1 00:00:00,000 --> 00:00:18,010 *36C3 preroll music* 2 00:00:18,480 --> 00:00:25,050 Herald: So Sasha is a doctor with a weak spot for LEDs and, completely abstains 3 00:00:25,050 --> 00:00:30,539 from HDMI adapters these days. He wanted to share with us the experience of 4 00:00:30,539 --> 00:00:36,360 attempting to build a delivery robots in the past 2.5 years in the Bay Area. And 5 00:00:36,360 --> 00:00:38,529 so, yeah, let's give him a big welcome. 6 00:00:46,420 --> 00:00:51,660 Sasha: Thank you, Mikael. So just a show of hands who here has built robots before? 7 00:00:52,530 --> 00:00:56,050 Well, it's quite a few people. What about autonomous robots, anybody built 8 00:00:56,050 --> 00:01:03,360 autonomous robots? Still quite a few people. Well, today I'm gonna be sharing 9 00:01:03,360 --> 00:01:09,150 with you the story of how not to build autonomous robots. Over the course of the 10 00:01:09,150 --> 00:01:12,690 past two and a half years together with my team, we built the world's largest robotic 11 00:01:12,690 --> 00:01:17,740 delivery infrastructure. We went from a concept sketch to a commercially viable 12 00:01:17,740 --> 00:01:22,520 service running in three cities. We've had lots of successes and one or two failures. 13 00:01:27,200 --> 00:01:29,950 So over the course of the next 45 minutes, I'm going to be sharing with you a couple 14 00:01:29,950 --> 00:01:33,410 of different stories. First of all, I'm going to briefly introduce myself and I'm 15 00:01:33,410 --> 00:01:37,280 going to share the story of how we built robots, the different prototypes we had, 16 00:01:37,280 --> 00:01:40,820 the different iterations that we tried. Then I'm going to jump on manufacturing. 17 00:01:40,820 --> 00:01:44,340 We actually went to China and scaled up our manufacturing, our production line. 18 00:01:44,340 --> 00:01:47,880 I'm gonna share with you the story of how he did that. And finally, I'm gonna talk 19 00:01:47,880 --> 00:01:52,320 about A.I. and all the magic that is artificial intelligence. So we'll be able 20 00:01:52,320 --> 00:01:58,013 to see how we were able to crack that puzzle. So without further ado, let's do 21 00:01:58,013 --> 00:02:04,380 the introduction. This is me. Right here. I like to build things. I built my first 22 00:02:04,380 --> 00:02:08,160 website when I was eleven and I built my first business when I was 13. I was the 23 00:02:08,160 --> 00:02:11,950 iPhone repair business that I was running in my bedroom. I've been really, really 24 00:02:11,950 --> 00:02:16,170 passionate about building things and over the course of many years I built a couple 25 00:02:16,170 --> 00:02:19,980 of different startups. One of them was a food delivery platform. We ended up 26 00:02:19,980 --> 00:02:24,450 running three different cities and doing hundreds of deliveries a day. By the time 27 00:02:24,450 --> 00:02:29,590 I was 19. So I got to experience startups pretty early on. I've been really enjoying 28 00:02:29,590 --> 00:02:33,100 that time. After this food delivery startup failed I went to some 29 00:02:33,100 --> 00:02:37,731 cryptocurrency startups and then went to work for big corporations. And that's 30 00:02:37,731 --> 00:02:44,900 actually very boring. I dorned my office with some supplementary graphics. After a 31 00:02:44,900 --> 00:02:47,790 while, I got a little bit bored of this corporate life. It wasn't really for me. 32 00:02:47,790 --> 00:02:51,670 So I decided to get a one way ticket to San Francisco. So I ended up in San 33 00:02:51,670 --> 00:02:56,740 Francisco staying on a friend's couch, not really knowing anybody. And I was really 34 00:02:56,740 --> 00:03:00,590 fortunate to be introduced to an incredible group of people. And over the 35 00:03:00,590 --> 00:03:07,336 course of about two and a half years, we started to take a concepts. A scatch that 36 00:03:07,336 --> 00:03:10,970 we had and we built up a robot. At first it was something that barely even worked. 37 00:03:10,970 --> 00:03:14,380 But then we gradually got to something that worked a little bit better and better 38 00:03:14,380 --> 00:03:18,950 and better. After a while, we actually managed to build a whole fleet of robots. 39 00:03:18,950 --> 00:03:23,680 I think at the peak we had 150 robots. So it was a really, really cool experience. 40 00:03:23,680 --> 00:03:27,130 And during that time, I got to meet the lieutenant governor of California, how to 41 00:03:27,130 --> 00:03:31,210 figure out how to do manufacturing in China and most importantly, work with an 42 00:03:31,210 --> 00:03:35,940 incredible team which who I had a lot of fun with building these robots. So, yeah, 43 00:03:35,940 --> 00:03:40,180 it's a little bit about me and what we were building. And maybe now we can jump 44 00:03:40,180 --> 00:03:45,730 in to how not to build robots. So this is our very first robot,this is a really 45 00:03:45,730 --> 00:03:51,000 small prototype. We built is basically a shopping basket on wheels. There is a RC 46 00:03:51,000 --> 00:03:55,640 car there below, there's a shopping basket and Arduino Raspberry Pi. The thing barely 47 00:03:55,640 --> 00:03:59,550 work. Honestly, it was really, really hacky. And what ended up happening is that 48 00:03:59,550 --> 00:04:03,750 most of the time we just dropped off the robot in front of the customer that 49 00:04:03,750 --> 00:04:06,680 literally just dropped it in front of the door just to see if they would like order 50 00:04:06,680 --> 00:04:11,190 food with robots. The answer was overwhelmingly yes. So we decided to spend 51 00:04:11,190 --> 00:04:15,480 some more time building our technology. There is a small - I don't know if you can 52 00:04:15,480 --> 00:04:18,510 see it here. Yeah, there you go. There's a small orange holder, that's actually a 53 00:04:18,510 --> 00:04:23,779 phone holder. So our very first prototype, it had a phone sitting on top of it doing 54 00:04:23,779 --> 00:04:30,659 a video call so that somebody can remotely control it from Colombia. So we really 55 00:04:30,659 --> 00:04:34,289 started out small, really humble just to see if it would work. And that's something 56 00:04:34,289 --> 00:04:38,460 that we did a lot of this being really resourceful in terms of trying out things. 57 00:04:38,460 --> 00:04:42,529 For about a year of this, we moved on to something that looks a little bit more 58 00:04:42,529 --> 00:04:45,830 like this. So we started playing around with the shape. We start playing around 59 00:04:45,830 --> 00:04:50,280 with the design. We noticed that people responded really positively to faces and 60 00:04:50,280 --> 00:04:56,669 to like things that looked like people. So we actually built in a face. So we took 61 00:04:56,669 --> 00:05:00,960 this little animation that we built and we put it onto the robot and there's actually 62 00:05:00,960 --> 00:05:05,619 really, really positive. We had a lot of good responses from the community, a lot 63 00:05:05,619 --> 00:05:09,979 of great feedback. And what we've seen is that people really love to have robots 64 00:05:09,979 --> 00:05:13,289 that are kind of friendly. There was another company that deployed robots that 65 00:05:13,289 --> 00:05:17,669 looked like vending machines or almost like tanks in San Francisco, and they got 66 00:05:17,669 --> 00:05:21,669 banned really, really quickly. So we decided that we would do our best to make 67 00:05:21,669 --> 00:05:26,620 sure our robots were as friendly as possible instead of threatening and scary. 68 00:05:26,620 --> 00:05:32,180 So that was a very important part of it. After another year, we ended up scaling up 69 00:05:32,180 --> 00:05:37,789 our production and we went to China to manufacture robots. And here this is what 70 00:05:37,789 --> 00:05:39,779 we ended up doing. *music* 71 00:05:39,779 --> 00:05:43,089 It's actually, a cool robot. We built it entirely from scratch. We got our own 72 00:05:43,089 --> 00:05:48,650 chassis, our cabin, our own compute module. Basically just about everything. 73 00:05:49,100 --> 00:05:56,070 That was a really cool experience. That was me. So yeah, that's a robot. That's 74 00:06:04,610 --> 00:06:11,080 the one we were rolling around the past six months. And we also had some failures 75 00:06:11,080 --> 00:06:16,860 in between, as you saw previously, this one. So we actually tried a couple of 76 00:06:16,860 --> 00:06:20,930 different concepts. So this was one of them. This was a Kiwi trike. We thought 77 00:06:20,930 --> 00:06:25,189 that maybe we can figure out how to have robots do part of the delivery and then 78 00:06:25,189 --> 00:06:29,589 trikes to another part of the delivery. We also tried to do restaurant robots. We had 79 00:06:29,589 --> 00:06:33,750 like robots that sit in the restaurant and bring food out from the counter to your 80 00:06:33,750 --> 00:06:38,430 doorstep. But what ended up happening is that it was actually pretty inefficient 81 00:06:38,430 --> 00:06:41,900 and people would wait a really long time for their deliveries. So it was very 82 00:06:41,900 --> 00:06:45,449 important for us to try a lot of different things. We tried this robot, the kiwi 83 00:06:45,449 --> 00:06:50,199 TRIKE that not quite worked out as we expected. We tried a restaurant robot. We 84 00:06:50,199 --> 00:06:55,070 tried a box that would sit behind our robot. We tryed a hub that would have like 85 00:06:55,070 --> 00:06:58,990 of a bunch of different robots inside of it. So we really, really tried a lot. And 86 00:06:58,990 --> 00:07:02,770 with every iteration, we constantly tried new techniques we costantly tried new 87 00:07:02,770 --> 00:07:06,590 manufacturing methods. We really tried just about everything to see if we can 88 00:07:06,590 --> 00:07:11,430 make it work. And what we ended up building is a platform that was really 89 00:07:11,430 --> 00:07:16,610 loved by people. We built a platform that students adored. That was our primary 90 00:07:16,610 --> 00:07:21,110 demographic we're delivering to college campuses and students really loved our 91 00:07:21,110 --> 00:07:24,940 products. We actually had people dressed up as Halloween costumes. We had entire 92 00:07:24,940 --> 00:07:29,539 classes go for Halloween in like kiwi bot costumes. So that was really, really cool 93 00:07:29,539 --> 00:07:33,949 stuff. Had a lot of great support. A lot of trust from the community as well. And 94 00:07:33,949 --> 00:07:37,279 that's like coming back to the design. That aspect of having a friendly robot, 95 00:07:37,279 --> 00:07:41,080 that meshes seamlessly within the fabric of a community is like super, super 96 00:07:41,080 --> 00:07:44,889 important. We've seen other robots around and they were maybe not as friendly, maybe 97 00:07:44,889 --> 00:07:49,300 they looked a little scary. Maybe they had something that was a bit off or maybe a 98 00:07:49,300 --> 00:07:52,569 little too industrial. But having like a friendly robot that could become a meme 99 00:07:52,569 --> 00:07:56,460 that was something truly revolutionary, something that really changed the 100 00:07:56,460 --> 00:08:00,389 landscape. And as a matter of fact, like these cute bots are the only robots that 101 00:08:00,389 --> 00:08:06,979 are deployed somewhere in the world where they coexist. Day to day with a community, 102 00:08:06,979 --> 00:08:10,399 with people. Like you have some limited deployments of robots here and there, 103 00:08:10,399 --> 00:08:13,599 maybe have a Roomba at home or some like that. But you don't have any large scale 104 00:08:13,599 --> 00:08:19,520 deployment. We have robots and people living in the same city all the time. So, 105 00:08:19,520 --> 00:08:22,649 of course, it took us a while to figure out what to do and how to do it. At first 106 00:08:22,649 --> 00:08:27,080 one of our models was to have robots deliver the entire meal, like go from the 107 00:08:27,080 --> 00:08:31,610 restaurant all the way to the customer and we would have a robot do that delivery. 108 00:08:31,610 --> 00:08:34,390 Turns out, it was pretty inefficient. People would wait like 60 minutes, 90 109 00:08:34,390 --> 00:08:39,039 minutes for their delivery. And we realized that maybe automating all of that 110 00:08:39,039 --> 00:08:43,840 was not the most efficient approach. So what we instead did is a multi-modal 111 00:08:43,840 --> 00:08:47,250 approach where we had people and robots. This is actually a really cool 112 00:08:47,250 --> 00:08:51,910 visualization that my team came up with. The blue lines are robots. So these are 113 00:08:51,910 --> 00:08:55,830 robots roaming around our Berkeley coverage area and the yellow lines are 114 00:08:55,830 --> 00:08:59,650 people. So how this would work is that people would go to restaurants, they pick 115 00:08:59,650 --> 00:09:03,500 up the food and they take you to a cluster. They take it to a cluster where 116 00:09:03,500 --> 00:09:06,940 you had a bunch of robots, they loaded into the robot and then the robot would 117 00:09:06,940 --> 00:09:11,190 actually do the last few hundred meters to your doorstep. And because we were able to 118 00:09:11,190 --> 00:09:16,930 do this, we were able to go and build a platform, that handled hundreds of orders a 119 00:09:16,930 --> 00:09:22,190 day with very, very few people. I mean, labor costs are really high for delivery. 120 00:09:22,190 --> 00:09:25,370 You'd be paying somewhere between five and thirteen dollars to get a meal delivered 121 00:09:25,370 --> 00:09:28,571 in the U.S. And as a student, that's like super expensive. That's not something that 122 00:09:28,571 --> 00:09:32,140 you can afford do every day. And also there is a pretty big shortage of people 123 00:09:32,140 --> 00:09:35,431 who want to do this job in the first place. The trend is really high. People 124 00:09:35,431 --> 00:09:38,730 are leaving all the time because they don't like to like sit in a car all day 125 00:09:38,730 --> 00:09:42,160 and deliver food. So that's why we have this parallel like this multi-modal 126 00:09:42,160 --> 00:09:47,160 approach where the people are biking around, they're enjoying their time outside and 127 00:09:47,160 --> 00:09:50,730 the robots are actually doing other boring stuff like the waiting. So the robot would 128 00:09:50,730 --> 00:09:54,580 go up to your doorstep and would wait for you to put on your pants, your shoes and 129 00:09:54,580 --> 00:09:58,000 actually walk outside. So that way we were able to change the dynamic. We're able to 130 00:09:58,000 --> 00:10:02,890 change our deficiency from one or two deliveries an hour, as you would have with 131 00:10:02,890 --> 00:10:07,530 like a traditional delivery service to as much as 15 deliveries an hour per person. 132 00:10:07,530 --> 00:10:11,040 So it made the delivery far more affordable and we were able to offer 133 00:10:11,040 --> 00:10:16,800 delivery at just one dollar a delivery, which is a cost that changes completely 134 00:10:16,800 --> 00:10:21,800 the way people approach delivery. In fact, if we look our top 20 percent of users, 135 00:10:21,800 --> 00:10:26,900 they were ordering over 14 times a week. So they were very, very happy that they 136 00:10:26,900 --> 00:10:30,510 could get whatever they wanted very quickly. Of course, not everybody was 137 00:10:30,510 --> 00:10:35,350 super happy. So we did have some people that didn't fully appreciate the magic 138 00:10:35,350 --> 00:10:40,690 that is kiwi bot. So we did have one person try to steal it, but they didn't 139 00:10:40,690 --> 00:10:48,100 get away with it. We found them pretty quickly. They hid it in the trunk. Not a 140 00:10:48,100 --> 00:10:56,310 very smart move. We ended up finding it with G.P.S. and also triangulating the Wi-Fi. 141 00:10:56,310 --> 00:11:00,750 So this guy decided to steal it because he doesn't like robots. I don't know why, but 142 00:11:00,750 --> 00:11:06,620 he was clearly very passionate about that topic. And he stole it and now he's in 143 00:11:06,620 --> 00:11:14,140 jail. So, yeah, don't steal robots. So maybe some conclusions from our robot 144 00:11:14,140 --> 00:11:18,351 part, like from building robots, from figuring out like what to do and what not 145 00:11:18,351 --> 00:11:21,920 to do. Really important thing that we do a lot in software and maybe not as much 146 00:11:21,920 --> 00:11:27,110 hardware is iteration. Like we iterated through three major revisions and like 147 00:11:27,110 --> 00:11:32,200 lots of small revisions during a really small period of time. It was really 148 00:11:32,200 --> 00:11:35,920 interesting to see like that transition. Every single time we try something new, we 149 00:11:35,920 --> 00:11:39,760 try it maybe for like 20 robots at a time, like not our whole fleet. We just try for 150 00:11:39,760 --> 00:11:43,690 a small portion of our fleet and that we were able to iterate really quickly and 151 00:11:43,690 --> 00:11:48,770 see what sensors worked or cameras worked. And just to see what we could do in order 152 00:11:48,770 --> 00:11:55,130 to grow the products, it was very important to iterate. Communication. 153 00:11:55,130 --> 00:11:59,780 Communication is absolutely fundamental. And not only communication like inside the 154 00:11:59,780 --> 00:12:03,071 company or anything, but more importantly, communication with your community. Because 155 00:12:03,071 --> 00:12:06,600 we weren't just building a product in isolation. We were building a product for 156 00:12:06,600 --> 00:12:11,530 people who live in a city, who have an established life. And we're kind of 157 00:12:11,530 --> 00:12:15,410 intruding into their lives by bringing in a new product that takes the sidewalks. So 158 00:12:15,410 --> 00:12:19,820 communicating what we're doing, showing them what this is and what this robot does 159 00:12:19,820 --> 00:12:24,340 is super important. Actually, very early on our designs have no text on it. They 160 00:12:24,340 --> 00:12:28,731 had like no information it was just like a basket case on RC car. And people were 161 00:12:28,731 --> 00:12:32,471 like really confused. The police were like: "Hey, what is this?", so we had to 162 00:12:32,471 --> 00:12:35,710 add a lot of communication, we had to put food delivery on the robots really 163 00:12:35,710 --> 00:12:38,980 clearly, we had to add a license plate with like a phone number that somebody 164 00:12:38,980 --> 00:12:43,390 could reach out to us. So communication is very, very, very important when it comes 165 00:12:43,390 --> 00:12:50,190 to robots. Also: Scaling hardware is hard, super hard. I mean, it was crazy. When we 166 00:12:50,190 --> 00:12:55,230 first started it was just Arduinos and Raspberry Pis and that did not scale 167 00:12:55,230 --> 00:13:00,170 really well. Like, sure, we could have maybe 10 or 20 units at once. But then how 168 00:13:00,170 --> 00:13:04,660 do you handle updates? How do you handle those weird things that happen all the 169 00:13:04,660 --> 00:13:08,660 time? So it was really challenging to do this. We actually killed a bunch of SD 170 00:13:08,660 --> 00:13:12,790 cards. Didn't really know you could destroy SD cards, but you can. And we 171 00:13:12,790 --> 00:13:17,260 learned a lot of things about hardware pushing it beyond its normal boundaries. 172 00:13:17,260 --> 00:13:20,680 So yeah, iteration, super important. Communication is key, like getting buy in 173 00:13:20,680 --> 00:13:25,590 from the community and scaling hardware is super, super hard. That's something we 174 00:13:25,590 --> 00:13:33,050 actually figured out how to solve by going into China. So how to do or how not to do 175 00:13:33,050 --> 00:13:42,020 manufacturing? So as every China story goes, I hopped on a plane and I ended up 176 00:13:42,020 --> 00:13:46,180 in China. And it's really interesting to see because like you have this perception 177 00:13:46,180 --> 00:13:50,950 of China from the media, you have this idea of what it would look like. But the 178 00:13:50,950 --> 00:13:55,900 reality is it doesn't look anything like what you would expect. It was a completely 179 00:13:55,900 --> 00:14:00,400 different world. It was at the same time Bladerunner and like the most modern city 180 00:14:00,400 --> 00:14:05,780 in the world and it was truly an awesome experience. I highly recommend anybody who 181 00:14:05,780 --> 00:14:10,750 has the opportunity to go in and explore the world. But of course, the culture is a 182 00:14:10,750 --> 00:14:14,760 little bit different. We were surprised to see some things happening there. Was a 183 00:14:14,760 --> 00:14:18,880 weird dichotomy between communism and consumerism. This is kind of interesting 184 00:14:18,880 --> 00:14:27,730 to see that sometimes. But the reason why we came to China is for manufacturing and 185 00:14:27,730 --> 00:14:31,911 there is no better place for that than Shenzhen. In Shenzhen, you have 186 00:14:31,911 --> 00:14:37,560 Huaqiangbei. This huge market. It's a market that spans several city blocks and 187 00:14:37,560 --> 00:14:41,910 you can actually find anything and everything you want. We were able to get 188 00:14:41,910 --> 00:14:46,650 components super quickly, super easily. And you could spend days just walking to a 189 00:14:46,650 --> 00:14:52,150 single building finding different things. There were entire city blocks dedicated to 190 00:14:52,150 --> 00:14:57,970 like just LEDs or just connectors or just processors. It was absolutely crazy. You 191 00:14:57,970 --> 00:15:02,290 could really, really, really get lost inside of these mazes. And what was really 192 00:15:02,290 --> 00:15:06,410 incredible to see and something I've never seen anywhere else in the world is just 193 00:15:06,410 --> 00:15:12,960 how easy it is to get hardware, to get things, to get parts. It was super easy. 194 00:15:12,960 --> 00:15:17,300 Just go in and get something and you could get it at one piece, two pieces, a 195 00:15:17,300 --> 00:15:21,090 thousand pieces like instantly. If you're anywhere else in the world, that's super 196 00:15:21,090 --> 00:15:24,980 hard to do. So just by this virtue, you're actually able to prototype things. You're 197 00:15:24,980 --> 00:15:28,810 able to build things incredibly fast. You're able to go in, you're able to 198 00:15:28,810 --> 00:15:33,960 comission a PCB and get all the parts almost instantly, which is not something 199 00:15:33,960 --> 00:15:37,850 you see anywhere else in the world. And also a lot of the manufacturers have their 200 00:15:37,850 --> 00:15:41,630 booths here so these would be direct booths from the manufacturers so you could 201 00:15:41,630 --> 00:15:45,290 say go up to them, start talking to them and ask, hey, can you make this product 202 00:15:45,290 --> 00:15:49,960 this specific way? Can you do it how I want it? And they'll be like, sure, why 203 00:15:49,960 --> 00:15:53,750 not? They'll do it for you. So it was really, really valuable to just learn from 204 00:15:53,750 --> 00:15:58,529 these people, from the vendors here, from manufacturers about how to build things. 205 00:15:58,529 --> 00:16:03,380 And it was actually really surprising to see everything they had in stock. Two 206 00:16:03,380 --> 00:16:07,960 years ago, we built an art installation here that covered a tunnel with LEDs. We 207 00:16:07,960 --> 00:16:13,490 covered one of the tunnels at 34C3 with LEDs. And we used this tiny, tiny chip. It 208 00:16:13,490 --> 00:16:18,270 was a five dollar ESP 8266 chip that basically was able to control all your 209 00:16:18,270 --> 00:16:23,370 LEDs. And over the course of five years, up to that point, I spent a lot of time 210 00:16:23,370 --> 00:16:28,300 figuring out how to build it myself. I played with Raspberry Pis I played with 211 00:16:28,300 --> 00:16:33,210 PCA controllers over serial and like I finally managed to get a prototype to 212 00:16:33,210 --> 00:16:36,950 work, but it was super clunky, it was super expensive and it wasn't very 213 00:16:36,950 --> 00:16:41,260 reliable and I go to China and I find that it's available there and much better 214 00:16:41,260 --> 00:16:46,340 quality, much cheaper, much faster, so it was a really, really interesting shift in 215 00:16:46,340 --> 00:16:49,390 perspective. It's something you can't appreciate when you're abroad. Even if 216 00:16:49,390 --> 00:16:52,900 you're browsing like eBay or Ali Express it's kind of hard to appreciate just how 217 00:16:52,900 --> 00:16:57,910 much selection you have and how you can find just about any tool, anything you 218 00:16:57,910 --> 00:17:02,500 need to find. So it's really, really incredible. But these markets were cool 219 00:17:02,500 --> 00:17:08,010 but was even cooler are the factories and during a course in China, we were able to 220 00:17:08,010 --> 00:17:11,579 visit a lot of factories. All these factories there are super, super 221 00:17:11,579 --> 00:17:15,159 welcoming. They always love having you over. They invited you to really, really 222 00:17:15,159 --> 00:17:20,819 luxurioius dinners. We had way too much food. And it was a feast of celebration 223 00:17:20,819 --> 00:17:24,149 every time. Actually, relationships are super, super important in China. Like a 224 00:17:24,149 --> 00:17:28,189 lot of people in the West, like they have contracts and they say, OK, this is the 225 00:17:28,189 --> 00:17:32,650 terms of the contracts. Well, China, you do sort of have contracts, but they don't 226 00:17:32,650 --> 00:17:36,910 matter as much as relationships. Like when you have a relationship with the 227 00:17:36,910 --> 00:17:39,680 manufacturer, you have to like always go to dinner with them, drink beer, smoke, go 228 00:17:39,680 --> 00:17:43,850 to KTV like it's a really evolved relationship. And you're only able to have 229 00:17:43,850 --> 00:17:46,700 good communication based on that relationship, because if you don't have a 230 00:17:46,700 --> 00:17:49,960 relationship they kind of forget about you. We actually had a couple of instances 231 00:17:49,960 --> 00:17:54,520 where manufacturers ghosted us. Like they had a critical component and they just 232 00:17:54,520 --> 00:17:57,610 stopped answering our emails, they stop answering our weechats, they just 233 00:17:57,610 --> 00:18:01,210 completely ignored us. And for some pieces they were completely irreplaceable, we 234 00:18:01,210 --> 00:18:05,179 could not just go out and find another factory to produce a specific part the way 235 00:18:05,179 --> 00:18:09,009 we wanted. And the only way you can ensure that this doesn't happen is by really 236 00:18:09,009 --> 00:18:13,610 explicitly making sure that you have good communication, a good relationship with 237 00:18:13,610 --> 00:18:17,450 that manufacturers. It's super, super important. This is one of the factories we 238 00:18:17,450 --> 00:18:23,379 worked with. It's really crazy. I mean, we went there and we're just absolutely blown 239 00:18:23,379 --> 00:18:27,830 away by the scale of everything and also blown away by how manual everything is. 240 00:18:27,830 --> 00:18:32,159 There's actually audio here. Everything was super manual. People were just like 241 00:18:32,159 --> 00:18:36,309 there with minimal or no protective equipment whatsoever. Just like building 242 00:18:36,309 --> 00:18:39,540 things that look like they were made by robots or machines, but they were in 243 00:18:39,540 --> 00:18:45,889 reality built by people with their hands, which is super crazy to see. And there 244 00:18:45,889 --> 00:18:51,570 were a lot of Blade Runner esque designs, really bizarre contraptions there in this 245 00:18:51,570 --> 00:18:57,450 factory. This is our fiberglass factory. The way we build our casing, was actually 246 00:18:57,450 --> 00:19:03,940 prototyping it first in carbon fiber, sorry, fiberglass and then moving onto a 247 00:19:03,940 --> 00:19:08,610 mold in carbon fiber. And actually Scotty, he made a really cool video on YouTube. So 248 00:19:08,610 --> 00:19:13,429 if you search for hockey stick factory on YouTube, you can see a huge video. My 249 00:19:13,429 --> 00:19:17,540 buddy Scotty actually goes and meets his factory, discover how they make this mold 250 00:19:17,540 --> 00:19:21,179 and how they make these carbon fiber things. It was actually really crazy to 251 00:19:21,179 --> 00:19:26,029 see it. It was cheaper to make a carbon fiber mold than it was to make a plastic 252 00:19:26,029 --> 00:19:29,460 mold. So since the tolerances were a little bit different, since like the 253 00:19:29,460 --> 00:19:32,739 process was a little bit simpler, you were able to make a mold, that was very, very 254 00:19:32,739 --> 00:19:38,440 strong and very indestructible without necessarily having to have all of that 255 00:19:38,440 --> 00:19:43,909 expense upfront for like a plastic mold. So, yeah, that our a fiberglass factory. 256 00:19:43,909 --> 00:19:47,370 Really exciting stuff. Really crazy scale. These folks like the first night we came 257 00:19:47,370 --> 00:19:51,059 there, we arrived at like 8 pm and there was 100 people in the factory just like 258 00:19:51,059 --> 00:19:56,720 working at 8pm. Really crazy to see. This is another factory worked with. So this 259 00:19:56,720 --> 00:20:01,539 was a metal factory. It was actually really, really, really interesting to see 260 00:20:01,539 --> 00:20:06,019 how they built all these things and at one hand, you can build super complex things, 261 00:20:06,019 --> 00:20:10,559 you can build a super complex designs. But on the other hand, we got surprised a 262 00:20:10,559 --> 00:20:15,750 couple of times by being unable to manufacture really simple designs. And it 263 00:20:15,750 --> 00:20:19,690 took us a while to get a grasp was like, oh, OK, so we can make really complex 264 00:20:19,690 --> 00:20:24,089 metal that's bent, but as soon as we add a well to aluminum, you start to have a big, 265 00:20:24,089 --> 00:20:28,059 big problem. So we had to like change a lot of our designs. We had to really adapt 266 00:20:28,059 --> 00:20:32,200 to the way things were being made in China. And sometimes you could adapt 267 00:20:32,200 --> 00:20:36,140 yourself, but like at an insane cost. So it was better to adapt to the way things 268 00:20:36,140 --> 00:20:41,190 were being done there. So, again, very, very interesting to see how things are 269 00:20:41,190 --> 00:20:45,480 done. No protective equipment, this is like a two ton press and his hands are 270 00:20:45,480 --> 00:20:50,139 millimeters away from it. So, yeah, it's a different world out there. Very, very 271 00:20:50,139 --> 00:20:57,830 different. Another factory we visited was a PCB factory. So this one has a really 272 00:20:57,830 --> 00:21:02,240 interesting story. This factory is not in Shenzhen. It's just across the border from 273 00:21:02,240 --> 00:21:07,299 Shenzhen. The city actually passed a law a couple of years ago that has very, very 274 00:21:07,299 --> 00:21:11,899 strict environmental policies. So you're no longer able to do PCB manufacturing 275 00:21:11,899 --> 00:21:15,900 inside the city anymore. So we actually had to drive for a couple of hours outside 276 00:21:15,900 --> 00:21:21,200 of the city and over there was a huge plant. And this plant was kind of semi 277 00:21:21,200 --> 00:21:26,649 automated, semi handmade. Were part of the process were done by hand, as you see 278 00:21:26,649 --> 00:21:31,280 here. But then parts of the process were done with machines. So they had this giant 279 00:21:31,280 --> 00:21:33,834 machine, which is basically a black box we can't really see inside of it. But you had 280 00:21:33,834 --> 00:21:38,980 a bunch of chemicals and it's like take a PCB and just like move it forward through 281 00:21:38,980 --> 00:21:43,220 a chain. This is really intresting to see. And this factory also had a really quick 282 00:21:43,220 --> 00:21:46,070 turnaround, they had a three hour turnaround if you paid a premium and the 283 00:21:46,070 --> 00:21:50,580 standard was 24 hours. You could also ask them to do PCBA so you can actually get 284 00:21:50,580 --> 00:21:55,070 them to assemble the PCB for you. And we ended up doing that for some of our PCBs. 285 00:21:55,070 --> 00:22:00,230 We'd give them build materials and we'd give them our designs and then they 286 00:22:00,230 --> 00:22:04,080 manifacture it. We actually got in a little bit of a situation with that 287 00:22:04,080 --> 00:22:08,119 because we sent them some designs, we sent them some parts that we wanted to put in 288 00:22:08,119 --> 00:22:12,350 our PCB and it turns out that one of these parts was unavailable and they didn't tell 289 00:22:12,350 --> 00:22:17,169 that to us until it was almost Chinese New Year. So we had to scramble all that to 290 00:22:17,169 --> 00:22:21,429 find another solution. Was very exciting to see how you would deal with these 291 00:22:21,429 --> 00:22:25,060 factories. There are some even cooler factories. I think the coolest factory I 292 00:22:25,060 --> 00:22:29,259 visited was a battery factory where they made lithium ion and lithium polymer 293 00:22:29,259 --> 00:22:34,120 batteries and it was almost entirely automated. It had giant films of things 294 00:22:34,120 --> 00:22:39,129 going into a machine and then you had all sorts of liquids and powders it was all 295 00:22:39,129 --> 00:22:43,000 combined together. It was super, super cool. Didn't allow us to film it 296 00:22:43,000 --> 00:22:45,760 unfortunately, there may be only a dozen or so such factories in the world. They're 297 00:22:45,760 --> 00:22:49,379 very protective about their technology, but the scale of how quickly they're 298 00:22:49,379 --> 00:22:53,070 manufacturing these batteries was just incredible. They would manufacture them at 299 00:22:53,070 --> 00:22:58,690 a crazy, crazy scale. So all these factors are cool, but actually building things is 300 00:22:58,690 --> 00:23:03,080 even cooler. So we ended up partnering with a contract manufacturer. I was really 301 00:23:03,080 --> 00:23:07,460 fortunate to find one through my network. Otherwise, I would have been totally lost. 302 00:23:07,460 --> 00:23:11,460 A couple of days before I ended up going to China I found a contract manufacturer 303 00:23:11,460 --> 00:23:16,360 that liked to work with start ups and small scale people and we ended up working 304 00:23:16,360 --> 00:23:20,340 with them to build our first batch of 50 robots. It was really interesting to see 305 00:23:20,340 --> 00:23:25,690 how different our designs were to what they expected. So they expected things are 306 00:23:25,690 --> 00:23:30,730 really ready. They are very explicit, very clearly specified. But we didn't have 307 00:23:30,730 --> 00:23:34,620 that. The difference between manufacturing in the U.S., for example, against China is 308 00:23:34,620 --> 00:23:39,019 that in the U.S., like it's a super long process and the back and forth takes super 309 00:23:39,019 --> 00:23:42,299 long just to get an idea of what kind of files they need. Whereas in China, you're 310 00:23:42,299 --> 00:23:46,429 able to sit down directly with the engineer, with the person in charge, and 311 00:23:46,429 --> 00:23:49,770 you can figure out what they need and they can help you out instantly. Actually, just 312 00:23:49,770 --> 00:23:54,990 here, I just want to show you one thing. So this is my designer, Alehandrew, and he was 313 00:23:54,990 --> 00:23:59,710 translating from English to Chinese with his phone with Google Translate, and it 314 00:23:59,710 --> 00:24:03,470 worked surprisingly well. Google Translate actually is not blocked in China for some 315 00:24:03,470 --> 00:24:09,019 reason. We were able to communicate almost all the time with that. Also, Weechat has 316 00:24:09,019 --> 00:24:12,450 a built in translate feature. So Weechat is like the universal app that everyone in 317 00:24:12,450 --> 00:24:16,159 China uses and has this built in translation feature that can translate 318 00:24:16,159 --> 00:24:21,730 your text automatically. So it's really, really cool to see how that worked. One 319 00:24:21,730 --> 00:24:25,871 question that we get commonly asked is like how do we find our manufacturers? How 320 00:24:25,871 --> 00:24:30,740 do we get this relationship? So about 20 percent of that was through Alibaba. So 321 00:24:30,740 --> 00:24:35,830 our fiberglass manufacturer. We quoted like 30 different manufacturers and went 322 00:24:35,830 --> 00:24:40,889 with the cheapest one. Of course, it was far more expensive than we expected and we 323 00:24:40,889 --> 00:24:45,669 ended up producing with them. 20 percent that, for example, our chassey, it was 324 00:24:45,669 --> 00:24:49,629 built with companies that we already had a relationship with. So we were just able to 325 00:24:49,629 --> 00:24:54,350 continue working with them. And then 60 percent was through just references so 326 00:24:54,350 --> 00:24:58,149 select networks or getting to know people and talking to them saying, "Oh, hey, who 327 00:24:58,149 --> 00:25:01,950 did you use for this or this" or "How did you make these PCBs?" or just getting a 328 00:25:01,950 --> 00:25:06,690 conversation going. So having that kind of network was really, really helpful in 329 00:25:06,690 --> 00:25:12,929 order to build these robots. So as you can actually see right here, our design, this 330 00:25:12,929 --> 00:25:16,640 is what we had when we came into China. When we left, we had our own computer 331 00:25:16,640 --> 00:25:22,090 module like super sophisticated. But this was like a Raspberry Pi, a pix hock and a 332 00:25:22,090 --> 00:25:26,539 voltage converter like a DC to DC converter. That was pretty much it. As you 333 00:25:26,539 --> 00:25:30,870 can see, it was not very reliable. It would break a lot. So it took us quite a 334 00:25:30,870 --> 00:25:36,201 while to translate this into something that was manufacturing well. So thanks to 335 00:25:36,201 --> 00:25:42,460 the dedication of my incredible team that we're able to do that. And we kind of did 336 00:25:42,460 --> 00:25:46,380 not know what we were doing so, we ended up having all of our parts and all of the 337 00:25:46,380 --> 00:25:51,960 components ready just days before Chinese New Year. So we actually had to do all of 338 00:25:51,960 --> 00:25:55,889 that someday ourselves. We didn't have any Chinese workers who could help us do that. 339 00:25:55,889 --> 00:26:01,249 So there's our team just assembling things in the factory like we wanted two days 340 00:26:01,249 --> 00:26:06,519 before Chinese New Year. So that was very, very interesting. We kind of hacked or 341 00:26:06,519 --> 00:26:12,309 tried to hack Chinese New Year. We assembled all the robots literally days if 342 00:26:12,309 --> 00:26:16,850 not hours before Chinese New Year and we shipped them out and everything was great, 343 00:26:16,850 --> 00:26:22,359 except our robots got stuck in customs. We had a trademark on our box and the customs 344 00:26:22,359 --> 00:26:25,519 agents, they open the box and saw more trademarks on some parts. We had 3D 345 00:26:25,519 --> 00:26:28,779 printed parts and they were like, no, this is not going to go through without the 346 00:26:28,779 --> 00:26:34,690 proper paperwork. So our robots got stuck for three weeks in China, which was really 347 00:26:34,690 --> 00:26:39,330 fun. Little problematic. So, yeah, those kind of things happen you have to be ready 348 00:26:39,330 --> 00:26:45,749 for it. After we received our robots in California, we had to spend another like 349 00:26:45,749 --> 00:26:51,789 maybe one or two months refinishing them, redoing some parts, tweaking them, 350 00:26:51,789 --> 00:26:56,029 flashing them. So there's still a lot of work to get them to work. The pieces we 351 00:26:56,029 --> 00:26:59,950 shipped out to China was maybe just like a case with most of the electronics in, but 352 00:26:59,950 --> 00:27:06,100 not all of it. So we still had to do a lot of tweaking over back home. And of course, 353 00:27:06,100 --> 00:27:09,000 all this going to be impossible without an incredible team so I was really fortunate 354 00:27:09,000 --> 00:27:14,539 to be with some really, really passionate people who would work four months in a row 355 00:27:14,539 --> 00:27:18,899 continuously without virtually taking any breaks. We had plenty of opportunities to 356 00:27:18,899 --> 00:27:23,029 go and take the high speed rail or go to Shanghai or even Tokyo, but we all stayed 357 00:27:23,029 --> 00:27:27,220 in Shenzhen and spend a lot of time together building these robots. It was a 358 00:27:27,220 --> 00:27:33,809 really, really arduous journey. So maybe some conclusions for scaling, 359 00:27:33,809 --> 00:27:37,149 manufacturing, some of the failures we've had in relationships. I mean, 360 00:27:37,149 --> 00:27:41,149 relationships are super important, like super, super important, in China far more 361 00:27:41,149 --> 00:27:44,619 important then contracts. If you're able to have a good line of communication with 362 00:27:44,619 --> 00:27:47,799 your manufacturer, that really, really helps out. Because if you don't, things go 363 00:27:47,799 --> 00:27:50,789 bad. We've had manufacturers that ghosted us. We have had manufacturers that 364 00:27:50,789 --> 00:27:56,139 completely ignored us or manufacturers that just replaced components because they 365 00:27:56,139 --> 00:28:02,299 just felt like it. So relationships, super important. Don't hack Chinese New Year. We 366 00:28:02,299 --> 00:28:06,210 tried it, doesn't work. It's a thing. China just sit's down for like two or 367 00:28:06,210 --> 00:28:09,590 three weeks. So it's really, really important to respect that. People buy 368 00:28:09,590 --> 00:28:13,980 tickets to go to their hometowns like months in advance and they're not going to 369 00:28:13,980 --> 00:28:17,179 move it for just like some pesky thing that you're building, especially for like 370 00:28:17,179 --> 00:28:20,850 some small scale thing. So, yeah, don't try to hack Chinese New Year, it did not 371 00:28:20,850 --> 00:28:27,109 work out well for us. Also, do it with a team . While I was in China, I saw a 372 00:28:27,109 --> 00:28:30,450 couple of sole entrepreneurs try to build their own thing and it was super, super 373 00:28:30,450 --> 00:28:34,169 hard, super stressful, having a team is really great, especially if like a foreign 374 00:28:34,169 --> 00:28:37,759 place where you don't really know anybody. Having that team there together, to 375 00:28:37,759 --> 00:28:41,390 support you is super, super important, especially since you can multitask, you 376 00:28:41,390 --> 00:28:44,840 can split responsibilities and do something together. So it's really, really 377 00:28:44,840 --> 00:28:50,929 important aspect. So that's how we manufactured and some of the failures 378 00:28:50,929 --> 00:28:58,909 we've had. Now let's talk about how not to build A.I.. So as we all know, A.I. is 379 00:28:58,909 --> 00:29:03,190 magic, right? Just as Blockchain and IoT and the cloud. It's absolutely magic, 380 00:29:03,190 --> 00:29:09,109 right? Well, the reality is it's it's not that magic. So we decided to have a very 381 00:29:09,109 --> 00:29:15,039 pragmatic approach to A.I.. We said, let's not do anything crazy. Let's just make 382 00:29:15,039 --> 00:29:20,019 something that works. So our very first iteration of a robot was this. This is 383 00:29:20,019 --> 00:29:24,029 like the control panel for a robot. It was super simple. We had a video call coming 384 00:29:24,029 --> 00:29:27,980 in from the robot, on the left over there is literally an iframe, super simple 385 00:29:27,980 --> 00:29:31,461 stuff. And on the right, we had a map, on the bottom we had some controls so you can 386 00:29:31,461 --> 00:29:36,230 move the robot forwards, backwards. It was very, very simple. It barely worked. On 387 00:29:36,230 --> 00:29:40,139 the robot we had our Arduino, Raspberry Pi all running in python and the server was 388 00:29:40,139 --> 00:29:45,340 Java communicating over web sockets. But this barely worked. So we decided, OK, 389 00:29:45,340 --> 00:29:50,090 what can we do? Maybe we can build an autonomous robot and we can both say that 390 00:29:50,090 --> 00:29:55,340 would work entirely by itself. We actually did that. So we built a robot that could 391 00:29:55,340 --> 00:30:00,009 go entirely by itself. It was fully autonomous. And it was actually really 392 00:30:00,009 --> 00:30:04,299 cool. The way we built it is we had pretty beefy computer inside. We had it a Nvidia 393 00:30:04,299 --> 00:30:08,940 Jetson TX2. On that, we were running ROSS and inside of ROSS we were running 394 00:30:08,940 --> 00:30:12,859 TensorFlow and a couple of other technologies. We had YODA for object 395 00:30:12,859 --> 00:30:17,190 detection and some other cool tech that I am not entirely familiar with it since I 396 00:30:17,190 --> 00:30:22,460 didn't write that code, but over here what the robot did is it looked at objects. So 397 00:30:22,460 --> 00:30:27,700 it was detecting objects. It was also measuring the distance to the objects. And 398 00:30:27,700 --> 00:30:30,979 it also had an inference neural network. And you can see that on the top left of 399 00:30:30,979 --> 00:30:36,519 the screen here. Basically, based on trained data, it would know where not to 400 00:30:36,519 --> 00:30:40,799 drive into and it would try to plot a path based on 12 different directions it could 401 00:30:40,799 --> 00:30:44,980 go into. So it had 12 directions and it would go in the direction which had the 402 00:30:44,980 --> 00:30:50,119 highest probability of not colliding with somebody or something. And this worked, 403 00:30:50,119 --> 00:30:55,659 OK. We were able to get like 99 percent autonomy. But the problem is, since we're 404 00:30:55,659 --> 00:31:00,119 doing a commercially viable delivery service, that's like offering deliveries 405 00:31:00,119 --> 00:31:04,190 to regular people and not something in the lab, it really had to do something that 406 00:31:04,190 --> 00:31:07,471 worked all the time. And the challenge with this is we still needed to have 407 00:31:07,471 --> 00:31:10,970 people in the loop. We still had to have people who looked at the robot to make 408 00:31:10,970 --> 00:31:15,649 sure it would actually not crash. And what happens if you have something that's fully 409 00:31:15,649 --> 00:31:20,399 autonomous and people assume it works well, when it doesn't work well instead 410 00:31:20,399 --> 00:31:23,349 of looking at the screen and being ready to take over, they're just looking at the 411 00:31:23,349 --> 00:31:28,950 phone and Instagram. So this approach wasn't the best one. And instead, we 412 00:31:28,950 --> 00:31:33,889 decided to use a supervision approach. So we spent a lot of time building this. So 413 00:31:33,889 --> 00:31:37,879 this is our supervisors console and it's actually really, really cool platform. 414 00:31:37,879 --> 00:31:42,479 It's a platform that allows you to connect to a robot and the robot streams to you 415 00:31:42,479 --> 00:31:46,909 video over Web RTC or like the 4G network and you're able to control it over web 416 00:31:46,909 --> 00:31:51,570 sockets. So the way to work is you'd have a supervisor that sets waypoints for the 417 00:31:51,570 --> 00:31:56,139 robot to follow. So the supervisor would click on the image and he or she would 418 00:31:56,139 --> 00:32:00,150 tell the robot to move 10 meters at a time. So typically they'd set waypoints 419 00:32:00,150 --> 00:32:04,240 every 5 to 10 seconds. It was a very interesting approach. We tried a couple of 420 00:32:04,240 --> 00:32:08,489 different approaches. We tried to do slam, that really did not work out for us. It 421 00:32:08,489 --> 00:32:15,220 took too much resources and it didn't give us a significant gain. We tried other 422 00:32:15,220 --> 00:32:19,399 things as well. We tried traffic light detection. So we tried traffic light 423 00:32:19,399 --> 00:32:23,149 detection. There are some amazing models available online, some great Github repos. 424 00:32:23,149 --> 00:32:28,970 The problem is, yes, they do work on a very clean data set. But when you actually 425 00:32:28,970 --> 00:32:34,590 have data, we actually have a real life scenario where we have like glare, you 426 00:32:34,590 --> 00:32:38,979 have rain, you have weird situations, you have homeless people. It doesn't really 427 00:32:38,979 --> 00:32:43,890 translate that well in the real world. So we kind of struggled with that. Instead, 428 00:32:43,890 --> 00:32:49,019 we actually had a more middle ground approach. So we are able to detect traffic 429 00:32:49,019 --> 00:32:53,889 lights really well, but we're not able to detect the color really well or the which 430 00:32:53,889 --> 00:32:57,769 kind of signal it's giving. So instead, all we do over here, this automatically 431 00:32:57,769 --> 00:33:01,979 zooms in to traffic lights. So it's very easy to see. This video actually that 432 00:33:01,979 --> 00:33:07,040 you're seeing is transmitted over very low frame rate, very low bit rate as well. I 433 00:33:07,040 --> 00:33:13,960 think we're doing 480p at 100 kilobytes a second. So it's very, very low bit rate. 434 00:33:13,960 --> 00:33:17,649 And when the robot isn't moving, we actually make it go black and white and 435 00:33:17,649 --> 00:33:22,129 even lower rate frame rate so that it doesn't waste resources. So yeah, it's 436 00:33:22,129 --> 00:33:28,109 pretty cool stuff. Over here on the top left we actualy have our latency. So we 437 00:33:28,109 --> 00:33:31,429 managed to build the infrastructure that allowed us to supervise this robots from 438 00:33:31,429 --> 00:33:36,590 Columbia for 200 milliseconds and less than 20 milliseconds. So it's like a blink of an 439 00:33:36,590 --> 00:33:40,929 eye. It was a really, really cool technology, it worked or 4G and we did a 440 00:33:40,929 --> 00:33:45,889 lot to optimize that. We had also a map over here. So this map is really, really 441 00:33:45,889 --> 00:33:50,130 cool. A lot of people ask us like, hey, did you do mapping? Did you map out your 442 00:33:50,130 --> 00:33:56,009 environment? Did you need to have something there before you came into a new 443 00:33:56,009 --> 00:34:02,259 place? And well the answer is no. But what we do instead is we actually map out the 444 00:34:02,259 --> 00:34:08,190 network conditions. So we would map out the network conditions of a city and we'd 445 00:34:08,190 --> 00:34:11,940 say, OK, these areas like over here. This is like high latency. We should avoid 446 00:34:11,940 --> 00:34:15,840 those areas because the robot could get stuck there. It is actually very 447 00:34:15,840 --> 00:34:19,770 interesting to see the network conditions change continuously. You didn't have the 448 00:34:19,770 --> 00:34:23,230 same network conditions every day, all day, all year. They'd actually change 449 00:34:23,230 --> 00:34:26,300 every few hours. So it was something that took us a while to figure out. 450 00:34:26,300 --> 00:34:30,230 *Takes a sip of Mate* So, of course, the way this works is we 451 00:34:30,230 --> 00:34:33,820 had two or three people supervising, sorry, two or three robots for a 452 00:34:33,820 --> 00:34:38,090 supervisor in Colombia, and we have just a bunch of people. Typically, students who 453 00:34:38,090 --> 00:34:41,360 would just be working part time and they were sitting in an office in Colombia 454 00:34:41,360 --> 00:34:47,580 doing this. Of course, the press found out about this and they wrote a very small bit 455 00:34:47,580 --> 00:34:53,270 of text in the article saying like, oh, Kiwi hires Colombians and pays them two 456 00:34:53,270 --> 00:34:59,330 dollars an hour. And people were really frustrated about that. We had a lot of 457 00:34:59,330 --> 00:35:03,500 interesting feedback about that. But what was interesting to see is that this 458 00:35:03,500 --> 00:35:07,890 technology actually helps people in Colombia. If you're there, it's a third 459 00:35:07,890 --> 00:35:11,051 world country it's a developing country. You can get a job at a factory. You can 460 00:35:11,051 --> 00:35:15,380 get a job at a textile shop, you can get a job maybe McDonald's. But there aren't 461 00:35:15,380 --> 00:35:21,860 that many tech jobs per say. The biggest employer in the country is a phone support 462 00:35:21,860 --> 00:35:25,950 company. So like when you call in to support line, you get connected to 463 00:35:25,950 --> 00:35:29,280 Colombia sometimes. And that's the biggest employer in the country. So in order to 464 00:35:29,280 --> 00:35:33,270 get like a tech job, it's really, really hard and giving people the ability to, go 465 00:35:33,270 --> 00:35:38,430 and supervise robots it's something that helped them get something on their CV and 466 00:35:38,430 --> 00:35:41,820 help them step up It helped them learn a little bit more about the technology and 467 00:35:41,820 --> 00:35:48,800 helped them progress in terms of their careers. Our lead A.I. guy, he actually 468 00:35:48,800 --> 00:35:53,290 started off as a supervisor and he went up through the ranks and then he ended up 469 00:35:53,290 --> 00:35:56,590 leading the A.I. and robotics team. So it was really interesting, really inspiring 470 00:35:56,590 --> 00:36:01,250 to see how that transition happened. And we managed to get our technology to work 471 00:36:01,250 --> 00:36:09,340 so well that we can do this. *Video of the inside of an airplane is shown* 472 00:36:09,340 --> 00:36:14,770 So we were able to get it to work with up to eight seconds latency, which meant that 473 00:36:14,770 --> 00:36:17,860 you can control it literally from anywhere in the world. So even from an airplane 474 00:36:17,860 --> 00:36:24,361 above the Pacific Ocean. So it was a really interesting experience. And we 475 00:36:24,361 --> 00:36:30,240 really try to make it simple. So in conclusion, for A.I., we realized that the 476 00:36:30,240 --> 00:36:34,030 best approach was to keep it simple. We tried a lot, a lot of different 477 00:36:34,030 --> 00:36:38,180 approaches, like we tried the traffic light detection. We tried a yellow pad 478 00:36:38,180 --> 00:36:42,680 detection. I didn't mention that. So in Berkeley, you have these accessibility 479 00:36:42,680 --> 00:36:46,650 ramps and you have yellow pads that blind people can actually feel them and see them 480 00:36:46,650 --> 00:36:51,910 easier. So we built the algorithm to detect that and we thought, OK, maybe if 481 00:36:51,910 --> 00:36:56,350 the robot is stuck in the middle of the intersection, you can automatically detect 482 00:36:56,350 --> 00:37:00,520 this yellow pattern and navigate to it. It's an approach that worked in theory, in 483 00:37:00,520 --> 00:37:06,260 practice it did not quite work. We tried segmentation. So that was an approach that 484 00:37:06,260 --> 00:37:12,070 worked OK. But some weird things broke it. So for example, any lamp posts or bicycle 485 00:37:12,070 --> 00:37:17,380 posts would crash the robot because it didn't see it. So yeah, keeping it simple 486 00:37:17,380 --> 00:37:22,590 was the best approach, really not going too crazy. And the approach we ended up 487 00:37:22,590 --> 00:37:27,160 going in the end was to have it more of like a driver assist type, like a parallel 488 00:37:27,160 --> 00:37:32,380 approach, parallel autonomy approach, where our robots would help people the 489 00:37:32,380 --> 00:37:37,000 same way that cars would help people stay in lanes or have cruise control or like 490 00:37:37,000 --> 00:37:39,460 with parking assistance. That's kind of the approach we're having. I think long 491 00:37:39,460 --> 00:37:43,621 term it is gonna be possible to build robots more autonomous, it could be 492 00:37:43,621 --> 00:37:48,720 Starship that have some interesting ideas about how to solve that. But I don't think 493 00:37:48,720 --> 00:37:53,220 it's quite something I could be scaled to every city just yet. Another really 494 00:37:53,220 --> 00:37:58,900 important thing is, the lab does not equal the real world. So there are many, many 495 00:37:58,900 --> 00:38:04,870 great examples of fantastic research papers from some great groups and they 496 00:38:04,870 --> 00:38:09,751 were great with very polished, very clean datasets. But they did not work when you 497 00:38:09,751 --> 00:38:14,740 deployed them on 100 robots, there were all different. They all had slightly 498 00:38:14,740 --> 00:38:18,470 different camera calibration that all had slightly different hardware, it all had 499 00:38:18,470 --> 00:38:23,250 slightly different chassis. It did not really translate as well. So these 500 00:38:23,250 --> 00:38:28,640 algorithms, these lab best case scenarios, will need to be modified a little bit. 501 00:38:28,640 --> 00:38:35,050 What else? Yeah, one thing, maybe jumping back to the keep it simple. We decided to 502 00:38:35,050 --> 00:38:39,790 put in a very simple safety mechanism. So the robot actually breaks if it sees 503 00:38:39,790 --> 00:38:43,660 something within 50 centimeters in front of it. So as kind of like a last measure, 504 00:38:43,660 --> 00:38:47,110 precaution, as you saw before, there is a video like you can supervise the robot 505 00:38:47,110 --> 00:38:51,501 from anywhere in the world, but a lot of latency. But having this 50 centimeter 506 00:38:51,501 --> 00:38:56,780 like hard break, actually saves us in case the robot loses connectivity or the 507 00:38:56,780 --> 00:39:01,680 supervisor is no longer able to supervise the robot. So it's always breaking 50 508 00:39:01,680 --> 00:39:07,940 centimeters away from any collision with like a baby or a car or whatever. So the 509 00:39:07,940 --> 00:39:12,420 approach we really thought about is, how can we expand human potential? There is a 510 00:39:12,420 --> 00:39:18,840 lot of talk about A.I. taking jobs or A.I. replacing people's roles, but we sort of 511 00:39:18,840 --> 00:39:22,290 kind of try to do that and it didn't work. Like we try to build robots that were 512 00:39:22,290 --> 00:39:26,170 fully autonomous that went from the restaurant to your door and that didn't 513 00:39:26,170 --> 00:39:29,190 work. People were waiting a very long time. These robots required an obscene 514 00:39:29,190 --> 00:39:34,230 amount of maintenance. So we ended up going for an approach that was far more 515 00:39:34,230 --> 00:39:37,850 parallel autonomy where these robots were like helping people to do more. Same way 516 00:39:37,850 --> 00:39:41,980 the supervisors are getting these assistive technologies where they able to 517 00:39:41,980 --> 00:39:45,530 set a waypoint to do the path finding and the robot does the motion planning on 518 00:39:45,530 --> 00:39:49,500 board. We also had the couriers who would just load food into the robots instead of 519 00:39:49,500 --> 00:39:54,190 the robots picking up food from the restaurant directly, so really expanding 520 00:39:54,190 --> 00:39:57,620 human potential. I think that's where it's at. And over the course of the past 521 00:39:57,620 --> 00:40:01,270 century, we've seen a lot of examples of this. Like we've seen operators of 522 00:40:01,270 --> 00:40:05,830 elevators. Like before, elevators had operators who would make it go up or down. 523 00:40:05,830 --> 00:40:09,320 And now they're fully automated. We had switchboard operators who were there to 524 00:40:09,320 --> 00:40:12,210 connect phone calls. Now we can make a phone call to anywhere in the world 525 00:40:12,210 --> 00:40:17,260 instantly for free. So we're seeing this transformation of work and transformation 526 00:40:17,260 --> 00:40:21,260 of the way things are done. And I think this is just the start. The way I see 527 00:40:21,260 --> 00:40:27,140 these robots is really meshing into the fabric of our societies and solving 528 00:40:27,140 --> 00:40:31,830 physical transportation. Like, sure, you can move bits from anywhere to anywhere in 529 00:40:31,830 --> 00:40:35,770 the world, but can you move atoms? It's really expensive to do that. It's really 530 00:40:35,770 --> 00:40:43,630 hard to do that. That's why I see robots expanding human potential. So. 531 00:40:43,630 --> 00:40:50,360 Conclusions. What we did was really cool and I think it was a cool experience. One 532 00:40:50,360 --> 00:40:55,650 thing that we realized is that tech isn't the hardest part, right? We spent a lot of 533 00:40:55,650 --> 00:41:01,630 time thinking how to build something, but figuring what to build is sometimes very 534 00:41:01,630 --> 00:41:05,930 important as well. And I don't think we spent enough time asking ourselves that 535 00:41:05,930 --> 00:41:09,840 question. We kind of went in all sorts of directions we didn't focus as much on 536 00:41:09,840 --> 00:41:15,590 making the best product possible. We kind of tried things that were really weird and 537 00:41:15,590 --> 00:41:18,630 not well thought out. So like having that more long term thinking, like thinking 538 00:41:18,630 --> 00:41:22,180 what should we build is very important because like how, you can just look up a 539 00:41:22,180 --> 00:41:27,350 tutorial on Google and figure out how to build robots. It's not the end of the 540 00:41:27,350 --> 00:41:31,310 world. One really important thing for us was interaction. So interacting with 541 00:41:31,310 --> 00:41:34,720 people, figuring out how to make the door open, when you actually received your food 542 00:41:34,720 --> 00:41:39,870 was super hard, super, super challenging to do. Actually, the only robot that opens 543 00:41:39,870 --> 00:41:43,720 the door for you. Other companies like Starship, for example, they have a button 544 00:41:43,720 --> 00:41:46,730 that unlocks a solenoid. So it's like the experience is not quite there yet to bend 545 00:41:46,730 --> 00:41:51,030 it down. You have to figure out how the door actually opens. So we spent a lot of 546 00:41:51,030 --> 00:41:55,400 time, a lot of effort in order to optimize that experience to make it as smooth as 547 00:41:55,400 --> 00:42:00,630 possible for people. Also, one thing we didn't figure out is financing. I'll come 548 00:42:00,630 --> 00:42:04,440 back to that in a second. That was really, really hard to do as well. So like tech, 549 00:42:04,440 --> 00:42:08,130 you know, not the hardest, financing figuring out like how to manage cash flow, 550 00:42:08,130 --> 00:42:12,300 super important. But I think the most important thing is to work with a great 551 00:42:12,300 --> 00:42:17,471 team. If you're going to be spending a lot of time with people who you eat, live and 552 00:42:17,471 --> 00:42:21,650 breathe with, it's really important to choose a team that you really connect with 553 00:42:21,650 --> 00:42:25,880 and then share the same passion as you do, because you could be miserable making an 554 00:42:25,880 --> 00:42:30,320 amazing amount of money, but if your with a really crappy team with a high turnover, 555 00:42:30,320 --> 00:42:32,751 it's really boring. I was really fortunate to work with one of the best teams in the 556 00:42:32,751 --> 00:42:36,800 world and over the course of the past two and a half years we managed to do quite a 557 00:42:36,800 --> 00:42:41,310 lot. And just last month we actually got an article in The New York Times. So that 558 00:42:41,310 --> 00:42:44,890 was a really big accomplishment for our team and we got to share it with our 559 00:42:44,890 --> 00:42:50,310 families. My mom was really proud. So a lot of great traction and a lot of great 560 00:42:50,310 --> 00:42:54,563 coverage. But unfortunately, we actually ran out of money, so we kind of ran out of 561 00:42:54,563 --> 00:43:00,300 money last month and we are no longer delivering things. So I decided to leave 562 00:43:00,300 --> 00:43:05,730 and start my own thing instead of doing robots. I decided to do data. So now I'm 563 00:43:05,730 --> 00:43:11,141 actually focusing more on building a tool that helps you tell stories with data. So 564 00:43:11,141 --> 00:43:16,450 this is Glint. This is a data storytelling tool. You're able to drag in some files 565 00:43:16,450 --> 00:43:21,900 and it tells you the story of your data without you having to write any code. So 566 00:43:21,900 --> 00:43:25,830 my hope for this is to allow anybody in the world without any knowledge about how 567 00:43:25,830 --> 00:43:29,500 to wrangle data, how to clean data, how to analyze data, to be able to tell stories 568 00:43:29,500 --> 00:43:35,320 with their data directly from their computers. I'm imagining a tool where you 569 00:43:35,320 --> 00:43:41,750 can say, oh, "In December there were X X visitors to Congress" or "Last summer we 570 00:43:41,750 --> 00:43:45,310 had X X sales" and automatically filled out for you. That's kind of what I'm 571 00:43:45,310 --> 00:43:49,131 thinking about. If you want to join the effort, there is a Github. I'm more than 572 00:43:49,131 --> 00:43:53,880 happy to have any contributors. And if you have any questions or comments we're happy 573 00:43:53,880 --> 00:44:02,280 to answer on Twitter or here in person. Thank you. 574 00:44:02,280 --> 00:44:11,370 *Applause* 575 00:44:11,370 --> 00:44:18,460 H: So, as usual, feel free to line up in front of the microphones or write your 576 00:44:18,460 --> 00:44:23,970 question to the signal angel over there. That already has one. Um, it's all the way 577 00:44:23,970 --> 00:44:28,690 down. Go ahead. Signal Angel: OK. Here is a user of your 578 00:44:28,690 --> 00:44:34,340 service who apparently got an e-mail from you that announced some changes. So he's 579 00:44:34,340 --> 00:44:38,200 wondering, what's up? What you're going, what you're planning to do there, whether 580 00:44:38,200 --> 00:44:40,300 you're continuing your service or closing shop? 581 00:44:40,300 --> 00:44:45,020 S: Yeah, it's unclear. We ran out of funds, so I think the CEO is still trying 582 00:44:45,020 --> 00:44:48,860 to figure out what to do with that. I wish him the best of luck, but I ended up 583 00:44:48,860 --> 00:44:52,840 leaving with a lot of other people. So we have like 50 people in November, now we 584 00:44:52,840 --> 00:44:57,150 have like 10 people left in the country. So it's very ambiguous what's happening, 585 00:44:57,150 --> 00:45:01,630 but yeah, I left. H: Yeah, microphone? 586 00:45:01,630 --> 00:45:08,910 Audience Member 1: No audio. Okay, now it works. I'm a little bit confused because 587 00:45:08,910 --> 00:45:14,540 you are presenting a 1970s concept of a manipulator, because a robot is something 588 00:45:14,540 --> 00:45:18,690 that works by itself, a manipulator is somebody who has some joysticks and moves 589 00:45:18,690 --> 00:45:22,930 things. So it's nothing special. You just have a interlinked Internet link for 590 00:45:22,930 --> 00:45:27,040 manipulator and in the 70s, there were cables. So what's the special thing? 591 00:45:27,040 --> 00:45:30,730 S: Yeah, that's a good question. I think the magic here is connecting everything 592 00:45:30,730 --> 00:45:34,510 together, figuring out for us how to build these robots, how to build a reliable 593 00:45:34,510 --> 00:45:38,640 connection, and how about a platform that works. And as I mentioned, like the how 594 00:45:38,640 --> 00:45:43,310 that's not that interesting. It's more of the what you build. It's that experience 595 00:45:43,310 --> 00:45:46,190 where you're able to order anything you want at any time and get it delivered in 596 00:45:46,190 --> 00:45:48,890 under 30 minutes virtually for free. So that's good. 597 00:45:48,890 --> 00:45:53,450 AM1: So, so far, so good, but evil people could just buy a remote control car, put a 598 00:45:53,450 --> 00:45:59,630 bomb in it, drive under a police car and make boom. And so it's the same use case. 599 00:45:59,630 --> 00:46:06,520 You deliver something by remote control. Audience Member 2: Yeah. You talked about 600 00:46:06,520 --> 00:46:13,910 iterating quickly and rapidly and that's very good model for conceptual stage and 601 00:46:13,910 --> 00:46:18,530 software. Were you in the stage where you were leasing your hardware with your 602 00:46:18,530 --> 00:46:22,580 iterations? Because usually a thick stack of certification has to come in between. 603 00:46:22,580 --> 00:46:27,640 S:So I'm not entirely sure. Are you asking if we got certified at every single 604 00:46:27,640 --> 00:46:30,720 release? AM2: I suppose. Yeah. What level of like 605 00:46:30,720 --> 00:46:36,080 recertification was it totally released. So you had to meet like regulations for 606 00:46:36,080 --> 00:46:39,660 each iteration of that? S: Yeah, absolutely. We didn't really get 607 00:46:39,660 --> 00:46:43,470 certified because we're not building hardware product for consumers. So we're 608 00:46:43,470 --> 00:46:46,670 not selling it to anybody. We're operating it ourselves. So we don't fit under the 609 00:46:46,670 --> 00:46:50,940 same kind of requirements. However, we did have to have some permits. And part of the 610 00:46:50,940 --> 00:46:54,241 conditions that these permits was that we had to meet some expectations. But they're 611 00:46:54,241 --> 00:46:58,320 very, very basic. And there were rigid like an FCC or a CE certification, for 612 00:46:58,320 --> 00:47:01,120 example. AM2: That was the question. Yeah. Thanks. 613 00:47:01,120 --> 00:47:07,320 S: Thank you. SA: Another question from the Internet. 614 00:47:07,320 --> 00:47:11,520 "Why did you develop different applications for Android and iOS?" 615 00:47:11,520 --> 00:47:18,040 S: For the consumer application? SA: I haven't got any more details. 616 00:47:18,040 --> 00:47:25,290 S: We just did. I mean, we had first an iOS application. I mean, 80 percent of our 617 00:47:25,290 --> 00:47:29,030 customers are using iOS. So we really spent a lot of effort like polishing that 618 00:47:29,030 --> 00:47:33,620 iOS experience, making sure that worked. And at one point, our Android app was 619 00:47:33,620 --> 00:47:39,240 working super badly. So we decided to kill it. And everybody was really, really 620 00:47:39,240 --> 00:47:43,730 pissed off, extremely pissed off. So we actually reintroduced it and we started 621 00:47:43,730 --> 00:47:47,860 catching up with features to the iOS version. Internally, all of our apps are 622 00:47:47,860 --> 00:47:53,850 built in React and React native. So we had a common framework for all of our internal 623 00:47:53,850 --> 00:47:56,890 apps, but we didn't have that experience. Where you're expecting the quality of 624 00:47:56,890 --> 00:48:02,100 experience, that we're expecting from a consumer app using React. That's why we 625 00:48:02,100 --> 00:48:13,650 had two different code bases. H: Have tried different methods regarding 626 00:48:13,650 --> 00:48:19,520 perception? For example, lidar, radar and what are your conclusions from that? 627 00:48:19,520 --> 00:48:24,030 S: Yeah, we tried lidar, we tried the cheap lidar we didn't try the really high 628 00:48:24,030 --> 00:48:27,740 end lidar. So the challenge with having like point clouds is that you have to 629 00:48:27,740 --> 00:48:32,570 compute, spent a lot of time competing. We were using a relatively low power device 630 00:48:32,570 --> 00:48:35,900 and it was running from batteries. So we didn't have the luxury of having like 10 631 00:48:35,900 --> 00:48:40,610 GPUs is in the trunk of a car, for example. So that was one approach. One 632 00:48:40,610 --> 00:48:45,910 question. Another question is how much does it cost? So lidars, they can cost ten 633 00:48:45,910 --> 00:48:49,040 thousand, hundred thousand dollars. Our bill of materials was 'round two and a 634 00:48:49,040 --> 00:48:53,140 half thousand. The last versions are two and a half thousand. So all of our sensors 635 00:48:53,140 --> 00:48:57,770 were very minimal. In terms of what sensors we tried? We tried a lot of 636 00:48:57,770 --> 00:49:04,480 different sensors. We tried ultrasonic sensors. We tried near field infrared 637 00:49:04,480 --> 00:49:11,360 sensors. We tried other sensors. Yeah, we tried a lot of different sensors. We are 638 00:49:11,360 --> 00:49:15,880 ended up just going with cameras. So we have cameras. We have six cameras onboard, 639 00:49:15,880 --> 00:49:22,880 all of them full HD. We stitch them into an image on our compute module and then 640 00:49:22,880 --> 00:49:25,781 the supervisor decides which portion of the image they want streams. They can 641 00:49:25,781 --> 00:49:29,620 manipulate with the keyboard to see which portion of the image is streamed. So we 642 00:49:29,620 --> 00:49:32,540 don't stream the whole image. We just stream a part of it. The really important 643 00:49:32,540 --> 00:49:36,740 part for us was to make something that's viable, that can be used commercially. I'm 644 00:49:36,740 --> 00:49:40,720 sure lidar is really cool, but I'm not seeing any commercial deployments of lidar 645 00:49:40,720 --> 00:49:45,710 based autonomous vehicles or robots yet. Audience Member 3: Thank you. 646 00:49:45,710 --> 00:49:54,501 Audience Member 4: You've tried out many different concepts how to do it. And you 647 00:49:54,501 --> 00:49:59,490 saw that your company ran out of money. Do you still believe in the business concept 648 00:49:59,490 --> 00:50:06,780 of robots delivering packages of food? S: Who knows? I think I think it was a 649 00:50:06,780 --> 00:50:11,500 great learning experience. We learned a lot. We had a great team. And I think 650 00:50:11,500 --> 00:50:14,230 we'll see some concept of robots. Maybe not exactly what we were building, maybe 651 00:50:14,230 --> 00:50:17,840 something a little bit different, but I think it's a little bit inevitable, 652 00:50:17,840 --> 00:50:20,530 especially with the rise of self-driving cars. Maybe we'll have cars delivering 653 00:50:20,530 --> 00:50:24,760 packages instead of robots. Not entirely sure what it would look like. I could tell 654 00:50:24,760 --> 00:50:27,861 you, Amazon, they bought one of our competitors dispatch labs. So they're 655 00:50:27,861 --> 00:50:31,650 making a big bet on this. There are two delivery companies in the US, Postmates 656 00:50:31,650 --> 00:50:35,800 and DoorDash, that are building products internally also for... with delivery 657 00:50:35,800 --> 00:50:39,520 robots. And also companies like FedEx are also building delivery robots. And then 658 00:50:39,520 --> 00:50:42,900 we have companies as Starship, for example, which are building robots and 659 00:50:42,900 --> 00:50:47,590 doing B2B with companies all over the world. So, yeah, I think we'll see some 660 00:50:47,590 --> 00:50:50,610 form of delivery robots. I don't know if it's going to be what we had or what 661 00:50:50,610 --> 00:50:58,540 somebody else is going to have. Audience Member 5: Were there any safety 662 00:50:58,540 --> 00:51:02,920 certifications you had to satisfy in order to operate around people? 663 00:51:02,920 --> 00:51:10,050 S: No. So. *laughter* Well, the thing is, in the US, like, it's kind of just do 664 00:51:10,050 --> 00:51:13,830 whatever you want. It's very different from Germany. You can kind of just do 665 00:51:13,830 --> 00:51:17,360 things and you can do them until you get in trouble. So we kind of had that 666 00:51:17,360 --> 00:51:20,470 approach don't ask for permission, don't ask for forgiveness. We ended up having to have a 667 00:51:20,470 --> 00:51:24,731 permit in the cities we operate in. But it was very simple. It was like, OK, you have 668 00:51:24,731 --> 00:51:30,520 to have lights. You have to have a phone number and you can not go in these areas. 669 00:51:30,520 --> 00:51:33,640 That was essentially all the authorization, all the permitting and 670 00:51:33,640 --> 00:51:36,550 certification that we had. H: Yeah? 671 00:51:36,550 --> 00:51:44,100 Audience Member 6: I wanted to ask, did you try other markets? Like, autonomous 672 00:51:44,100 --> 00:51:49,320 driving is very hard, even way more than... manage it fully. So like perhaps 673 00:51:49,320 --> 00:51:53,800 elderly care, like you could use this robots in elder care where you have a 674 00:51:53,800 --> 00:51:59,900 controlled environment where everything is the same. Did you search after other 675 00:51:59,900 --> 00:52:05,880 markets where it's less... S: Yeah, that's a great question. Yeah, 676 00:52:05,880 --> 00:52:10,000 there is a lot of potential for markets like elderly care, for example, also for 677 00:52:10,000 --> 00:52:14,930 mail delivery, for applications inside of factories. We had a couple different 678 00:52:14,930 --> 00:52:18,660 medical companies that reached out to us and like, hey, we want to move items, move 679 00:52:18,660 --> 00:52:22,372 packages inside of our facilities. So we did have a lot of interest. We tried to 680 00:52:22,372 --> 00:52:25,480 keep a focus on the consumer space, like really building a consumer experience that 681 00:52:25,480 --> 00:52:29,400 worked out before branching out into these more B2B approaches. Where elderly care 682 00:52:29,400 --> 00:52:34,300 could be one of them. I think one important thing about elderly care and 683 00:52:34,300 --> 00:52:38,160 services like Meals on Wheels, for example, is that human contact. So I think 684 00:52:38,160 --> 00:52:41,830 people who are maybe not seeing as much of their family, of their relatives, they 685 00:52:41,830 --> 00:52:45,300 really cherish that connection they get from people who deliver them food. So I 686 00:52:45,300 --> 00:52:48,620 think it's a multifaceted approach they have to have. You have a couple of 687 00:52:48,620 --> 00:52:54,640 different considerations with these kind of services for the elderly, for example. 688 00:52:54,640 --> 00:52:59,900 AM 6: Thank you. Audience Member 7: What kind of 689 00:52:59,900 --> 00:53:06,770 personality do Chinese entrepreneurs have? S: *laughs* I think, as I mentioned like, 690 00:53:06,770 --> 00:53:10,740 it's really important to have relationships. So they were very 691 00:53:10,740 --> 00:53:16,051 interesting. They were very deeply in belief of their government. They had 692 00:53:16,051 --> 00:53:20,380 nothing bad to say about it. They believe they would bring them everything... the 693 00:53:20,380 --> 00:53:23,330 best possible, even though they still try to access Facebook and Twitter with VPNs. 694 00:53:23,330 --> 00:53:28,230 So they were very, very loyal to their governments. They were very, very 695 00:53:28,230 --> 00:53:32,820 diligent. If they committed to something, they would usually deliver on that. They 696 00:53:32,820 --> 00:53:38,170 really wanted to make sure you had a good experience. And also what we saw, for 697 00:53:38,170 --> 00:53:41,060 example, with building up these relationships, like the first few times we 698 00:53:41,060 --> 00:53:45,300 talk, they would try everything to impress us. So we got taken to these ridiculously 699 00:53:45,300 --> 00:53:50,490 expensive restaurants to make sure that we were welcomed well and make sure 700 00:53:50,490 --> 00:53:55,200 everything was right. I actually had an interesting episode earlier this year. I 701 00:53:55,200 --> 00:53:59,560 was going to go to Burning Man and then all of a sudden one of my colleagues had 702 00:53:59,560 --> 00:54:04,080 an argument with my manufacturer about whether Hong Kong is another country or 703 00:54:04,080 --> 00:54:09,900 not. And I ended up having to go to China to deal with our manufacturer instead of 704 00:54:09,900 --> 00:54:14,200 going to Burning Man to make sure we're aligned in terms of our beliefs. So, 705 00:54:14,200 --> 00:54:17,830 sometimes it's really delicate. You cannot, like talk too much about the 706 00:54:17,830 --> 00:54:20,610 government there. You can't talk too much about politics. It's best to just stick to 707 00:54:20,610 --> 00:54:25,150 business and, yeah, focus on building a product. 708 00:54:25,150 --> 00:54:30,055 H: I guess this was it. Thank you. S: Thank you so much. *Applause* 709 00:54:30,055 --> 00:55:00,000 subtitles created by c3subtitles.de in the year 2020. Join, and help us!