03:35.94,Default,,0000,0000,0000,,incredible team which who I had a lot of\Nfun with building these robots. So, yeah,
Dialogue: 0,0:03:35.94,0:03:40.18,Default,,0000,0000,0000,,it's a little bit about me and what we\Nwere building. And maybe now we can jump
Dialogue: 0,0:03:40.18,0:03:45.73,Default,,0000,0000,0000,,in to how not to build robots. So this is\Nour very first robot,this is a really
Dialogue: 0,0:03:45.73,0:03:51.00,Default,,0000,0000,0000,,small prototype. We built is basically a\Nshopping basket on wheels. There is a RC
Dialogue: 0,0:03:51.00,0:03:55.64,Default,,0000,0000,0000,,car there below, there's a shopping basket\Nand Arduino Raspberry Pi. The thing barely
Dialogue: 0,0:03:55.64,0:03:59.55,Default,,0000,0000,0000,,work. Honestly, it was really, really\Nhacky. And what ended up happening is that
Dialogue: 0,0:03:59.55,0:04:03.75,Default,,0000,0000,0000,,most of the time we just dropped off the\Nrobot in front of the customer that
Dialogue: 0,0:04:03.75,0:04:06.68,Default,,0000,0000,0000,,literally just dropped it in front of the\Ndoor just to see if they would like order
Dialogue: 0,0:04:06.68,0:04:11.19,Default,,0000,0000,0000,,food with robots. The answer was\Noverwhelmingly yes. So we decided to spend
Dialogue: 0,0:04:11.19,0:04:15.48,Default,,0000,0000,0000,,some more time building our technology.\NThere is a small - I don't know if you can
Dialogue: 0,0:04:15.48,0:04:18.51,Default,,0000,0000,0000,,see it here. Yeah, there you go. There's a\Nsmall orange holder, that's actually a
Dialogue: 0,0:04:18.51,0:04:23.78,Default,,0000,0000,0000,,phone holder. So our very first prototype,\Nit had a phone sitting on top of it doing
Dialogue: 0,0:04:23.78,0:04:30.66,Default,,0000,0000,0000,,a video call so that somebody can remotely\Ncontrol it from Colombia. So we really
Dialogue: 0,0:04:30.66,0:04:34.29,Default,,0000,0000,0000,,started out small, really humble just to\Nsee if it would work. And that's something
Dialogue: 0,0:04:34.29,0:04:38.46,Default,,0000,0000,0000,,that we did a lot of this being really\Nresourceful in terms of trying out things.
Dialogue: 0,0:04:38.46,0:04:42.53,Default,,0000,0000,0000,,For about a year of this, we moved on to\Nsomething that looks a little bit more
Dialogue: 0,0:04:42.53,0:04:45.83,Default,,0000,0000,0000,,like this. So we started playing around\Nwith the shape. We start playing around
Dialogue: 0,0:04:45.83,0:04:50.28,Default,,0000,0000,0000,,with the design. We noticed that people\Nresponded really positively to faces and
Dialogue: 0,0:04:50.28,0:04:56.67,Default,,0000,0000,0000,,to like things that looked like people. So\Nwe actually built in a face. So we took
Dialogue: 0,0:04:56.67,0:05:00.96,Default,,0000,0000,0000,,this little animation that we built and we\Nput it onto the robot and there's actually
Dialogue: 0,0:05:00.96,0:05:05.62,Default,,0000,0000,0000,,really, really positive. We had a lot of\Ngood responses from the community, a lot
Dialogue: 0,0:05:05.62,0:05:09.98,Default,,0000,0000,0000,,of great feedback. And what we've seen is\Nthat people really love to have robots
Dialogue: 0,0:05:09.98,0:05:13.29,Default,,0000,0000,0000,,that are kind of friendly. There was\Nanother company that deployed robots that
Dialogue: 0,0:05:13.29,0:05:17.67,Default,,0000,0000,0000,,looked like vending machines or almost\Nlike tanks in San Francisco, and they got
Dialogue: 0,0:05:17.67,0:05:21.67,Default,,0000,0000,0000,,banned really, really quickly. So we\Ndecided that we would do our best to make
Dialogue: 0,0:05:21.67,0:05:26.62,Default,,0000,0000,0000,,sure our robots were as friendly as\Npossible instead of threatening and scary.
Dialogue: 0,0:05:26.62,0:05:32.18,Default,,0000,0000,0000,,So that was a very important part of it.\NAfter another year, we ended up scaling up
Dialogue: 0,0:05:32.18,0:05:37.79,Default,,0000,0000,0000,,our production and we went to China to\Nmanufacture robots. And here this is what
Dialogue: 0,0:05:37.79,0:05:39.78,Default,,0000,0000,0000,,we ended up doing.\N{\i1}music{\i0}
Dialogue: 0,0:05:39.78,0:05:43.09,Default,,0000,0000,0000,,It's actually, a cool robot. We built it\Nentirely from scratch. We got our own
Dialogue: 0,0:05:43.09,0:05:48.65,Default,,0000,0000,0000,,chassis, our cabin, our own compute\Nmodule. Basically just about everything.
Dialogue: 0,0:05:49.10,0:05:56.07,Default,,0000,0000,0000,,That was a really cool experience. That\Nwas me. So yeah, that's a robot. That's
Dialogue: 0,0:06:04.61,0:06:11.08,Default,,0000,0000,0000,,the one we were rolling around the past\Nsix months. And we also had some failures
Dialogue: 0,0:06:11.08,0:06:16.86,Default,,0000,0000,0000,,in between, as you saw previously, this\None. So we actually tried a couple of
Dialogue: 0,0:06:16.86,0:06:20.93,Default,,0000,0000,0000,,different concepts. So this was one of\Nthem. This was a Kiwi trike. We thought
Dialogue: 0,0:06:20.93,0:06:25.19,Default,,0000,0000,0000,,that maybe we can figure out how to have\Nrobots do part of the delivery and then
Dialogue: 0,0:06:25.19,0:06:29.59,Default,,0000,0000,0000,,trikes to another part of the delivery. We\Nalso tried to do restaurant robots. We had
Dialogue: 0,0:06:29.59,0:06:33.75,Default,,0000,0000,0000,,like robots that sit in the restaurant and\Nbring food out from the counter to your
Dialogue: 0,0:06:33.75,0:06:38.43,Default,,0000,0000,0000,,doorstep. But what ended up happening is\Nthat it was actually pretty inefficient
Dialogue: 0,0:06:38.43,0:06:41.90,Default,,0000,0000,0000,,and people would wait a really long time\Nfor their deliveries. So it was very
Dialogue: 0,0:06:41.90,0:06:45.45,Default,,0000,0000,0000,,important for us to try a lot of different\Nthings. We tried this robot, the kiwi
Dialogue: 0,0:06:45.45,0:06:50.20,Default,,0000,0000,0000,,TRIKE that not quite worked out as we\Nexpected. We tried a restaurant robot. We
Dialogue: 0,0:06:50.20,0:06:55.07,Default,,0000,0000,0000,,tried a box that would sit behind our\Nrobot. We tryed a hub that would have like
Dialogue: 0,0:06:55.07,0:06:58.99,Default,,0000,0000,0000,,of a bunch of different robots inside of\Nit. So we really, really tried a lot. And
Dialogue: 0,0:06:58.99,0:07:02.77,Default,,0000,0000,0000,,with every iteration, we constantly tried\Nnew techniques we costantly tried new
Dialogue: 0,0:07:02.77,0:07:06.59,Default,,0000,0000,0000,,manufacturing methods. We really tried\Njust about everything to see if we can
Dialogue: 0,0:07:06.59,0:07:11.43,Default,,0000,0000,0000,,make it work. And what we ended up\Nbuilding is a platform that was really
Dialogue: 0,0:07:11.43,0:07:16.61,Default,,0000,0000,0000,,loved by people. We built a platform that\Nstudents adored. That was our primary
Dialogue: 0,0:07:16.61,0:07:21.11,Default,,0000,0000,0000,,demographic we're delivering to college\Ncampuses and students really loved our
Dialogue: 0,0:07:21.11,0:07:24.94,Default,,0000,0000,0000,,products. We actually had people dressed\Nup as Halloween costumes. We had entire
Dialogue: 0,0:07:24.94,0:07:29.54,Default,,0000,0000,0000,,classes go for Halloween in like kiwi bot\Ncostumes. So that was really, really cool
Dialogue: 0,0:07:29.54,0:07:33.95,Default,,0000,0000,0000,,stuff. Had a lot of great support. A lot\Nof trust from the community as well. And
Dialogue: 0,0:07:33.95,0:07:37.28,Default,,0000,0000,0000,,that's like coming back to the design.\NThat aspect of having a friendly robot,
Dialogue: 0,0:07:37.28,0:07:41.08,Default,,0000,0000,0000,,that meshes seamlessly within the fabric\Nof a community is like super, super
Dialogue: 0,0:07:41.08,0:07:44.89,Default,,0000,0000,0000,,important. We've seen other robots around\Nand they were maybe not as friendly, maybe
Dialogue: 0,0:07:44.89,0:07:49.30,Default,,0000,0000,0000,,they looked a little scary. Maybe they had\Nsomething that was a bit off or maybe a
Dialogue: 0,0:07:49.30,0:07:52.57,Default,,0000,0000,0000,,little too industrial. But having like a\Nfriendly robot that could become a meme
Dialogue: 0,0:07:52.57,0:07:56.46,Default,,0000,0000,0000,,that was something truly revolutionary,\Nsomething that really changed the
Dialogue: 0,0:07:56.46,0:08:00.39,Default,,0000,0000,0000,,landscape. And as a matter of fact, like\Nthese cute bots are the only robots that
Dialogue: 0,0:08:00.39,0:08:06.98,Default,,0000,0000,0000,,are deployed somewhere in the world where\Nthey coexist. Day to day with a community,
Dialogue: 0,0:08:06.98,0:08:10.40,Default,,0000,0000,0000,,with people. Like you have some limited\Ndeployments of robots here and there,
Dialogue: 0,0:08:10.40,0:08:13.60,Default,,0000,0000,0000,,maybe have a Roomba at home or some like\Nthat. But you don't have any large scale
Dialogue: 0,0:08:13.60,0:08:19.52,Default,,0000,0000,0000,,deployment. We have robots and people\Nliving in the same city all the time. So,
Dialogue: 0,0:08:19.52,0:08:22.65,Default,,0000,0000,0000,,of course, it took us a while to figure\Nout what to do and how to do it. At first
Dialogue: 0,0:08:22.65,0:08:27.08,Default,,0000,0000,0000,,one of our models was to have robots\Ndeliver the entire meal, like go from the
Dialogue: 0,0:08:27.08,0:08:31.61,Default,,0000,0000,0000,,restaurant all the way to the customer and\Nwe would have a robot do that delivery.
Dialogue: 0,0:08:31.61,0:08:34.39,Default,,0000,0000,0000,,Turns out, it was pretty inefficient.\NPeople would wait like 60 minutes, 90
Dialogue: 0,0:08:34.39,0:08:39.04,Default,,0000,0000,0000,,minutes for their delivery. And we\Nrealized that maybe automating all of that
Dialogue: 0,0:08:39.04,0:08:43.84,Default,,0000,0000,0000,,was not the most efficient approach. So\Nwhat we instead did is a multi-modal
Dialogue: 0,0:08:43.84,0:08:47.25,Default,,0000,0000,0000,,approach where we had people and robots.\NThis is actually a really cool
Dialogue: 0,0:08:47.25,0:08:51.91,Default,,0000,0000,0000,,visualization that my team came up with.\NThe blue lines are robots. So these are
Dialogue: 0,0:08:51.91,0:08:55.83,Default,,0000,0000,0000,,robots roaming around our Berkeley\Ncoverage area and the yellow lines are
Dialogue: 0,0:08:55.83,0:08:59.65,Default,,0000,0000,0000,,people. So how this would work is that\Npeople would go to restaurants, they pick
Dialogue: 0,0:08:59.65,0:09:03.50,Default,,0000,0000,0000,,up the food and they take you to a\Ncluster. They take it to a cluster where
Dialogue: 0,0:09:03.50,0:09:06.94,Default,,0000,0000,0000,,you had a bunch of robots, they loaded\Ninto the robot and then the robot would
Dialogue: 0,0:09:06.94,0:09:11.19,Default,,0000,0000,0000,,actually do the last few hundred meters to\Nyour doorstep. And because we were able to
Dialogue: 0,0:09:11.19,0:09:16.93,Default,,0000,0000,0000,,do this, we were able to go and build a\Nplatform, that handled hundreds of orders a
Dialogue: 0,0:09:16.93,0:09:22.19,Default,,0000,0000,0000,,day with very, very few people. I mean,\Nlabor costs are really high for delivery.
Dialogue: 0,0:09:22.19,0:09:25.37,Default,,0000,0000,0000,,You'd be paying somewhere between five and\Nthirteen dollars to get a meal delivered
Dialogue: 0,0:09:25.37,0:09:28.57,Default,,0000,0000,0000,,in the U.S. And as a student, that's like\Nsuper expensive. That's not something that
Dialogue: 0,0:09:28.57,0:09:32.14,Default,,0000,0000,0000,,you can afford do every day. And also\Nthere is a pretty big shortage of people
Dialogue: 0,0:09:32.14,0:09:35.43,Default,,0000,0000,0000,,who want to do this job in the first\Nplace. The trend is really high. People
Dialogue: 0,0:09:35.43,0:09:38.73,Default,,0000,0000,0000,,are leaving all the time because they\Ndon't like to like sit in a car all day
Dialogue: 0,0:09:38.73,0:09:42.16,Default,,0000,0000,0000,,and deliver food. So that's why we have\Nthis parallel like this multi-modal
Dialogue: 0,0:09:42.16,0:09:47.16,Default,,0000,0000,0000,,approach where the people are biking around,\Nthey're enjoying their time outside and
Dialogue: 0,0:09:47.16,0:09:50.73,Default,,0000,0000,0000,,the robots are actually doing other boring\Nstuff like the waiting. So the robot would
Dialogue: 0,0:09:50.73,0:09:54.58,Default,,0000,0000,0000,,go up to your doorstep and would wait for\Nyou to put on your pants, your shoes and
Dialogue: 0,0:09:54.58,0:09:58.00,Default,,0000,0000,0000,,actually walk outside. So that way we were\Nable to change the dynamic. We're able to
Dialogue: 0,0:09:58.00,0:10:02.89,Default,,0000,0000,0000,,change our deficiency from one or two\Ndeliveries an hour, as you would have with
Dialogue: 0,0:10:02.89,0:10:07.53,Default,,0000,0000,0000,,like a traditional delivery service to as\Nmuch as 15 deliveries an hour per person.
Dialogue: 0,0:10:07.53,0:10:11.04,Default,,0000,0000,0000,,So it made the delivery far more\Naffordable and we were able to offer
Dialogue: 0,0:10:11.04,0:10:16.80,Default,,0000,0000,0000,,delivery at just one dollar a delivery,\Nwhich is a cost that changes completely
Dialogue: 0,0:10:16.80,0:10:21.80,Default,,0000,0000,0000,,the way people approach delivery. In fact,\Nif we look our top 20 percent of users,
Dialogue: 0,0:10:21.80,0:10:26.90,Default,,0000,0000,0000,,they were ordering over 14 times a week.\NSo they were very, very happy that they
Dialogue: 0,0:10:26.90,0:10:30.51,Default,,0000,0000,0000,,could get whatever they wanted very\Nquickly. Of course, not everybody was
Dialogue: 0,0:10:30.51,0:10:35.35,Default,,0000,0000,0000,,super happy. So we did have some people\Nthat didn't fully appreciate the magic
Dialogue: 0,0:10:35.35,0:10:40.69,Default,,0000,0000,0000,,that is kiwi bot. So we did have one\Nperson try to steal it, but they didn't
Dialogue: 0,0:10:40.69,0:10:48.10,Default,,0000,0000,0000,,get away with it. We found them pretty\Nquickly. They hid it in the trunk. Not a
Dialogue: 0,0:10:48.10,0:10:56.31,Default,,0000,0000,0000,,very smart move. We ended up finding it with\NG.P.S. and also triangulating the Wi-Fi.
Dialogue: 0,0:10:56.31,0:11:00.75,Default,,0000,0000,0000,,So this guy decided to steal it because he\Ndoesn't like robots. I don't know why, but
Dialogue: 0,0:11:00.75,0:11:06.62,Default,,0000,0000,0000,,he was clearly very passionate about that\Ntopic. And he stole it and now he's in
Dialogue: 0,0:11:06.62,0:11:14.14,Default,,0000,0000,0000,,jail. So, yeah, don't steal robots. So\Nmaybe some conclusions from our robot
Dialogue: 0,0:11:14.14,0:11:18.35,Default,,0000,0000,0000,,part, like from building robots, from\Nfiguring out like what to do and what not
Dialogue: 0,0:11:18.35,0:11:21.92,Default,,0000,0000,0000,,to do. Really important thing that we do a\Nlot in software and maybe not as much
Dialogue: 0,0:11:21.92,0:11:27.11,Default,,0000,0000,0000,,hardware is iteration. Like we iterated\Nthrough three major revisions and like
Dialogue: 0,0:11:27.11,0:11:32.20,Default,,0000,0000,0000,,lots of small revisions during a really\Nsmall period of time. It was really
Dialogue: 0,0:11:32.20,0:11:35.92,Default,,0000,0000,0000,,interesting to see like that transition.\NEvery single time we try something new, we
Dialogue: 0,0:11:35.92,0:11:39.76,Default,,0000,0000,0000,,try it maybe for like 20 robots at a time,\Nlike not our whole fleet. We just try for
Dialogue: 0,0:11:39.76,0:11:43.69,Default,,0000,0000,0000,,a small portion of our fleet and that we\Nwere able to iterate really quickly and
Dialogue: 0,0:11:43.69,0:11:48.77,Default,,0000,0000,0000,,see what sensors worked or cameras worked.\NAnd just to see what we could do in order
Dialogue: 0,0:11:48.77,0:11:55.13,Default,,0000,0000,0000,,to grow the products, it was very\Nimportant to iterate. Communication.
Dialogue: 0,0:11:55.13,0:11:59.78,Default,,0000,0000,0000,,Communication is absolutely fundamental.\NAnd not only communication like inside the
Dialogue: 0,0:11:59.78,0:12:03.07,Default,,0000,0000,0000,,company or anything, but more importantly,\Ncommunication with your community. Because
Dialogue: 0,0:12:03.07,0:12:06.60,Default,,0000,0000,0000,,we weren't just building a product in\Nisolation. We were building a product for
Dialogue: 0,0:12:06.60,0:12:11.53,Default,,0000,0000,0000,,people who live in a city, who have an\Nestablished life. And we're kind of
Dialogue: 0,0:12:11.53,0:12:15.41,Default,,0000,0000,0000,,intruding into their lives by bringing in\Na new product that takes the sidewalks. So
Dialogue: 0,0:12:15.41,0:12:19.82,Default,,0000,0000,0000,,communicating what we're doing, showing\Nthem what this is and what this robot does
Dialogue: 0,0:12:19.82,0:12:24.34,Default,,0000,0000,0000,,is super important. Actually, very early\Non our designs have no text on it. They
Dialogue: 0,0:12:24.34,0:12:28.73,Default,,0000,0000,0000,,had like no information it was just like a\Nbasket case on RC car. And people were
Dialogue: 0,0:12:28.73,0:12:32.47,Default,,0000,0000,0000,,like really confused. The police were\Nlike: "Hey, what is this?", so we had to
Dialogue: 0,0:12:32.47,0:12:35.71,Default,,0000,0000,0000,,add a lot of communication, we had to put\Nfood delivery on the robots really
Dialogue: 0,0:12:35.71,0:12:38.98,Default,,0000,0000,0000,,clearly, we had to add a license plate\Nwith like a phone number that somebody
Dialogue: 0,0:12:38.98,0:12:43.39,Default,,0000,0000,0000,,could reach out to us. So communication is\Nvery, very, very important when it comes
Dialogue: 0,0:12:43.39,0:12:50.19,Default,,0000,0000,0000,,to robots. Also: Scaling hardware is hard,\Nsuper hard. I mean, it was crazy. When we
Dialogue: 0,0:12:50.19,0:12:55.23,Default,,0000,0000,0000,,first started it was just Arduinos and\NRaspberry Pis and that did not scale
Dialogue: 0,0:12:55.23,0:13:00.17,Default,,0000,0000,0000,,really well. Like, sure, we could have\Nmaybe 10 or 20 units at once. But then how
Dialogue: 0,0:13:00.17,0:13:04.66,Default,,0000,0000,0000,,do you handle updates? How do you handle\Nthose weird things that happen all the
Dialogue: 0,0:13:04.66,0:13:08.66,Default,,0000,0000,0000,,time? So it was really challenging to do\Nthis. We actually killed a bunch of SD
Dialogue: 0,0:13:08.66,0:13:12.79,Default,,0000,0000,0000,,cards. Didn't really know you could\Ndestroy SD cards, but you can. And we
Dialogue: 0,0:13:12.79,0:13:17.26,Default,,0000,0000,0000,,learned a lot of things about hardware\Npushing it beyond its normal boundaries.
Dialogue: 0,0:13:17.26,0:13:20.68,Default,,0000,0000,0000,,So yeah, iteration, super important.\NCommunication is key, like getting buy in
Dialogue: 0,0:13:20.68,0:13:25.59,Default,,0000,0000,0000,,from the community and scaling hardware is\Nsuper, super hard. That's something we
Dialogue: 0,0:13:25.59,0:13:33.05,Default,,0000,0000,0000,,actually figured out how to solve by going\Ninto China. So how to do or how not to do
Dialogue: 0,0:13:33.05,0:13:42.02,Default,,0000,0000,0000,,manufacturing? So as every China story\Ngoes, I hopped on a plane and I ended up
Dialogue: 0,0:13:42.02,0:13:46.18,Default,,0000,0000,0000,,in China. And it's really interesting to\Nsee because like you have this perception
Dialogue: 0,0:13:46.18,0:13:50.95,Default,,0000,0000,0000,,of China from the media, you have this\Nidea of what it would look like. But the
Dialogue: 0,0:13:50.95,0:13:55.90,Default,,0000,0000,0000,,reality is it doesn't look anything like\Nwhat you would expect. It was a completely
Dialogue: 0,0:13:55.90,0:14:00.40,Default,,0000,0000,0000,,different world. It was at the same time\NBladerunner and like the most modern city
Dialogue: 0,0:14:00.40,0:14:05.78,Default,,0000,0000,0000,,in the world and it was truly an awesome\Nexperience. I highly recommend anybody who
Dialogue: 0,0:14:05.78,0:14:10.75,Default,,0000,0000,0000,,has the opportunity to go in and explore\Nthe world. But of course, the culture is a
Dialogue: 0,0:14:10.75,0:14:14.76,Default,,0000,0000,0000,,little bit different. We were surprised to\Nsee some things happening there. Was a
Dialogue: 0,0:14:14.76,0:14:18.88,Default,,0000,0000,0000,,weird dichotomy between communism and\Nconsumerism. This is kind of interesting
Dialogue: 0,0:14:18.88,0:14:27.73,Default,,0000,0000,0000,,to see that sometimes. But the reason why\Nwe came to China is for manufacturing and
Dialogue: 0,0:14:27.73,0:14:31.91,Default,,0000,0000,0000,,there is no better place for that than\NShenzhen. In Shenzhen, you have
Dialogue: 0,0:14:31.91,0:14:37.56,Default,,0000,0000,0000,,Huaqiangbei. This huge market. It's a\Nmarket that spans several city blocks and
Dialogue: 0,0:14:37.56,0:14:41.91,Default,,0000,0000,0000,,you can actually find anything and\Neverything you want. We were able to get
Dialogue: 0,0:14:41.91,0:14:46.65,Default,,0000,0000,0000,,components super quickly, super easily.\NAnd you could spend days just walking to a
Dialogue: 0,0:14:46.65,0:14:52.15,Default,,0000,0000,0000,,single building finding different things.\NThere were entire city blocks dedicated to
Dialogue: 0,0:14:52.15,0:14:57.97,Default,,0000,0000,0000,,like just LEDs or just connectors or just\Nprocessors. It was absolutely crazy. You
Dialogue: 0,0:14:57.97,0:15:02.29,Default,,0000,0000,0000,,could really, really, really get lost\Ninside of these mazes. And what was really
Dialogue: 0,0:15:02.29,0:15:06.41,Default,,0000,0000,0000,,incredible to see and something I've never\Nseen anywhere else in the world is just
Dialogue: 0,0:15:06.41,0:15:12.96,Default,,0000,0000,0000,,how easy it is to get hardware, to get\Nthings, to get parts. It was super easy.
Dialogue: 0,0:15:12.96,0:15:17.30,Default,,0000,0000,0000,,Just go in and get something and you could\Nget it at one piece, two pieces, a
Dialogue: 0,0:15:17.30,0:15:21.09,Default,,0000,0000,0000,,thousand pieces like instantly. If you're\Nanywhere else in the world, that's super
Dialogue: 0,0:15:21.09,0:15:24.98,Default,,0000,0000,0000,,hard to do. So just by this virtue, you're\Nactually able to prototype things. You're
Dialogue: 0,0:15:24.98,0:15:28.81,Default,,0000,0000,0000,,able to build things incredibly fast.\NYou're able to go in, you're able to
Dialogue: 0,0:15:28.81,0:15:33.96,Default,,0000,0000,0000,,comission a PCB and get all the parts\Nalmost instantly, which is not something
Dialogue: 0,0:15:33.96,0:15:37.85,Default,,0000,0000,0000,,you see anywhere else in the world. And\Nalso a lot of the manufacturers have their
Dialogue: 0,0:15:37.85,0:15:41.63,Default,,0000,0000,0000,,booths here so these would be direct\Nbooths from the manufacturers so you could
Dialogue: 0,0:15:41.63,0:15:45.29,Default,,0000,0000,0000,,say go up to them, start talking to them\Nand ask, hey, can you make this product
Dialogue: 0,0:15:45.29,0:15:49.96,Default,,0000,0000,0000,,this specific way? Can you do it how I\Nwant it? And they'll be like, sure, why
Dialogue: 0,0:15:49.96,0:15:53.75,Default,,0000,0000,0000,,not? They'll do it for you. So it was\Nreally, really valuable to just learn from
Dialogue: 0,0:15:53.75,0:15:58.53,Default,,0000,0000,0000,,these people, from the vendors here, from\Nmanufacturers about how to build things.
Dialogue: 0,0:15:58.53,0:16:03.38,Default,,0000,0000,0000,,And it was actually really surprising to\Nsee everything they had in stock. Two
Dialogue: 0,0:16:03.38,0:16:07.96,Default,,0000,0000,0000,,years ago, we built an art installation\Nhere that covered a tunnel with LEDs. We
Dialogue: 0,0:16:07.96,0:16:13.49,Default,,0000,0000,0000,,covered one of the tunnels at 34C3 with\NLEDs. And we used this tiny, tiny chip. It
Dialogue: 0,0:16:13.49,0:16:18.27,Default,,0000,0000,0000,,was a five dollar ESP 8266 chip that\Nbasically was able to control all your
Dialogue: 0,0:16:18.27,0:16:23.37,Default,,0000,0000,0000,,LEDs. And over the course of five years,\Nup to that point, I spent a lot of time
Dialogue: 0,0:16:23.37,0:16:28.30,Default,,0000,0000,0000,,figuring out how to build it myself. I\Nplayed with Raspberry Pis I played with
Dialogue: 0,0:16:28.30,0:16:33.21,Default,,0000,0000,0000,,PCA controllers over serial and like I\Nfinally managed to get a prototype to
Dialogue: 0,0:16:33.21,0:16:36.95,Default,,0000,0000,0000,,work, but it was super clunky, it was\Nsuper expensive and it wasn't very
Dialogue: 0,0:16:36.95,0:16:41.26,Default,,0000,0000,0000,,reliable and I go to China and I find that\Nit's available there and much better
Dialogue: 0,0:16:41.26,0:16:46.34,Default,,0000,0000,0000,,quality, much cheaper, much faster, so it\Nwas a really, really interesting shift in
Dialogue: 0,0:16:46.34,0:16:49.39,Default,,0000,0000,0000,,perspective. It's something you can't\Nappreciate when you're abroad. Even if
Dialogue: 0,0:16:49.39,0:16:52.90,Default,,0000,0000,0000,,you're browsing like eBay or Ali Express\Nit's kind of hard to appreciate just how
Dialogue: 0,0:16:52.90,0:16:57.91,Default,,0000,0000,0000,,much selection you have and how you can\Nfind just about any tool, anything you
Dialogue: 0,0:16:57.91,0:17:02.50,Default,,0000,0000,0000,,need to find. So it's really, really\Nincredible. But these markets were cool
Dialogue: 0,0:17:02.50,0:17:08.01,Default,,0000,0000,0000,,but was even cooler are the factories and\Nduring a course in China, we were able to
Dialogue: 0,0:17:08.01,0:17:11.58,Default,,0000,0000,0000,,visit a lot of factories. All these\Nfactories there are super, super
Dialogue: 0,0:17:11.58,0:17:15.16,Default,,0000,0000,0000,,welcoming. They always love having you\Nover. They invited you to really, really
Dialogue: 0,0:17:15.16,0:17:20.82,Default,,0000,0000,0000,,luxurioius dinners. We had way too much\Nfood. And it was a feast of celebration
Dialogue: 0,0:17:20.82,0:17:24.15,Default,,0000,0000,0000,,every time. Actually, relationships are\Nsuper, super important in China. Like a
Dialogue: 0,0:17:24.15,0:17:28.19,Default,,0000,0000,0000,,lot of people in the West, like they have\Ncontracts and they say, OK, this is the
Dialogue: 0,0:17:28.19,0:17:32.65,Default,,0000,0000,0000,,terms of the contracts. Well, China, you\Ndo sort of have contracts, but they don't
Dialogue: 0,0:17:32.65,0:17:36.91,Default,,0000,0000,0000,,matter as much as relationships. Like when\Nyou have a relationship with the
Dialogue: 0,0:17:36.91,0:17:39.68,Default,,0000,0000,0000,,manufacturer, you have to like always go\Nto dinner with them, drink beer, smoke, go
Dialogue: 0,0:17:39.68,0:17:43.85,Default,,0000,0000,0000,,to KTV like it's a really evolved\Nrelationship. And you're only able to have
Dialogue: 0,0:17:43.85,0:17:46.70,Default,,0000,0000,0000,,good communication based on that\Nrelationship, because if you don't have a
Dialogue: 0,0:17:46.70,0:17:49.96,Default,,0000,0000,0000,,relationship they kind of forget about\Nyou. We actually had a couple of instances
Dialogue: 0,0:17:49.96,0:17:54.52,Default,,0000,0000,0000,,where manufacturers ghosted us. Like they\Nhad a critical component and they just
Dialogue: 0,0:17:54.52,0:17:57.61,Default,,0000,0000,0000,,stopped answering our emails, they stop\Nanswering our weechats, they just
Dialogue: 0,0:17:57.61,0:18:01.21,Default,,0000,0000,0000,,completely ignored us. And for some pieces\Nthey were completely irreplaceable, we
Dialogue: 0,0:18:01.21,0:18:05.18,Default,,0000,0000,0000,,could not just go out and find another\Nfactory to produce a specific part the way
Dialogue: 0,0:18:05.18,0:18:09.01,Default,,0000,0000,0000,,we wanted. And the only way you can ensure\Nthat this doesn't happen is by really
Dialogue: 0,0:18:09.01,0:18:13.61,Default,,0000,0000,0000,,explicitly making sure that you have good\Ncommunication, a good relationship with
Dialogue: 0,0:18:13.61,0:18:17.45,Default,,0000,0000,0000,,that manufacturers. It's super, super\Nimportant. This is one of the factories we
Dialogue: 0,0:18:17.45,0:18:23.38,Default,,0000,0000,0000,,worked with. It's really crazy. I mean, we\Nwent there and we're just absolutely blown
Dialogue: 0,0:18:23.38,0:18:27.83,Default,,0000,0000,0000,,away by the scale of everything and also\Nblown away by how manual everything is.
Dialogue: 0,0:18:27.83,0:18:32.16,Default,,0000,0000,0000,,There's actually audio here. Everything\Nwas super manual. People were just like
Dialogue: 0,0:18:32.16,0:18:36.31,Default,,0000,0000,0000,,there with minimal or no protective\Nequipment whatsoever. Just like building
Dialogue: 0,0:18:36.31,0:18:39.54,Default,,0000,0000,0000,,things that look like they were made by\Nrobots or machines, but they were in
Dialogue: 0,0:18:39.54,0:18:45.89,Default,,0000,0000,0000,,reality built by people with their hands,\Nwhich is super crazy to see. And there
Dialogue: 0,0:18:45.89,0:18:51.57,Default,,0000,0000,0000,,were a lot of Blade Runner esque designs,\Nreally bizarre contraptions there in this
Dialogue: 0,0:18:51.57,0:18:57.45,Default,,0000,0000,0000,,factory. This is our fiberglass factory.\NThe way we build our casing, was actually
Dialogue: 0,0:18:57.45,0:19:03.94,Default,,0000,0000,0000,,prototyping it first in carbon fiber,\Nsorry, fiberglass and then moving onto a
Dialogue: 0,0:19:03.94,0:19:08.61,Default,,0000,0000,0000,,mold in carbon fiber. And actually Scotty,\Nhe made a really cool video on YouTube. So
Dialogue: 0,0:19:08.61,0:19:13.43,Default,,0000,0000,0000,,if you search for hockey stick factory on\NYouTube, you can see a huge video. My
Dialogue: 0,0:19:13.43,0:19:17.54,Default,,0000,0000,0000,,buddy Scotty actually goes and meets his\Nfactory, discover how they make this mold
Dialogue: 0,0:19:17.54,0:19:21.18,Default,,0000,0000,0000,,and how they make these carbon fiber\Nthings. It was actually really crazy to
Dialogue: 0,0:19:21.18,0:19:26.03,Default,,0000,0000,0000,,see it. It was cheaper to make a carbon\Nfiber mold than it was to make a plastic
Dialogue: 0,0:19:26.03,0:19:29.46,Default,,0000,0000,0000,,mold. So since the tolerances were a\Nlittle bit different, since like the
Dialogue: 0,0:19:29.46,0:19:32.74,Default,,0000,0000,0000,,process was a little bit simpler, you were\Nable to make a mold, that was very, very
Dialogue: 0,0:19:32.74,0:19:38.44,Default,,0000,0000,0000,,strong and very indestructible without\Nnecessarily having to have all of that
Dialogue: 0,0:19:38.44,0:19:43.91,Default,,0000,0000,0000,,expense upfront for like a plastic mold.\NSo, yeah, that our a fiberglass factory.
Dialogue: 0,0:19:43.91,0:19:47.37,Default,,0000,0000,0000,,Really exciting stuff. Really crazy scale.\NThese folks like the first night we came
Dialogue: 0,0:19:47.37,0:19:51.06,Default,,0000,0000,0000,,there, we arrived at like 8 pm and there\Nwas 100 people in the factory just like
Dialogue: 0,0:19:51.06,0:19:56.72,Default,,0000,0000,0000,,working at 8pm. Really crazy to see. This\Nis another factory worked with. So this
Dialogue: 0,0:19:56.72,0:20:01.54,Default,,0000,0000,0000,,was a metal factory. It was actually\Nreally, really, really interesting to see
Dialogue: 0,0:20:01.54,0:20:06.02,Default,,0000,0000,0000,,how they built all these things and at one\Nhand, you can build super complex things,
Dialogue: 0,0:20:06.02,0:20:10.56,Default,,0000,0000,0000,,you can build a super complex designs. But\Non the other hand, we got surprised a
Dialogue: 0,0:20:10.56,0:20:15.75,Default,,0000,0000,0000,,couple of times by being unable to\Nmanufacture really simple designs. And it
Dialogue: 0,0:20:15.75,0:20:19.69,Default,,0000,0000,0000,,took us a while to get a grasp was like,\Noh, OK, so we can make really complex
Dialogue: 0,0:20:19.69,0:20:24.09,Default,,0000,0000,0000,,metal that's bent, but as soon as we add a\Nwell to aluminum, you start to have a big,
Dialogue: 0,0:20:24.09,0:20:28.06,Default,,0000,0000,0000,,big problem. So we had to like change a\Nlot of our designs. We had to really adapt
Dialogue: 0,0:20:28.06,0:20:32.20,Default,,0000,0000,0000,,to the way things were being made in\NChina. And sometimes you could adapt
Dialogue: 0,0:20:32.20,0:20:36.14,Default,,0000,0000,0000,,yourself, but like at an insane cost. So\Nit was better to adapt to the way things
Dialogue: 0,0:20:36.14,0:20:41.19,Default,,0000,0000,0000,,were being done there. So, again, very,\Nvery interesting to see how things are
Dialogue: 0,0:20:41.19,0:20:45.48,Default,,0000,0000,0000,,done. No protective equipment, this is\Nlike a two ton press and his hands are
Dialogue: 0,0:20:45.48,0:20:50.14,Default,,0000,0000,0000,,millimeters away from it. So, yeah, it's a\Ndifferent world out there. Very, very
Dialogue: 0,0:20:50.14,0:20:57.83,Default,,0000,0000,0000,,different. Another factory we visited was\Na PCB factory. So this one has a really
Dialogue: 0,0:20:57.83,0:21:02.24,Default,,0000,0000,0000,,interesting story. This factory is not in\NShenzhen. It's just across the border from
Dialogue: 0,0:21:02.24,0:21:07.30,Default,,0000,0000,0000,,Shenzhen. The city actually passed a law a\Ncouple of years ago that has very, very
Dialogue: 0,0:21:07.30,0:21:11.90,Default,,0000,0000,0000,,strict environmental policies. So you're\Nno longer able to do PCB manufacturing
Dialogue: 0,0:21:11.90,0:21:15.90,Default,,0000,0000,0000,,inside the city anymore. So we actually\Nhad to drive for a couple of hours outside
Dialogue: 0,0:21:15.90,0:21:21.20,Default,,0000,0000,0000,,of the city and over there was a huge\Nplant. And this plant was kind of semi
Dialogue: 0,0:21:21.20,0:21:26.65,Default,,0000,0000,0000,,automated, semi handmade. Were part of the\Nprocess were done by hand, as you see
Dialogue: 0,0:21:26.65,0:21:31.28,Default,,0000,0000,0000,,here. But then parts of the process were\Ndone with machines. So they had this giant
Dialogue: 0,0:21:31.28,0:21:33.83,Default,,0000,0000,0000,,machine, which is basically a black box we\Ncan't really see inside of it. But you had
Dialogue: 0,0:21:33.83,0:21:38.98,Default,,0000,0000,0000,,a bunch of chemicals and it's like take a\NPCB and just like move it forward through
Dialogue: 0,0:21:38.98,0:21:43.22,Default,,0000,0000,0000,,a chain. This is really intresting to see.\NAnd this factory also had a really quick
Dialogue: 0,0:21:43.22,0:21:46.07,Default,,0000,0000,0000,,turnaround, they had a three hour\Nturnaround if you paid a premium and the
Dialogue: 0,0:21:46.07,0:21:50.58,Default,,0000,0000,0000,,standard was 24 hours. You could also ask\Nthem to do PCBA so you can actually get
Dialogue: 0,0:21:50.58,0:21:55.07,Default,,0000,0000,0000,,them to assemble the PCB for you. And we\Nended up doing that for some of our PCBs.
Dialogue: 0,0:21:55.07,0:22:00.23,Default,,0000,0000,0000,,We'd give them build materials and we'd\Ngive them our designs and then they
Dialogue: 0,0:22:00.23,0:22:04.08,Default,,0000,0000,0000,,manifacture it. We actually got in a\Nlittle bit of a situation with that
Dialogue: 0,0:22:04.08,0:22:08.12,Default,,0000,0000,0000,,because we sent them some designs, we sent\Nthem some parts that we wanted to put in
Dialogue: 0,0:22:08.12,0:22:12.35,Default,,0000,0000,0000,,our PCB and it turns out that one of these\Nparts was unavailable and they didn't tell
Dialogue: 0,0:22:12.35,0:22:17.17,Default,,0000,0000,0000,,that to us until it was almost Chinese New\NYear. So we had to scramble all that to
Dialogue: 0,0:22:17.17,0:22:21.43,Default,,0000,0000,0000,,find another solution. Was very exciting\Nto see how you would deal with these
Dialogue: 0,0:22:21.43,0:22:25.06,Default,,0000,0000,0000,,factories. There are some even cooler\Nfactories. I think the coolest factory I
Dialogue: 0,0:22:25.06,0:22:29.26,Default,,0000,0000,0000,,visited was a battery factory where they\Nmade lithium ion and lithium polymer
Dialogue: 0,0:22:29.26,0:22:34.12,Default,,0000,0000,0000,,batteries and it was almost entirely\Nautomated. It had giant films of things
Dialogue: 0,0:22:34.12,0:22:39.13,Default,,0000,0000,0000,,going into a machine and then you had all\Nsorts of liquids and powders it was all
Dialogue: 0,0:22:39.13,0:22:43.00,Default,,0000,0000,0000,,combined together. It was super, super\Ncool. Didn't allow us to film it
Dialogue: 0,0:22:43.00,0:22:45.76,Default,,0000,0000,0000,,unfortunately, there may be only a dozen\Nor so such factories in the world. They're
Dialogue: 0,0:22:45.76,0:22:49.38,Default,,0000,0000,0000,,very protective about their technology,\Nbut the scale of how quickly they're
Dialogue: 0,0:22:49.38,0:22:53.07,Default,,0000,0000,0000,,manufacturing these batteries was just\Nincredible. They would manufacture them at
Dialogue: 0,0:22:53.07,0:22:58.69,Default,,0000,0000,0000,,a crazy, crazy scale. So all these factors\Nare cool, but actually building things is
Dialogue: 0,0:22:58.69,0:23:03.08,Default,,0000,0000,0000,,even cooler. So we ended up partnering\Nwith a contract manufacturer. I was really
Dialogue: 0,0:23:03.08,0:23:07.46,Default,,0000,0000,0000,,fortunate to find one through my network.\NOtherwise, I would have been totally lost.
Dialogue: 0,0:23:07.46,0:23:11.46,Default,,0000,0000,0000,,A couple of days before I ended up going\Nto China I found a contract manufacturer
Dialogue: 0,0:23:11.46,0:23:16.36,Default,,0000,0000,0000,,that liked to work with start ups and\Nsmall scale people and we ended up working
Dialogue: 0,0:23:16.36,0:23:20.34,Default,,0000,0000,0000,,with them to build our first batch of 50\Nrobots. It was really interesting to see
Dialogue: 0,0:23:20.34,0:23:25.69,Default,,0000,0000,0000,,how different our designs were to what\Nthey expected. So they expected things are
Dialogue: 0,0:23:25.69,0:23:30.73,Default,,0000,0000,0000,,really ready. They are very explicit, very\Nclearly specified. But we didn't have
Dialogue: 0,0:23:30.73,0:23:34.62,Default,,0000,0000,0000,,that. The difference between manufacturing\Nin the U.S., for example, against China is
Dialogue: 0,0:23:34.62,0:23:39.02,Default,,0000,0000,0000,,that in the U.S., like it's a super long\Nprocess and the back and forth takes super
Dialogue: 0,0:23:39.02,0:23:42.30,Default,,0000,0000,0000,,long just to get an idea of what kind of\Nfiles they need. Whereas in China, you're
Dialogue: 0,0:23:42.30,0:23:46.43,Default,,0000,0000,0000,,able to sit down directly with the\Nengineer, with the person in charge, and
Dialogue: 0,0:23:46.43,0:23:49.77,Default,,0000,0000,0000,,you can figure out what they need and they\Ncan help you out instantly. Actually, just
Dialogue: 0,0:23:49.77,0:23:54.99,Default,,0000,0000,0000,,here, I just want to show you one thing.\NSo this is my designer, Alehandrew, and he was
Dialogue: 0,0:23:54.99,0:23:59.71,Default,,0000,0000,0000,,translating from English to Chinese with\Nhis phone with Google Translate, and it
Dialogue: 0,0:23:59.71,0:24:03.47,Default,,0000,0000,0000,,worked surprisingly well. Google Translate\Nactually is not blocked in China for some
Dialogue: 0,0:24:03.47,0:24:09.02,Default,,0000,0000,0000,,reason. We were able to communicate almost\Nall the time with that. Also, Weechat has
Dialogue: 0,0:24:09.02,0:24:12.45,Default,,0000,0000,0000,,a built in translate feature. So Weechat\Nis like the universal app that everyone in
Dialogue: 0,0:24:12.45,0:24:16.16,Default,,0000,0000,0000,,China uses and has this built in\Ntranslation feature that can translate
Dialogue: 0,0:24:16.16,0:24:21.73,Default,,0000,0000,0000,,your text automatically. So it's really,\Nreally cool to see how that worked. One
Dialogue: 0,0:24:21.73,0:24:25.87,Default,,0000,0000,0000,,question that we get commonly asked is\Nlike how do we find our manufacturers? How
Dialogue: 0,0:24:25.87,0:24:30.74,Default,,0000,0000,0000,,do we get this relationship? So about 20\Npercent of that was through Alibaba. So
Dialogue: 0,0:24:30.74,0:24:35.83,Default,,0000,0000,0000,,our fiberglass manufacturer. We quoted\Nlike 30 different manufacturers and went
Dialogue: 0,0:24:35.83,0:24:40.89,Default,,0000,0000,0000,,with the cheapest one. Of course, it was\Nfar more expensive than we expected and we
Dialogue: 0,0:24:40.89,0:24:45.67,Default,,0000,0000,0000,,ended up producing with them. 20 percent\Nthat, for example, our chassey, it was
Dialogue: 0,0:24:45.67,0:24:49.63,Default,,0000,0000,0000,,built with companies that we already had a\Nrelationship with. So we were just able to
Dialogue: 0,0:24:49.63,0:24:54.35,Default,,0000,0000,0000,,continue working with them. And then 60\Npercent was through just references so
Dialogue: 0,0:24:54.35,0:24:58.15,Default,,0000,0000,0000,,select networks or getting to know people\Nand talking to them saying, "Oh, hey, who
Dialogue: 0,0:24:58.15,0:25:01.95,Default,,0000,0000,0000,,did you use for this or this" or "How did\Nyou make these PCBs?" or just getting a
Dialogue: 0,0:25:01.95,0:25:06.69,Default,,0000,0000,0000,,conversation going. So having that kind of\Nnetwork was really, really helpful in
Dialogue: 0,0:25:06.69,0:25:12.93,Default,,0000,0000,0000,,order to build these robots. So as you can\Nactually see right here, our design, this
Dialogue: 0,0:25:12.93,0:25:16.64,Default,,0000,0000,0000,,is what we had when we came into China.\NWhen we left, we had our own computer
Dialogue: 0,0:25:16.64,0:25:22.09,Default,,0000,0000,0000,,module like super sophisticated. But this\Nwas like a Raspberry Pi, a pix hock and a
Dialogue: 0,0:25:22.09,0:25:26.54,Default,,0000,0000,0000,,voltage converter like a DC to DC\Nconverter. That was pretty much it. As you
Dialogue: 0,0:25:26.54,0:25:30.87,Default,,0000,0000,0000,,can see, it was not very reliable. It\Nwould break a lot. So it took us quite a
Dialogue: 0,0:25:30.87,0:25:36.20,Default,,0000,0000,0000,,while to translate this into something\Nthat was manufacturing well. So thanks to
Dialogue: 0,0:25:36.20,0:25:42.46,Default,,0000,0000,0000,,the dedication of my incredible team that\Nwe're able to do that. And we kind of did
Dialogue: 0,0:25:42.46,0:25:46.38,Default,,0000,0000,0000,,not know what we were doing so, we ended\Nup having all of our parts and all of the
Dialogue: 0,0:25:46.38,0:25:51.96,Default,,0000,0000,0000,,components ready just days before Chinese\NNew Year. So we actually had to do all of
Dialogue: 0,0:25:51.96,0:25:55.89,Default,,0000,0000,0000,,that someday ourselves. We didn't have any\NChinese workers who could help us do that.
Dialogue: 0,0:25:55.89,0:26:01.25,Default,,0000,0000,0000,,So there's our team just assembling things\Nin the factory like we wanted two days
Dialogue: 0,0:26:01.25,0:26:06.52,Default,,0000,0000,0000,,before Chinese New Year. So that was very,\Nvery interesting. We kind of hacked or
Dialogue: 0,0:26:06.52,0:26:12.31,Default,,0000,0000,0000,,tried to hack Chinese New Year. We\Nassembled all the robots literally days if
Dialogue: 0,0:26:12.31,0:26:16.85,Default,,0000,0000,0000,,not hours before Chinese New Year and we\Nshipped them out and everything was great,
Dialogue: 0,0:26:16.85,0:26:22.36,Default,,0000,0000,0000,,except our robots got stuck in customs. We\Nhad a trademark on our box and the customs
Dialogue: 0,0:26:22.36,0:26:25.52,Default,,0000,0000,0000,,agents, they open the box and saw more\Ntrademarks on some parts. We had 3D
Dialogue: 0,0:26:25.52,0:26:28.78,Default,,0000,0000,0000,,printed parts and they were like, no, this\Nis not going to go through without the
Dialogue: 0,0:26:28.78,0:26:34.69,Default,,0000,0000,0000,,proper paperwork. So our robots got stuck\Nfor three weeks in China, which was really
Dialogue: 0,0:26:34.69,0:26:39.33,Default,,0000,0000,0000,,fun. Little problematic. So, yeah, those\Nkind of things happen you have to be ready
Dialogue: 0,0:26:39.33,0:26:45.75,Default,,0000,0000,0000,,for it. After we received our robots in\NCalifornia, we had to spend another like
Dialogue: 0,0:26:45.75,0:26:51.79,Default,,0000,0000,0000,,maybe one or two months refinishing them,\Nredoing some parts, tweaking them,
Dialogue: 0,0:26:51.79,0:26:56.03,Default,,0000,0000,0000,,flashing them. So there's still a lot of\Nwork to get them to work. The pieces we
Dialogue: 0,0:26:56.03,0:26:59.95,Default,,0000,0000,0000,,shipped out to China was maybe just like a\Ncase with most of the electronics in, but
Dialogue: 0,0:26:59.95,0:27:06.10,Default,,0000,0000,0000,,not all of it. So we still had to do a lot\Nof tweaking over back home. And of course,
Dialogue: 0,0:27:06.10,0:27:09.00,Default,,0000,0000,0000,,all this going to be impossible without an\Nincredible team so I was really fortunate
Dialogue: 0,0:27:09.00,0:27:14.54,Default,,0000,0000,0000,,to be with some really, really passionate\Npeople who would work four months in a row
Dialogue: 0,0:27:14.54,0:27:18.90,Default,,0000,0000,0000,,continuously without virtually taking any\Nbreaks. We had plenty of opportunities to
Dialogue: 0,0:27:18.90,0:27:23.03,Default,,0000,0000,0000,,go and take the high speed rail or go to\NShanghai or even Tokyo, but we all stayed
Dialogue: 0,0:27:23.03,0:27:27.22,Default,,0000,0000,0000,,in Shenzhen and spend a lot of time\Ntogether building these robots. It was a
Dialogue: 0,0:27:27.22,0:27:33.81,Default,,0000,0000,0000,,really, really arduous journey. So maybe\Nsome conclusions for scaling,
Dialogue: 0,0:27:33.81,0:27:37.15,Default,,0000,0000,0000,,manufacturing, some of the failures we've\Nhad in relationships. I mean,
Dialogue: 0,0:27:37.15,0:27:41.15,Default,,0000,0000,0000,,relationships are super important, like\Nsuper, super important, in China far more
Dialogue: 0,0:27:41.15,0:27:44.62,Default,,0000,0000,0000,,important then contracts. If you're able\Nto have a good line of communication with
Dialogue: 0,0:27:44.62,0:27:47.80,Default,,0000,0000,0000,,your manufacturer, that really, really\Nhelps out. Because if you don't, things go
Dialogue: 0,0:27:47.80,0:27:50.79,Default,,0000,0000,0000,,bad. We've had manufacturers that ghosted\Nus. We have had manufacturers that
Dialogue: 0,0:27:50.79,0:27:56.14,Default,,0000,0000,0000,,completely ignored us or manufacturers\Nthat just replaced components because they
Dialogue: 0,0:27:56.14,0:28:02.30,Default,,0000,0000,0000,,just felt like it. So relationships, super\Nimportant. Don't hack Chinese New Year. We
Dialogue: 0,0:28:02.30,0:28:06.21,Default,,0000,0000,0000,,tried it, doesn't work. It's a thing.\NChina just sit's down for like two or
Dialogue: 0,0:28:06.21,0:28:09.59,Default,,0000,0000,0000,,three weeks. So it's really, really\Nimportant to respect that. People buy
Dialogue: 0,0:28:09.59,0:28:13.98,Default,,0000,0000,0000,,tickets to go to their hometowns like\Nmonths in advance and they're not going to
Dialogue: 0,0:28:13.98,0:28:17.18,Default,,0000,0000,0000,,move it for just like some pesky thing\Nthat you're building, especially for like
Dialogue: 0,0:28:17.18,0:28:20.85,Default,,0000,0000,0000,,some small scale thing. So, yeah, don't\Ntry to hack Chinese New Year, it did not
Dialogue: 0,0:28:20.85,0:28:27.11,Default,,0000,0000,0000,,work out well for us. Also, do it with a\Nteam . While I was in China, I saw a
Dialogue: 0,0:28:27.11,0:28:30.45,Default,,0000,0000,0000,,couple of sole entrepreneurs try to build\Ntheir own thing and it was super, super
Dialogue: 0,0:28:30.45,0:28:34.17,Default,,0000,0000,0000,,hard, super stressful, having a team is\Nreally great, especially if like a foreign
Dialogue: 0,0:28:34.17,0:28:37.76,Default,,0000,0000,0000,,place where you don't really know anybody.\NHaving that team there together, to
Dialogue: 0,0:28:37.76,0:28:41.39,Default,,0000,0000,0000,,support you is super, super important,\Nespecially since you can multitask, you
Dialogue: 0,0:28:41.39,0:28:44.84,Default,,0000,0000,0000,,can split responsibilities and do\Nsomething together. So it's really, really
Dialogue: 0,0:28:44.84,0:28:50.93,Default,,0000,0000,0000,,important aspect. So that's how we\Nmanufactured and some of the failures
Dialogue: 0,0:28:50.93,0:28:58.91,Default,,0000,0000,0000,,we've had. Now let's talk about how not to\Nbuild A.I.. So as we all know, A.I. is
Dialogue: 0,0:28:58.91,0:29:03.19,Default,,0000,0000,0000,,magic, right? Just as Blockchain and IoT\Nand the cloud. It's absolutely magic,
Dialogue: 0,0:29:03.19,0:29:09.11,Default,,0000,0000,0000,,right? Well, the reality is it's it's not\Nthat magic. So we decided to have a very
Dialogue: 0,0:29:09.11,0:29:15.04,Default,,0000,0000,0000,,pragmatic approach to A.I.. We said, let's\Nnot do anything crazy. Let's just make
Dialogue: 0,0:29:15.04,0:29:20.02,Default,,0000,0000,0000,,something that works. So our very first\Niteration of a robot was this. This is
Dialogue: 0,0:29:20.02,0:29:24.03,Default,,0000,0000,0000,,like the control panel for a robot. It was\Nsuper simple. We had a video call coming
Dialogue: 0,0:29:24.03,0:29:27.98,Default,,0000,0000,0000,,in from the robot, on the left over there\Nis literally an iframe, super simple
Dialogue: 0,0:29:27.98,0:29:31.46,Default,,0000,0000,0000,,stuff. And on the right, we had a map, on\Nthe bottom we had some controls so you can
Dialogue: 0,0:29:31.46,0:29:36.23,Default,,0000,0000,0000,,move the robot forwards, backwards. It was\Nvery, very simple. It barely worked. On
Dialogue: 0,0:29:36.23,0:29:40.14,Default,,0000,0000,0000,,the robot we had our Arduino, Raspberry Pi\Nall running in python and the server was
Dialogue: 0,0:29:40.14,0:29:45.34,Default,,0000,0000,0000,,Java communicating over web sockets. But\Nthis barely worked. So we decided, OK,
Dialogue: 0,0:29:45.34,0:29:50.09,Default,,0000,0000,0000,,what can we do? Maybe we can build an\Nautonomous robot and we can both say that
Dialogue: 0,0:29:50.09,0:29:55.34,Default,,0000,0000,0000,,would work entirely by itself. We actually\Ndid that. So we built a robot that could
Dialogue: 0,0:29:55.34,0:30:00.01,Default,,0000,0000,0000,,go entirely by itself. It was fully\Nautonomous. And it was actually really
Dialogue: 0,0:30:00.01,0:30:04.30,Default,,0000,0000,0000,,cool. The way we built it is we had pretty\Nbeefy computer inside. We had it a Nvidia
Dialogue: 0,0:30:04.30,0:30:08.94,Default,,0000,0000,0000,,Jetson TX2. On that, we were running ROSS\Nand inside of ROSS we were running
Dialogue: 0,0:30:08.94,0:30:12.86,Default,,0000,0000,0000,,TensorFlow and a couple of other\Ntechnologies. We had YODA for object
Dialogue: 0,0:30:12.86,0:30:17.19,Default,,0000,0000,0000,,detection and some other cool tech that I\Nam not entirely familiar with it since I
Dialogue: 0,0:30:17.19,0:30:22.46,Default,,0000,0000,0000,,didn't write that code, but over here what\Nthe robot did is it looked at objects. So
Dialogue: 0,0:30:22.46,0:30:27.70,Default,,0000,0000,0000,,it was detecting objects. It was also\Nmeasuring the distance to the objects. And
Dialogue: 0,0:30:27.70,0:30:30.98,Default,,0000,0000,0000,,it also had an inference neural network.\NAnd you can see that on the top left of
Dialogue: 0,0:30:30.98,0:30:36.52,Default,,0000,0000,0000,,the screen here. Basically, based on\Ntrained data, it would know where not to
Dialogue: 0,0:30:36.52,0:30:40.80,Default,,0000,0000,0000,,drive into and it would try to plot a path\Nbased on 12 different directions it could
Dialogue: 0,0:30:40.80,0:30:44.98,Default,,0000,0000,0000,,go into. So it had 12 directions and it\Nwould go in the direction which had the
Dialogue: 0,0:30:44.98,0:30:50.12,Default,,0000,0000,0000,,highest probability of not colliding with\Nsomebody or something. And this worked,
Dialogue: 0,0:30:50.12,0:30:55.66,Default,,0000,0000,0000,,OK. We were able to get like 99 percent\Nautonomy. But the problem is, since we're
Dialogue: 0,0:30:55.66,0:31:00.12,Default,,0000,0000,0000,,doing a commercially viable delivery\Nservice, that's like offering deliveries
Dialogue: 0,0:31:00.12,0:31:04.19,Default,,0000,0000,0000,,to regular people and not something in the\Nlab, it really had to do something that
Dialogue: 0,0:31:04.19,0:31:07.47,Default,,0000,0000,0000,,worked all the time. And the challenge\Nwith this is we still needed to have
Dialogue: 0,0:31:07.47,0:31:10.97,Default,,0000,0000,0000,,people in the loop. We still had to have\Npeople who looked at the robot to make
Dialogue: 0,0:31:10.97,0:31:15.65,Default,,0000,0000,0000,,sure it would actually not crash. And what\Nhappens if you have something that's fully
Dialogue: 0,0:31:15.65,0:31:20.40,Default,,0000,0000,0000,,autonomous and people assume it works\Nwell, when it doesn't work well instead
Dialogue: 0,0:31:20.40,0:31:23.35,Default,,0000,0000,0000,,of looking at the screen and being ready\Nto take over, they're just looking at the
Dialogue: 0,0:31:23.35,0:31:28.95,Default,,0000,0000,0000,,phone and Instagram. So this approach\Nwasn't the best one. And instead, we
Dialogue: 0,0:31:28.95,0:31:33.89,Default,,0000,0000,0000,,decided to use a supervision approach. So\Nwe spent a lot of time building this. So
Dialogue: 0,0:31:33.89,0:31:37.88,Default,,0000,0000,0000,,this is our supervisors console and it's\Nactually really, really cool platform.
Dialogue: 0,0:31:37.88,0:31:42.48,Default,,0000,0000,0000,,It's a platform that allows you to connect\Nto a robot and the robot streams to you
Dialogue: 0,0:31:42.48,0:31:46.91,Default,,0000,0000,0000,,video over Web RTC or like the 4G network\Nand you're able to control it over web
Dialogue: 0,0:31:46.91,0:31:51.57,Default,,0000,0000,0000,,sockets. So the way to work is you'd have\Na supervisor that sets waypoints for the
Dialogue: 0,0:31:51.57,0:31:56.14,Default,,0000,0000,0000,,robot to follow. So the supervisor would\Nclick on the image and he or she would
Dialogue: 0,0:31:56.14,0:32:00.15,Default,,0000,0000,0000,,tell the robot to move 10 meters at a\Ntime. So typically they'd set waypoints
Dialogue: 0,0:32:00.15,0:32:04.24,Default,,0000,0000,0000,,every 5 to 10 seconds. It was a very\Ninteresting approach. We tried a couple of
Dialogue: 0,0:32:04.24,0:32:08.49,Default,,0000,0000,0000,,different approaches. We tried to do slam,\Nthat really did not work out for us. It
Dialogue: 0,0:32:08.49,0:32:15.22,Default,,0000,0000,0000,,took too much resources and it didn't give\Nus a significant gain. We tried other
Dialogue: 0,0:32:15.22,0:32:19.40,Default,,0000,0000,0000,,things as well. We tried traffic light\Ndetection. So we tried traffic light
Dialogue: 0,0:32:19.40,0:32:23.15,Default,,0000,0000,0000,,detection. There are some amazing models\Navailable online, some great Github repos.
Dialogue: 0,0:32:23.15,0:32:28.97,Default,,0000,0000,0000,,The problem is, yes, they do work on a\Nvery clean data set. But when you actually
Dialogue: 0,0:32:28.97,0:32:34.59,Default,,0000,0000,0000,,have data, we actually have a real life\Nscenario where we have like glare, you
Dialogue: 0,0:32:34.59,0:32:38.98,Default,,0000,0000,0000,,have rain, you have weird situations, you\Nhave homeless people. It doesn't really
Dialogue: 0,0:32:38.98,0:32:43.89,Default,,0000,0000,0000,,translate that well in the real world. So\Nwe kind of struggled with that. Instead,
Dialogue: 0,0:32:43.89,0:32:49.02,Default,,0000,0000,0000,,we actually had a more middle ground\Napproach. So we are able to detect traffic
Dialogue: 0,0:32:49.02,0:32:53.89,Default,,0000,0000,0000,,lights really well, but we're not able to\Ndetect the color really well or the which
Dialogue: 0,0:32:53.89,0:32:57.77,Default,,0000,0000,0000,,kind of signal it's giving. So instead,\Nall we do over here, this automatically
Dialogue: 0,0:32:57.77,0:33:01.98,Default,,0000,0000,0000,,zooms in to traffic lights. So it's very\Neasy to see. This video actually that
Dialogue: 0,0:33:01.98,0:33:07.04,Default,,0000,0000,0000,,you're seeing is transmitted over very low\Nframe rate, very low bit rate as well. I
Dialogue: 0,0:33:07.04,0:33:13.96,Default,,0000,0000,0000,,think we're doing 480p at 100 kilobytes a\Nsecond. So it's very, very low bit rate.
Dialogue: 0,0:33:13.96,0:33:17.65,Default,,0000,0000,0000,,And when the robot isn't moving, we\Nactually make it go black and white and
Dialogue: 0,0:33:17.65,0:33:22.13,Default,,0000,0000,0000,,even lower rate frame rate so that it\Ndoesn't waste resources. So yeah, it's
Dialogue: 0,0:33:22.13,0:33:28.11,Default,,0000,0000,0000,,pretty cool stuff. Over here on the top\Nleft we actualy have our latency. So we
Dialogue: 0,0:33:28.11,0:33:31.43,Default,,0000,0000,0000,,managed to build the infrastructure that\Nallowed us to supervise this robots from
Dialogue: 0,0:33:31.43,0:33:36.59,Default,,0000,0000,0000,,Columbia for 200 milliseconds and less than 20\Nmilliseconds. So it's like a blink of an
Dialogue: 0,0:33:36.59,0:33:40.93,Default,,0000,0000,0000,,eye. It was a really, really cool\Ntechnology, it worked or 4G and we did a
Dialogue: 0,0:33:40.93,0:33:45.89,Default,,0000,0000,0000,,lot to optimize that. We had also a map\Nover here. So this map is really, really
Dialogue: 0,0:33:45.89,0:33:50.13,Default,,0000,0000,0000,,cool. A lot of people ask us like, hey,\Ndid you do mapping? Did you map out your
Dialogue: 0,0:33:50.13,0:33:56.01,Default,,0000,0000,0000,,environment? Did you need to have\Nsomething there before you came into a new
Dialogue: 0,0:33:56.01,0:34:02.26,Default,,0000,0000,0000,,place? And well the answer is no. But what\Nwe do instead is we actually map out the
Dialogue: 0,0:34:02.26,0:34:08.19,Default,,0000,0000,0000,,network conditions. So we would map out\Nthe network conditions of a city and we'd
Dialogue: 0,0:34:08.19,0:34:11.94,Default,,0000,0000,0000,,say, OK, these areas like over here. This\Nis like high latency. We should avoid
Dialogue: 0,0:34:11.94,0:34:15.84,Default,,0000,0000,0000,,those areas because the robot could get\Nstuck there. It is actually very
Dialogue: 0,0:34:15.84,0:34:19.77,Default,,0000,0000,0000,,interesting to see the network conditions\Nchange continuously. You didn't have the
Dialogue: 0,0:34:19.77,0:34:23.23,Default,,0000,0000,0000,,same network conditions every day, all\Nday, all year. They'd actually change
Dialogue: 0,0:34:23.23,0:34:26.30,Default,,0000,0000,0000,,every few hours. So it was something that\Ntook us a while to figure out.
Dialogue: 0,0:34:26.30,0:34:30.23,Default,,0000,0000,0000,,{\i1}Takes a sip of Mate{\i0}\NSo, of course, the way this works is we
Dialogue: 0,0:34:30.23,0:34:33.82,Default,,0000,0000,0000,,had two or three people supervising,\Nsorry, two or three robots for a
Dialogue: 0,0:34:33.82,0:34:38.09,Default,,0000,0000,0000,,supervisor in Colombia, and we have just a\Nbunch of people. Typically, students who
Dialogue: 0,0:34:38.09,0:34:41.36,Default,,0000,0000,0000,,would just be working part time and they\Nwere sitting in an office in Colombia
Dialogue: 0,0:34:41.36,0:34:47.58,Default,,0000,0000,0000,,doing this. Of course, the press found out\Nabout this and they wrote a very small bit
Dialogue: 0,0:34:47.58,0:34:53.27,Default,,0000,0000,0000,,of text in the article saying like, oh,\NKiwi hires Colombians and pays them two
Dialogue: 0,0:34:53.27,0:34:59.33,Default,,0000,0000,0000,,dollars an hour. And people were really\Nfrustrated about that. We had a lot of
Dialogue: 0,0:34:59.33,0:35:03.50,Default,,0000,0000,0000,,interesting feedback about that. But what\Nwas interesting to see is that this
Dialogue: 0,0:35:03.50,0:35:07.89,Default,,0000,0000,0000,,technology actually helps people in\NColombia. If you're there, it's a third
Dialogue: 0,0:35:07.89,0:35:11.05,Default,,0000,0000,0000,,world country it's a developing country.\NYou can get a job at a factory. You can
Dialogue: 0,0:35:11.05,0:35:15.38,Default,,0000,0000,0000,,get a job at a textile shop, you can get a\Njob maybe McDonald's. But there aren't
Dialogue: 0,0:35:15.38,0:35:21.86,Default,,0000,0000,0000,,that many tech jobs per say. The biggest\Nemployer in the country is a phone support
Dialogue: 0,0:35:21.86,0:35:25.95,Default,,0000,0000,0000,,company. So like when you call in to\Nsupport line, you get connected to
Dialogue: 0,0:35:25.95,0:35:29.28,Default,,0000,0000,0000,,Colombia sometimes. And that's the biggest\Nemployer in the country. So in order to
Dialogue: 0,0:35:29.28,0:35:33.27,Default,,0000,0000,0000,,get like a tech job, it's really, really\Nhard and giving people the ability to, go
Dialogue: 0,0:35:33.27,0:35:38.43,Default,,0000,0000,0000,,and supervise robots it's something that\Nhelped them get something on their CV and
Dialogue: 0,0:35:38.43,0:35:41.82,Default,,0000,0000,0000,,help them step up It helped them learn a\Nlittle bit more about the technology and
Dialogue: 0,0:35:41.82,0:35:48.80,Default,,0000,0000,0000,,helped them progress in terms of their\Ncareers. Our lead A.I. guy, he actually
Dialogue: 0,0:35:48.80,0:35:53.29,Default,,0000,0000,0000,,started off as a supervisor and he went up\Nthrough the ranks and then he ended up
Dialogue: 0,0:35:53.29,0:35:56.59,Default,,0000,0000,0000,,leading the A.I. and robotics team. So it\Nwas really interesting, really inspiring
Dialogue: 0,0:35:56.59,0:36:01.25,Default,,0000,0000,0000,,to see how that transition happened. And\Nwe managed to get our technology to work
Dialogue: 0,0:36:01.25,0:36:09.34,Default,,0000,0000,0000,,so well that we can do this.\N{\i1}Video of the inside of an airplane is shown{\i0}
Dialogue: 0,0:36:09.34,0:36:14.77,Default,,0000,0000,0000,,So we were able to get it to work with up to\Neight seconds latency, which meant that
Dialogue: 0,0:36:14.77,0:36:17.86,Default,,0000,0000,0000,,you can control it literally from anywhere\Nin the world. So even from an airplane
Dialogue: 0,0:36:17.86,0:36:24.36,Default,,0000,0000,0000,,above the Pacific Ocean. So it was a\Nreally interesting experience. And we
Dialogue: 0,0:36:24.36,0:36:30.24,Default,,0000,0000,0000,,really try to make it simple. So in\Nconclusion, for A.I., we realized that the
Dialogue: 0,0:36:30.24,0:36:34.03,Default,,0000,0000,0000,,best approach was to keep it simple. We\Ntried a lot, a lot of different
Dialogue: 0,0:36:34.03,0:36:38.18,Default,,0000,0000,0000,,approaches, like we tried the traffic\Nlight detection. We tried a yellow pad
Dialogue: 0,0:36:38.18,0:36:42.68,Default,,0000,0000,0000,,detection. I didn't mention that. So in\NBerkeley, you have these accessibility
Dialogue: 0,0:36:42.68,0:36:46.65,Default,,0000,0000,0000,,ramps and you have yellow pads that blind\Npeople can actually feel them and see them
Dialogue: 0,0:36:46.65,0:36:51.91,Default,,0000,0000,0000,,easier. So we built the algorithm to\Ndetect that and we thought, OK, maybe if
Dialogue: 0,0:36:51.91,0:36:56.35,Default,,0000,0000,0000,,the robot is stuck in the middle of the\Nintersection, you can automatically detect
Dialogue: 0,0:36:56.35,0:37:00.52,Default,,0000,0000,0000,,this yellow pattern and navigate to it.\NIt's an approach that worked in theory, in
Dialogue: 0,0:37:00.52,0:37:06.26,Default,,0000,0000,0000,,practice it did not quite work. We tried\Nsegmentation. So that was an approach that
Dialogue: 0,0:37:06.26,0:37:12.07,Default,,0000,0000,0000,,worked OK. But some weird things broke it.\NSo for example, any lamp posts or bicycle
Dialogue: 0,0:37:12.07,0:37:17.38,Default,,0000,0000,0000,,posts would crash the robot because it\Ndidn't see it. So yeah, keeping it simple
Dialogue: 0,0:37:17.38,0:37:22.59,Default,,0000,0000,0000,,was the best approach, really not going\Ntoo crazy. And the approach we ended up
Dialogue: 0,0:37:22.59,0:37:27.16,Default,,0000,0000,0000,,going in the end was to have it more of\Nlike a driver assist type, like a parallel
Dialogue: 0,0:37:27.16,0:37:32.38,Default,,0000,0000,0000,,approach, parallel autonomy approach,\Nwhere our robots would help people the
Dialogue: 0,0:37:32.38,0:37:37.00,Default,,0000,0000,0000,,same way that cars would help people stay\Nin lanes or have cruise control or like
Dialogue: 0,0:37:37.00,0:37:39.46,Default,,0000,0000,0000,,with parking assistance. That's kind of\Nthe approach we're having. I think long
Dialogue: 0,0:37:39.46,0:37:43.62,Default,,0000,0000,0000,,term it is gonna be possible to build\Nrobots more autonomous, it could be
Dialogue: 0,0:37:43.62,0:37:48.72,Default,,0000,0000,0000,,Starship that have some interesting ideas\Nabout how to solve that. But I don't think
Dialogue: 0,0:37:48.72,0:37:53.22,Default,,0000,0000,0000,,it's quite something I could be scaled to\Nevery city just yet. Another really
Dialogue: 0,0:37:53.22,0:37:58.90,Default,,0000,0000,0000,,important thing is, the lab does not equal\Nthe real world. So there are many, many
Dialogue: 0,0:37:58.90,0:38:04.87,Default,,0000,0000,0000,,great examples of fantastic research\Npapers from some great groups and they
Dialogue: 0,0:38:04.87,0:38:09.75,Default,,0000,0000,0000,,were great with very polished, very clean\Ndatasets. But they did not work when you
Dialogue: 0,0:38:09.75,0:38:14.74,Default,,0000,0000,0000,,deployed them on 100 robots, there were\Nall different. They all had slightly
Dialogue: 0,0:38:14.74,0:38:18.47,Default,,0000,0000,0000,,different camera calibration that all had\Nslightly different hardware, it all had
Dialogue: 0,0:38:18.47,0:38:23.25,Default,,0000,0000,0000,,slightly different chassis. It did not\Nreally translate as well. So these
Dialogue: 0,0:38:23.25,0:38:28.64,Default,,0000,0000,0000,,algorithms, these lab best case scenarios,\Nwill need to be modified a little bit.
Dialogue: 0,0:38:28.64,0:38:35.05,Default,,0000,0000,0000,,What else? Yeah, one thing, maybe jumping\Nback to the keep it simple. We decided to
Dialogue: 0,0:38:35.05,0:38:39.79,Default,,0000,0000,0000,,put in a very simple safety mechanism. So\Nthe robot actually breaks if it sees
Dialogue: 0,0:38:39.79,0:38:43.66,Default,,0000,0000,0000,,something within 50 centimeters in front\Nof it. So as kind of like a last measure,
Dialogue: 0,0:38:43.66,0:38:47.11,Default,,0000,0000,0000,,precaution, as you saw before, there is a\Nvideo like you can supervise the robot
Dialogue: 0,0:38:47.11,0:38:51.50,Default,,0000,0000,0000,,from anywhere in the world, but a lot of\Nlatency. But having this 50 centimeter
Dialogue: 0,0:38:51.50,0:38:56.78,Default,,0000,0000,0000,,like hard break, actually saves us in case\Nthe robot loses connectivity or the
Dialogue: 0,0:38:56.78,0:39:01.68,Default,,0000,0000,0000,,supervisor is no longer able to supervise\Nthe robot. So it's always breaking 50
Dialogue: 0,0:39:01.68,0:39:07.94,Default,,0000,0000,0000,,centimeters away from any collision with\Nlike a baby or a car or whatever. So the
Dialogue: 0,0:39:07.94,0:39:12.42,Default,,0000,0000,0000,,approach we really thought about is, how\Ncan we expand human potential? There is a
Dialogue: 0,0:39:12.42,0:39:18.84,Default,,0000,0000,0000,,lot of talk about A.I. taking jobs or A.I.\Nreplacing people's roles, but we sort of
Dialogue: 0,0:39:18.84,0:39:22.29,Default,,0000,0000,0000,,kind of try to do that and it didn't work.\NLike we try to build robots that were
Dialogue: 0,0:39:22.29,0:39:26.17,Default,,0000,0000,0000,,fully autonomous that went from the\Nrestaurant to your door and that didn't
Dialogue: 0,0:39:26.17,0:39:29.19,Default,,0000,0000,0000,,work. People were waiting a very long\Ntime. These robots required an obscene
Dialogue: 0,0:39:29.19,0:39:34.23,Default,,0000,0000,0000,,amount of maintenance. So we ended up\Ngoing for an approach that was far more
Dialogue: 0,0:39:34.23,0:39:37.85,Default,,0000,0000,0000,,parallel autonomy where these robots were\Nlike helping people to do more. Same way
Dialogue: 0,0:39:37.85,0:39:41.98,Default,,0000,0000,0000,,the supervisors are getting these\Nassistive technologies where they able to
Dialogue: 0,0:39:41.98,0:39:45.53,Default,,0000,0000,0000,,set a waypoint to do the path finding and\Nthe robot does the motion planning on
Dialogue: 0,0:39:45.53,0:39:49.50,Default,,0000,0000,0000,,board. We also had the couriers who would\Njust load food into the robots instead of
Dialogue: 0,0:39:49.50,0:39:54.19,Default,,0000,0000,0000,,the robots picking up food from the\Nrestaurant directly, so really expanding
Dialogue: 0,0:39:54.19,0:39:57.62,Default,,0000,0000,0000,,human potential. I think that's where it's\Nat. And over the course of the past
Dialogue: 0,0:39:57.62,0:40:01.27,Default,,0000,0000,0000,,century, we've seen a lot of examples of\Nthis. Like we've seen operators of
Dialogue: 0,0:40:01.27,0:40:05.83,Default,,0000,0000,0000,,elevators. Like before, elevators had\Noperators who would make it go up or down.
Dialogue: 0,0:40:05.83,0:40:09.32,Default,,0000,0000,0000,,And now they're fully automated. We had\Nswitchboard operators who were there to
Dialogue: 0,0:40:09.32,0:40:12.21,Default,,0000,0000,0000,,connect phone calls. Now we can make a\Nphone call to anywhere in the world
Dialogue: 0,0:40:12.21,0:40:17.26,Default,,0000,0000,0000,,instantly for free. So we're seeing this\Ntransformation of work and transformation
Dialogue: 0,0:40:17.26,0:40:21.26,Default,,0000,0000,0000,,of the way things are done. And I think\Nthis is just the start. The way I see
Dialogue: 0,0:40:21.26,0:40:27.14,Default,,0000,0000,0000,,these robots is really meshing into the\Nfabric of our societies and solving
Dialogue: 0,0:40:27.14,0:40:31.83,Default,,0000,0000,0000,,physical transportation. Like, sure, you\Ncan move bits from anywhere to anywhere in
Dialogue: 0,0:40:31.83,0:40:35.77,Default,,0000,0000,0000,,the world, but can you move atoms? It's\Nreally expensive to do that. It's really
Dialogue: 0,0:40:35.77,0:40:43.63,Default,,0000,0000,0000,,hard to do that. That's why I see robots\Nexpanding human potential. So.
Dialogue: 0,0:40:43.63,0:40:50.36,Default,,0000,0000,0000,,Conclusions. What we did was really cool\Nand I think it was a cool experience. One
Dialogue: 0,0:40:50.36,0:40:55.65,Default,,0000,0000,0000,,thing that we realized is that tech isn't\Nthe hardest part, right? We spent a lot of
Dialogue: 0,0:40:55.65,0:41:01.63,Default,,0000,0000,0000,,time thinking how to build something, but\Nfiguring what to build is sometimes very
Dialogue: 0,0:41:01.63,0:41:05.93,Default,,0000,0000,0000,,important as well. And I don't think we\Nspent enough time asking ourselves that
Dialogue: 0,0:41:05.93,0:41:09.84,Default,,0000,0000,0000,,question. We kind of went in all sorts of\Ndirections we didn't focus as much on
Dialogue: 0,0:41:09.84,0:41:15.59,Default,,0000,0000,0000,,making the best product possible. We kind\Nof tried things that were really weird and
Dialogue: 0,0:41:15.59,0:41:18.63,Default,,0000,0000,0000,,not well thought out. So like having that\Nmore long term thinking, like thinking
Dialogue: 0,0:41:18.63,0:41:22.18,Default,,0000,0000,0000,,what should we build is very important\Nbecause like how, you can just look up a
Dialogue: 0,0:41:22.18,0:41:27.35,Default,,0000,0000,0000,,tutorial on Google and figure out how to\Nbuild robots. It's not the end of the
Dialogue: 0,0:41:27.35,0:41:31.31,Default,,0000,0000,0000,,world. One really important thing for us\Nwas interaction. So interacting with
Dialogue: 0,0:41:31.31,0:41:34.72,Default,,0000,0000,0000,,people, figuring out how to make the door\Nopen, when you actually received your food
Dialogue: 0,0:41:34.72,0:41:39.87,Default,,0000,0000,0000,,was super hard, super, super challenging\Nto do. Actually, the only robot that opens
Dialogue: 0,0:41:39.87,0:41:43.72,Default,,0000,0000,0000,,the door for you. Other companies like\NStarship, for example, they have a button
Dialogue: 0,0:41:43.72,0:41:46.73,Default,,0000,0000,0000,,that unlocks a solenoid. So it's like the\Nexperience is not quite there yet to bend
Dialogue: 0,0:41:46.73,0:41:51.03,Default,,0000,0000,0000,,it down. You have to figure out how the\Ndoor actually opens. So we spent a lot of
Dialogue: 0,0:41:51.03,0:41:55.40,Default,,0000,0000,0000,,time, a lot of effort in order to optimize\Nthat experience to make it as smooth as
Dialogue: 0,0:41:55.40,0:42:00.63,Default,,0000,0000,0000,,possible for people. Also, one thing we\Ndidn't figure out is financing. I'll come
Dialogue: 0,0:42:00.63,0:42:04.44,Default,,0000,0000,0000,,back to that in a second. That was really,\Nreally hard to do as well. So like tech,
Dialogue: 0,0:42:04.44,0:42:08.13,Default,,0000,0000,0000,,you know, not the hardest, financing\Nfiguring out like how to manage cash flow,
Dialogue: 0,0:42:08.13,0:42:12.30,Default,,0000,0000,0000,,super important. But I think the most\Nimportant thing is to work with a great
Dialogue: 0,0:42:12.30,0:42:17.47,Default,,0000,0000,0000,,team. If you're going to be spending a lot\Nof time with people who you eat, live and
Dialogue: 0,0:42:17.47,0:42:21.65,Default,,0000,0000,0000,,breathe with, it's really important to\Nchoose a team that you really connect with
Dialogue: 0,0:42:21.65,0:42:25.88,Default,,0000,0000,0000,,and then share the same passion as you do,\Nbecause you could be miserable making an
Dialogue: 0,0:42:25.88,0:42:30.32,Default,,0000,0000,0000,,amazing amount of money, but if your with\Na really crappy team with a high turnover,
Dialogue: 0,0:42:30.32,0:42:32.75,Default,,0000,0000,0000,,it's really boring. I was really fortunate\Nto work with one of the best teams in the
Dialogue: 0,0:42:32.75,0:42:36.80,Default,,0000,0000,0000,,world and over the course of the past two\Nand a half years we managed to do quite a
Dialogue: 0,0:42:36.80,0:42:41.31,Default,,0000,0000,0000,,lot. And just last month we actually got\Nan article in The New York Times. So that
Dialogue: 0,0:42:41.31,0:42:44.89,Default,,0000,0000,0000,,was a really big accomplishment for our\Nteam and we got to share it with our
Dialogue: 0,0:42:44.89,0:42:50.31,Default,,0000,0000,0000,,families. My mom was really proud. So a\Nlot of great traction and a lot of great
Dialogue: 0,0:42:50.31,0:42:54.56,Default,,0000,0000,0000,,coverage. But unfortunately, we actually\Nran out of money, so we kind of ran out of
Dialogue: 0,0:42:54.56,0:43:00.30,Default,,0000,0000,0000,,money last month and we are no longer\Ndelivering things. So I decided to leave
Dialogue: 0,0:43:00.30,0:43:05.73,Default,,0000,0000,0000,,and start my own thing instead of doing\Nrobots. I decided to do data. So now I'm
Dialogue: 0,0:43:05.73,0:43:11.14,Default,,0000,0000,0000,,actually focusing more on building a tool\Nthat helps you tell stories with data. So
Dialogue: 0,0:43:11.14,0:43:16.45,Default,,0000,0000,0000,,this is Glint. This is a data storytelling\Ntool. You're able to drag in some files
Dialogue: 0,0:43:16.45,0:43:21.90,Default,,0000,0000,0000,,and it tells you the story of your data\Nwithout you having to write any code. So
Dialogue: 0,0:43:21.90,0:43:25.83,Default,,0000,0000,0000,,my hope for this is to allow anybody in\Nthe world without any knowledge about how
Dialogue: 0,0:43:25.83,0:43:29.50,Default,,0000,0000,0000,,to wrangle data, how to clean data, how to\Nanalyze data, to be able to tell stories
Dialogue: 0,0:43:29.50,0:43:35.32,Default,,0000,0000,0000,,with their data directly from their\Ncomputers. I'm imagining a tool where you
Dialogue: 0,0:43:35.32,0:43:41.75,Default,,0000,0000,0000,,can say, oh, "In December there were X X\Nvisitors to Congress" or "Last summer we
Dialogue: 0,0:43:41.75,0:43:45.31,Default,,0000,0000,0000,,had X X sales" and automatically filled\Nout for you. That's kind of what I'm
Dialogue: 0,0:43:45.31,0:43:49.13,Default,,0000,0000,0000,,thinking about. If you want to join the\Neffort, there is a Github. I'm more than
Dialogue: 0,0:43:49.13,0:43:53.88,Default,,0000,0000,0000,,happy to have any contributors. And if you\Nhave any questions or comments we're happy
Dialogue: 0,0:43:53.88,0:44:02.28,Default,,0000,0000,0000,,to answer on Twitter or here in person.\NThank you.
Dialogue: 0,0:44:02.28,0:44:11.37,Default,,0000,0000,0000,,{\i1}Applause{\i0}
Dialogue: 0,0:44:11.37,0:44:18.46,Default,,0000,0000,0000,,H: So, as usual, feel free to line up in\Nfront of the microphones or write your
Dialogue: 0,0:44:18.46,0:44:23.97,Default,,0000,0000,0000,,question to the signal angel over there.\NThat already has one. Um, it's all the way
Dialogue: 0,0:44:23.97,0:44:28.69,Default,,0000,0000,0000,,down. Go ahead.\NSignal Angel: OK. Here is a user of your
Dialogue: 0,0:44:28.69,0:44:34.34,Default,,0000,0000,0000,,service who apparently got an e-mail from\Nyou that announced some changes. So he's
Dialogue: 0,0:44:34.34,0:44:38.20,Default,,0000,0000,0000,,wondering, what's up? What you're going,\Nwhat you're planning to do there, whether
Dialogue: 0,0:44:38.20,0:44:40.30,Default,,0000,0000,0000,,you're continuing your service or closing\Nshop?
Dialogue: 0,0:44:40.30,0:44:45.02,Default,,0000,0000,0000,,S: Yeah, it's unclear. We ran out of\Nfunds, so I think the CEO is still trying
Dialogue: 0,0:44:45.02,0:44:48.86,Default,,0000,0000,0000,,to figure out what to do with that. I wish\Nhim the best of luck, but I ended up
Dialogue: 0,0:44:48.86,0:44:52.84,Default,,0000,0000,0000,,leaving with a lot of other people. So we\Nhave like 50 people in November, now we
Dialogue: 0,0:44:52.84,0:44:57.15,Default,,0000,0000,0000,,have like 10 people left in the country.\NSo it's very ambiguous what's happening,
Dialogue: 0,0:44:57.15,0:45:01.63,Default,,0000,0000,0000,,but yeah, I left.\NH: Yeah, microphone?
Dialogue: 0,0:45:01.63,0:45:08.91,Default,,0000,0000,0000,,Audience Member 1: No audio. Okay, now it\Nworks. I'm a little bit confused because
Dialogue: 0,0:45:08.91,0:45:14.54,Default,,0000,0000,0000,,you are presenting a 1970s concept of a\Nmanipulator, because a robot is something
Dialogue: 0,0:45:14.54,0:45:18.69,Default,,0000,0000,0000,,that works by itself, a manipulator is\Nsomebody who has some joysticks and moves
Dialogue: 0,0:45:18.69,0:45:22.93,Default,,0000,0000,0000,,things. So it's nothing special. You just\Nhave a interlinked Internet link for
Dialogue: 0,0:45:22.93,0:45:27.04,Default,,0000,0000,0000,,manipulator and in the 70s, there were\Ncables. So what's the special thing?
Dialogue: 0,0:45:27.04,0:45:30.73,Default,,0000,0000,0000,,S: Yeah, that's a good question. I think\Nthe magic here is connecting everything
Dialogue: 0,0:45:30.73,0:45:34.51,Default,,0000,0000,0000,,together, figuring out for us how to build\Nthese robots, how to build a reliable
Dialogue: 0,0:45:34.51,0:45:38.64,Default,,0000,0000,0000,,connection, and how about a platform that\Nworks. And as I mentioned, like the how
Dialogue: 0,0:45:38.64,0:45:43.31,Default,,0000,0000,0000,,that's not that interesting. It's more of\Nthe what you build. It's that experience
Dialogue: 0,0:45:43.31,0:45:46.19,Default,,0000,0000,0000,,where you're able to order anything you\Nwant at any time and get it delivered in
Dialogue: 0,0:45:46.19,0:45:48.89,Default,,0000,0000,0000,,under 30 minutes virtually for free. So\Nthat's good.
Dialogue: 0,0:45:48.89,0:45:53.45,Default,,0000,0000,0000,,AM1: So, so far, so good, but evil people\Ncould just buy a remote control car, put a
Dialogue: 0,0:45:53.45,0:45:59.63,Default,,0000,0000,0000,,bomb in it, drive under a police car and\Nmake boom. And so it's the same use case.
Dialogue: 0,0:45:59.63,0:46:06.52,Default,,0000,0000,0000,,You deliver something by remote control.\NAudience Member 2: Yeah. You talked about
Dialogue: 0,0:46:06.52,0:46:13.91,Default,,0000,0000,0000,,iterating quickly and rapidly and that's\Nvery good model for conceptual stage and
Dialogue: 0,0:46:13.91,0:46:18.53,Default,,0000,0000,0000,,software. Were you in the stage where you\Nwere leasing your hardware with your
Dialogue: 0,0:46:18.53,0:46:22.58,Default,,0000,0000,0000,,iterations? Because usually a thick stack\Nof certification has to come in between.
Dialogue: 0,0:46:22.58,0:46:27.64,Default,,0000,0000,0000,,S:So I'm not entirely sure. Are you asking\Nif we got certified at every single
Dialogue: 0,0:46:27.64,0:46:30.72,Default,,0000,0000,0000,,release?\NAM2: I suppose. Yeah. What level of like
Dialogue: 0,0:46:30.72,0:46:36.08,Default,,0000,0000,0000,,recertification was it totally released.\NSo you had to meet like regulations for
Dialogue: 0,0:46:36.08,0:46:39.66,Default,,0000,0000,0000,,each iteration of that?\NS: Yeah, absolutely. We didn't really get
Dialogue: 0,0:46:39.66,0:46:43.47,Default,,0000,0000,0000,,certified because we're not building\Nhardware product for consumers. So we're
Dialogue: 0,0:46:43.47,0:46:46.67,Default,,0000,0000,0000,,not selling it to anybody. We're operating\Nit ourselves. So we don't fit under the
Dialogue: 0,0:46:46.67,0:46:50.94,Default,,0000,0000,0000,,same kind of requirements. However, we did\Nhave to have some permits. And part of the
Dialogue: 0,0:46:50.94,0:46:54.24,Default,,0000,0000,0000,,conditions that these permits was that we\Nhad to meet some expectations. But they're
Dialogue: 0,0:46:54.24,0:46:58.32,Default,,0000,0000,0000,,very, very basic. And there were rigid\Nlike an FCC or a CE certification, for
Dialogue: 0,0:46:58.32,0:47:01.12,Default,,0000,0000,0000,,example.\NAM2: That was the question. Yeah. Thanks.
Dialogue: 0,0:47:01.12,0:47:07.32,Default,,0000,0000,0000,,S: Thank you.\NSA: Another question from the Internet.
Dialogue: 0,0:47:07.32,0:47:11.52,Default,,0000,0000,0000,,"Why did you develop different\Napplications for Android and iOS?"
Dialogue: 0,0:47:11.52,0:47:18.04,Default,,0000,0000,0000,,S: For the consumer application?\NSA: I haven't got any more details.
Dialogue: 0,0:47:18.04,0:47:25.29,Default,,0000,0000,0000,,S: We just did. I mean, we had first an\NiOS application. I mean, 80 percent of our
Dialogue: 0,0:47:25.29,0:47:29.03,Default,,0000,0000,0000,,customers are using iOS. So we really\Nspent a lot of effort like polishing that
Dialogue: 0,0:47:29.03,0:47:33.62,Default,,0000,0000,0000,,iOS experience, making sure that worked.\NAnd at one point, our Android app was
Dialogue: 0,0:47:33.62,0:47:39.24,Default,,0000,0000,0000,,working super badly. So we decided to kill\Nit. And everybody was really, really
Dialogue: 0,0:47:39.24,0:47:43.73,Default,,0000,0000,0000,,pissed off, extremely pissed off. So we\Nactually reintroduced it and we started
Dialogue: 0,0:47:43.73,0:47:47.86,Default,,0000,0000,0000,,catching up with features to the iOS\Nversion. Internally, all of our apps are
Dialogue: 0,0:47:47.86,0:47:53.85,Default,,0000,0000,0000,,built in React and React native. So we had\Na common framework for all of our internal
Dialogue: 0,0:47:53.85,0:47:56.89,Default,,0000,0000,0000,,apps, but we didn't have that experience.\NWhere you're expecting the quality of
Dialogue: 0,0:47:56.89,0:48:02.10,Default,,0000,0000,0000,,experience, that we're expecting from a\Nconsumer app using React. That's why we
Dialogue: 0,0:48:02.10,0:48:13.65,Default,,0000,0000,0000,,had two different code bases.\NH: Have tried different methods regarding
Dialogue: 0,0:48:13.65,0:48:19.52,Default,,0000,0000,0000,,perception? For example, lidar, radar and\Nwhat are your conclusions from that?
Dialogue: 0,0:48:19.52,0:48:24.03,Default,,0000,0000,0000,,S: Yeah, we tried lidar, we tried the\Ncheap lidar we didn't try the really high
Dialogue: 0,0:48:24.03,0:48:27.74,Default,,0000,0000,0000,,end lidar. So the challenge with having\Nlike point clouds is that you have to
Dialogue: 0,0:48:27.74,0:48:32.57,Default,,0000,0000,0000,,compute, spent a lot of time competing. We\Nwere using a relatively low power device
Dialogue: 0,0:48:32.57,0:48:35.90,Default,,0000,0000,0000,,and it was running from batteries. So we\Ndidn't have the luxury of having like 10
Dialogue: 0,0:48:35.90,0:48:40.61,Default,,0000,0000,0000,,GPUs is in the trunk of a car, for\Nexample. So that was one approach. One
Dialogue: 0,0:48:40.61,0:48:45.91,Default,,0000,0000,0000,,question. Another question is how much\Ndoes it cost? So lidars, they can cost ten
Dialogue: 0,0:48:45.91,0:48:49.04,Default,,0000,0000,0000,,thousand, hundred thousand dollars. Our\Nbill of materials was 'round two and a
Dialogue: 0,0:48:49.04,0:48:53.14,Default,,0000,0000,0000,,half thousand. The last versions are two\Nand a half thousand. So all of our sensors
Dialogue: 0,0:48:53.14,0:48:57.77,Default,,0000,0000,0000,,were very minimal. In terms of what\Nsensors we tried? We tried a lot of
Dialogue: 0,0:48:57.77,0:49:04.48,Default,,0000,0000,0000,,different sensors. We tried ultrasonic\Nsensors. We tried near field infrared
Dialogue: 0,0:49:04.48,0:49:11.36,Default,,0000,0000,0000,,sensors. We tried other sensors. Yeah, we\Ntried a lot of different sensors. We are
Dialogue: 0,0:49:11.36,0:49:15.88,Default,,0000,0000,0000,,ended up just going with cameras. So we\Nhave cameras. We have six cameras onboard,
Dialogue: 0,0:49:15.88,0:49:22.88,Default,,0000,0000,0000,,all of them full HD. We stitch them into\Nan image on our compute module and then
Dialogue: 0,0:49:22.88,0:49:25.78,Default,,0000,0000,0000,,the supervisor decides which portion of\Nthe image they want streams. They can
Dialogue: 0,0:49:25.78,0:49:29.62,Default,,0000,0000,0000,,manipulate with the keyboard to see which\Nportion of the image is streamed. So we
Dialogue: 0,0:49:29.62,0:49:32.54,Default,,0000,0000,0000,,don't stream the whole image. We just\Nstream a part of it. The really important
Dialogue: 0,0:49:32.54,0:49:36.74,Default,,0000,0000,0000,,part for us was to make something that's\Nviable, that can be used commercially. I'm
Dialogue: 0,0:49:36.74,0:49:40.72,Default,,0000,0000,0000,,sure lidar is really cool, but I'm not\Nseeing any commercial deployments of lidar
Dialogue: 0,0:49:40.72,0:49:45.71,Default,,0000,0000,0000,,based autonomous vehicles or robots yet.\NAudience Member 3: Thank you.
Dialogue: 0,0:49:45.71,0:49:54.50,Default,,0000,0000,0000,,Audience Member 4: You've tried out many\Ndifferent concepts how to do it. And you
Dialogue: 0,0:49:54.50,0:49:59.49,Default,,0000,0000,0000,,saw that your company ran out of money. Do\Nyou still believe in the business concept
Dialogue: 0,0:49:59.49,0:50:06.78,Default,,0000,0000,0000,,of robots delivering packages of food?\NS: Who knows? I think I think it was a
Dialogue: 0,0:50:06.78,0:50:11.50,Default,,0000,0000,0000,,great learning experience. We learned a\Nlot. We had a great team. And I think
Dialogue: 0,0:50:11.50,0:50:14.23,Default,,0000,0000,0000,,we'll see some concept of robots. Maybe\Nnot exactly what we were building, maybe
Dialogue: 0,0:50:14.23,0:50:17.84,Default,,0000,0000,0000,,something a little bit different, but I\Nthink it's a little bit inevitable,
Dialogue: 0,0:50:17.84,0:50:20.53,Default,,0000,0000,0000,,especially with the rise of self-driving\Ncars. Maybe we'll have cars delivering
Dialogue: 0,0:50:20.53,0:50:24.76,Default,,0000,0000,0000,,packages instead of robots. Not entirely\Nsure what it would look like. I could tell
Dialogue: 0,0:50:24.76,0:50:27.86,Default,,0000,0000,0000,,you, Amazon, they bought one of our\Ncompetitors dispatch labs. So they're
Dialogue: 0,0:50:27.86,0:50:31.65,Default,,0000,0000,0000,,making a big bet on this. There are two\Ndelivery companies in the US, Postmates
Dialogue: 0,0:50:31.65,0:50:35.80,Default,,0000,0000,0000,,and DoorDash, that are building products\Ninternally also for... with delivery
Dialogue: 0,0:50:35.80,0:50:39.52,Default,,0000,0000,0000,,robots. And also companies like FedEx are\Nalso building delivery robots. And then
Dialogue: 0,0:50:39.52,0:50:42.90,Default,,0000,0000,0000,,we have companies as Starship, for\Nexample, which are building robots and
Dialogue: 0,0:50:42.90,0:50:47.59,Default,,0000,0000,0000,,doing B2B with companies all over the\Nworld. So, yeah, I think we'll see some
Dialogue: 0,0:50:47.59,0:50:50.61,Default,,0000,0000,0000,,form of delivery robots. I don't know if\Nit's going to be what we had or what
Dialogue: 0,0:50:50.61,0:50:58.54,Default,,0000,0000,0000,,somebody else is going to have.\NAudience Member 5: Were there any safety
Dialogue: 0,0:50:58.54,0:51:02.92,Default,,0000,0000,0000,,certifications you had to satisfy in order\Nto operate around people?
Dialogue: 0,0:51:02.92,0:51:10.05,Default,,0000,0000,0000,,S: No. So. {\i1}laughter{\i0} Well, the thing is,\Nin the US, like, it's kind of just do
Dialogue: 0,0:51:10.05,0:51:13.83,Default,,0000,0000,0000,,whatever you want. It's very different\Nfrom Germany. You can kind of just do
Dialogue: 0,0:51:13.83,0:51:17.36,Default,,0000,0000,0000,,things and you can do them until you get\Nin trouble. So we kind of had that
Dialogue: 0,0:51:17.36,0:51:20.47,Default,,0000,0000,0000,,approach don't ask for permission, don't ask for\Nforgiveness. We ended up having to have a
Dialogue: 0,0:51:20.47,0:51:24.73,Default,,0000,0000,0000,,permit in the cities we operate in. But it\Nwas very simple. It was like, OK, you have
Dialogue: 0,0:51:24.73,0:51:30.52,Default,,0000,0000,0000,,to have lights. You have to have a phone\Nnumber and you can not go in these areas.
Dialogue: 0,0:51:30.52,0:51:33.64,Default,,0000,0000,0000,,That was essentially all the\Nauthorization, all the permitting and
Dialogue: 0,0:51:33.64,0:51:36.55,Default,,0000,0000,0000,,certification that we had.\NH: Yeah?
Dialogue: 0,0:51:36.55,0:51:44.10,Default,,0000,0000,0000,,Audience Member 6: I wanted to ask, did\Nyou try other markets? Like, autonomous
Dialogue: 0,0:51:44.10,0:51:49.32,Default,,0000,0000,0000,,driving is very hard, even way more\Nthan... manage it fully. So like perhaps
Dialogue: 0,0:51:49.32,0:51:53.80,Default,,0000,0000,0000,,elderly care, like you could use this\Nrobots in elder care where you have a
Dialogue: 0,0:51:53.80,0:51:59.90,Default,,0000,0000,0000,,controlled environment where everything is\Nthe same. Did you search after other
Dialogue: 0,0:51:59.90,0:52:05.88,Default,,0000,0000,0000,,markets where it's less...\NS: Yeah, that's a great question. Yeah,
Dialogue: 0,0:52:05.88,0:52:10.00,Default,,0000,0000,0000,,there is a lot of potential for markets\Nlike elderly care, for example, also for
Dialogue: 0,0:52:10.00,0:52:14.93,Default,,0000,0000,0000,,mail delivery, for applications inside of\Nfactories. We had a couple different
Dialogue: 0,0:52:14.93,0:52:18.66,Default,,0000,0000,0000,,medical companies that reached out to us\Nand like, hey, we want to move items, move
Dialogue: 0,0:52:18.66,0:52:22.37,Default,,0000,0000,0000,,packages inside of our facilities. So we\Ndid have a lot of interest. We tried to
Dialogue: 0,0:52:22.37,0:52:25.48,Default,,0000,0000,0000,,keep a focus on the consumer space, like\Nreally building a consumer experience that
Dialogue: 0,0:52:25.48,0:52:29.40,Default,,0000,0000,0000,,worked out before branching out into these\Nmore B2B approaches. Where elderly care
Dialogue: 0,0:52:29.40,0:52:34.30,Default,,0000,0000,0000,,could be one of them. I think one\Nimportant thing about elderly care and
Dialogue: 0,0:52:34.30,0:52:38.16,Default,,0000,0000,0000,,services like Meals on Wheels, for\Nexample, is that human contact. So I think
Dialogue: 0,0:52:38.16,0:52:41.83,Default,,0000,0000,0000,,people who are maybe not seeing as much of\Ntheir family, of their relatives, they
Dialogue: 0,0:52:41.83,0:52:45.30,Default,,0000,0000,0000,,really cherish that connection they get\Nfrom people who deliver them food. So I
Dialogue: 0,0:52:45.30,0:52:48.62,Default,,0000,0000,0000,,think it's a multifaceted approach they\Nhave to have. You have a couple of
Dialogue: 0,0:52:48.62,0:52:54.64,Default,,0000,0000,0000,,different considerations with these kind\Nof services for the elderly, for example.
Dialogue: 0,0:52:54.64,0:52:59.90,Default,,0000,0000,0000,,AM 6: Thank you.\NAudience Member 7: What kind of
Dialogue: 0,0:52:59.90,0:53:06.77,Default,,0000,0000,0000,,personality do Chinese entrepreneurs have?\NS: {\i1}laughs{\i0} I think, as I mentioned like,
Dialogue: 0,0:53:06.77,0:53:10.74,Default,,0000,0000,0000,,it's really important to have\Nrelationships. So they were very
Dialogue: 0,0:53:10.74,0:53:16.05,Default,,0000,0000,0000,,interesting. They were very deeply in\Nbelief of their government. They had
Dialogue: 0,0:53:16.05,0:53:20.38,Default,,0000,0000,0000,,nothing bad to say about it. They believe\Nthey would bring them everything... the
Dialogue: 0,0:53:20.38,0:53:23.33,Default,,0000,0000,0000,,best possible, even though they still try\Nto access Facebook and Twitter with VPNs.
Dialogue: 0,0:53:23.33,0:53:28.23,Default,,0000,0000,0000,,So they were very, very loyal\Nto their governments. They were very, very
Dialogue: 0,0:53:28.23,0:53:32.82,Default,,0000,0000,0000,,diligent. If they committed to something,\Nthey would usually deliver on that. They
Dialogue: 0,0:53:32.82,0:53:38.17,Default,,0000,0000,0000,,really wanted to make sure you had a good\Nexperience. And also what we saw, for
Dialogue: 0,0:53:38.17,0:53:41.06,Default,,0000,0000,0000,,example, with building up these\Nrelationships, like the first few times we
Dialogue: 0,0:53:41.06,0:53:45.30,Default,,0000,0000,0000,,talk, they would try everything to impress\Nus. So we got taken to these ridiculously
Dialogue: 0,0:53:45.30,0:53:50.49,Default,,0000,0000,0000,,expensive restaurants to make sure that we\Nwere welcomed well and make sure
Dialogue: 0,0:53:50.49,0:53:55.20,Default,,0000,0000,0000,,everything was right. I actually had an\Ninteresting episode earlier this year. I
Dialogue: 0,0:53:55.20,0:53:59.56,Default,,0000,0000,0000,,was going to go to Burning Man and then\Nall of a sudden one of my colleagues had
Dialogue: 0,0:53:59.56,0:54:04.08,Default,,0000,0000,0000,,an argument with my manufacturer about\Nwhether Hong Kong is another country or
Dialogue: 0,0:54:04.08,0:54:09.90,Default,,0000,0000,0000,,not. And I ended up having to go to China\Nto deal with our manufacturer instead of
Dialogue: 0,0:54:09.90,0:54:14.20,Default,,0000,0000,0000,,going to Burning Man to make sure we're\Naligned in terms of our beliefs. So,
Dialogue: 0,0:54:14.20,0:54:17.83,Default,,0000,0000,0000,,sometimes it's really delicate. You\Ncannot, like talk too much about the
Dialogue: 0,0:54:17.83,0:54:20.61,Default,,0000,0000,0000,,government there. You can't talk too much\Nabout politics. It's best to just stick to
Dialogue: 0,0:54:20.61,0:54:25.15,Default,,0000,0000,0000,,business and, yeah, focus on building a\Nproduct.
Dialogue: 0,0:54:25.15,0:54:30.06,Default,,0000,0000,0000,,H: I guess this was it. Thank you.\NS: Thank you so much. {\i1}Applause{\i0}
Dialogue: 0,0:54:30.06,0:55:00.00,Default,,0000,0000,0000,,subtitles created by c3subtitles.de\Nin the year 2020. Join, and help us!