The Robots Are Coming with Peter Barrett (Playground Global), Helen Boniske (Lemnos VC), Claire Delaunay (Nvidia) and Cyril Ebersweiler (SOSV/HAX) | Disrupt SF (Day 1)
6:20AM Sep 6, 2018
self driving car
Now the robots are coming. Are the robots coming? That's the question. You know there's talking about AR the marriage between AI and robots is going to be a fascinating conference in the next few years. So here to discuss, this is Peter Barrett from playground. Global. Helen boniske from lemnos. vc, Claire Delaunay from Nvidia and Cyril ebersweiler from sov ha x x x. Very hard wary
to discuss robots, ai sensors, GPUs and the changing labor markets, remaking the robotics world to moderate this panel. We have once again Danny Creighton, who's Editorial Manager at TechCrunch. Big round of applause, everybody. Thank you.
Alright, everyone, thanks for joining. We got a sellout crowd here. So obviously, robotics is top of mind, I think for a lot of people. Thanks for coming back from lunch. We have a absolutely distinguished group of investors here. But what's crazy about the robotic space is just a couple of years ago, there was no real robotics investors at all. If we looked at seven, eight years ago, and now we're seeing dozens of startups being funded several have grown very quite large we have several large exits in the space and so I'm looking forward to the panel today we don't want to start out we obviously have a lot of founders in the room some of them who are considering starting a robotics company how what kind of backgrounds to people need today and I'm going to start with surreal on this one we're gonna backgrounds two people need in the robotics space today. Do you need to have a robotics PhD to do well in the space? Is that the critical kind of skill? Yeah, I think things have been evolving quite fast. Over the past few years. I'd like to take the you know, the example of one of the most well known robot that kind of Kickstarter, the reffing called pure to
a few years back and you know, the team that built all of this was literally rocket scientists. And, you know, maybe you call them roboticist as well. And what's been really interesting over the past two to four years is like, the evolution of technology in building on the shoulders of giants as well of those technologies really has kind of changed, I think, the profile of the founders and the group of founders that can build a robot at the first place. And I think the the technicalities of it, of course, are still extremely important. But the more we look at robots, the more we look, we don't get the technology, we really look at the applications.
And Helen, you're also sort of at the seed stage with lemnos Labs and mean, what are you seeing as a pattern for the backgrounds of founders these days in the robotics space, I think
the ideal founding team is going to consist of that roboticist or technical leader, you're also going to need some domain expertise around whatever vertical that you're considering. Selling into selling robotics is actually really hard. It often involves multiple people agreeing to a sale, which could be somebody who's head of security, security, head of it, head of facilities to all come together and say, yes, this works for us,
when you look at the, you know, starting profiles for all of your companies, you know, are they coming more from the hardware innovation side? Are they coming more from the software side? Obviously, Claire, you're coming from Nvidia. So you obviously have a lot of people coming in on the hardware side. But what sort of the pet interesting on that and that world?
Well, I'm a robot, it's really an assembly of, of hardware and software. And the more we we push intelligence to hardware by, for instance, flushing or deploying a neural network inside the sensor itself, the more the frontiers between the hardware or software are blurring. And that's what's beautiful about robotics. It's, it's that there's no magic answer to to build a robot, it's really an assembly of a lot of different skills that have to work very well together in order to solve a specific problem in Peter
you as well. I mean, you come from a hardware background, but you're also increasingly looking at different software layers in the space. And what what are you seeing with playground? Well, I
think the scope of what a robot is, is broadening my we consider it to be anything with sensing and acceleration. And so a camera is much a robot as an atlas city as a robot. So the breadth of disciplines required to animate those pieces of technology to continue to attach the sensing and acceleration to intelligence is actually rolling in a whole bunch of software disciplines that haven't actually been closely correlated with robots access yet,
let me ask you, I mean, what are the challenges as you have to bring so many different components together in order to make a robot, whether it's sensing and perception, and then the GPUs and some of the technology behind that arms and articulation to be able to actually act in space? Are you seeing improvements for early stage startups to be able to sort of get their MVP is in the market to actually
put those components together quicker, so that they're, you know, first kind of demo robot isn't the Series B, but something else. Yeah, I
mean, I think you want most of your capital to go into new technology and not reengineering, shaky the robot from 50 years ago. And I think probably 90% of the capital that goes into robotics now go sideways on stuff that just isn't important and isn't part of the key application. So there's a, there's a huge opportunity for platforms to make that capital more efficient. And you see that pattern for the other panelists as well. It's quite a, I will call it the prehistoric, you know,
of robotics. But, but we're still at the very early innings. If you're, you know, we, we are a few things that I haven't happened. Like, of course, you don't build an iPhone with an Arduino board. And so that's, you know, and video and others are working on much more powerful technologies and micro processors that will actually suit the needs of robots, which ritually were not available two to four years ago. And it's the same surrounding the, all the sensors, the motors, for example, you know, the resolution for four motors is a is probably more akin to a Gameboy than an HDTV today, about how much you know, control, you can have override the battery technology is just all of this actually needs to, to continue to, to move forward. And, and then we probably will have, or Macintosh moments of, of the robotics, which we haven't seen yet.
And hello. But you were
starting to see a lot of platform companies that are solving part of the stack that maybe they they're uniquely suited to do, and a lot of our companies are struggling with. And it's often our portfolio founders who bring these companies to us with, hey, they solve this piece really, really well. And then they're doing it, you know, for these five companies. Is that an interesting investment? And I think we're starting to get to the point where, where, yes, we are seeing fleets of robots in the field, doing interesting things, we're a long ways from, you know, 10s of thousands, hundreds of thousands, millions. But some of these companies, if maybe it's a component that they do really well, now can start to slice across different verticals. And how
do you think about I mean, there's sort of a chicken and egg problem, right? Like, you know, let's say you build a great arm you need robots actually uses ours, but those robots need the arms in the first place to be relevant to their, you know, particular applications, like how do you, you know, as an investor or from the video perspective, like as, as, you know, all your customers? How do you how do you get all the sales processes in place here in the right order?
So, yeah, I think it's, it's, it's a fascinating problem. And because this is really have been doing robotics for more than 15 years, and that's the recurrent pattern is is this chicken and egg problem is how do you as a startup, for instance, how do you survive the first two years by trying to put all the CES hardware that doesn't exist anywhere else together. So even just create a demo robot. And so the reality is that if we look at for instance, cell phone before the cell phones were so democratize and we have a smartphone now, everywhere, and I am you was extremely expensive. And the simple fact that we find we found a single product that concentrate all this hardware and that we can mass produce completely changed the prices of like IMU, magnetometers and all of that, and suddenly, we started seeing software around it. And like, people started building on top of it. And today, very interestingly, we see the same thing happening in self driving cars. Because Finally, I mean, self driving car easily robots, right? It's actually doing something very useful. So we call the self driving cars, what is actually supposed to do not robots. But
all the older the startup working on in companies working on self driving car, the old needs are some sensors. And he turns out the sensors are also the one we need for robotics, which is fantastic because we buy buy, buy some movement that's happening in the economy nowadays, we can actually capitalize on it into we can boost up a lot of like very brand new application in terms of robotics and things that would not be possible to walk on if we didn't have this Sonali momentum on the hardware. Yeah,
it would be awesome if Abella dine 128 cost the same as a cell phone I am you.
But there's definitely a sense that this is robotics is like cell phones in 2005. So what lies ahead or some normal platforms and an explosion in hardware, because all of a sudden, broader set of people can build really exotic hardware, very capable hardware if they're not hamstrung by the limitations of software. And if you look at industrial arms today, they're amazing machines. They're incredibly accurate, there only differentiated on how bad the software is right now. So I think there's some real opportunities to go through that same watershed that funds went through, we're
talking about autonomous cars mean robot today are very kind of independent, they don't really interact with each other, or they interact, you know, spatially but they don't interact in terms of data is sort of any sort of software layer. I mean, are you starting to see how robots are actually communicating? I know, Peter, you spent quite a bit of time thinking about this sort of, you know, network approach to your robot. Yeah,
I mean, I think we try and solve autonomy, very anthropic Lee, and then we try and make these things sentient and independent. And I think they work much better when they can collaborate with each other. And cities, you know, we can't read each other's minds will see through each other's eyes, because can do that robots can do that. And so if they talking to themselves, or talking to cities, they can do superhuman things, and we know how to write that software. So thinking about robots in aggregate thinking about these applications, and aggregate changes when you think about the platform? And
are you seeing any sort of open source or kind of open standards around robotics? I mean, is it too early to go on some of this sort of stuff that I don't even know if any of your seeing it submit? The answer might be No. Are there any sort of patterns in that category?
There's a lot of there's a lot of open source software Ross is a very good example of high level software.
But we also see standardization in ADS as a sensor level on like, what kind of data do we need forms a sensor, which is very interesting because
because we need we need again, we need to, we need to be able to leverage very stable blocks to be the robots because building a robot is extremely complicated. And so you have to standardize somewhere in yourself back. And when you can standardize on the image formats, for instance, or on the on the detection detection that are coming directly from the sensor, you are able to exchange and compare different approaches, different sensors, with the same results on your side, on the software side, on the decision making side and,
and that's Yeah, that's, that's very, very interesting attended 10, I
will say, there's one big missing piece, which is there's, there's standardization of knowledge and, and representations of knowledge and experience are completely missing. So we we have JPG and videos so we can train neural networks to do perception tasks. We don't have standards for robotic forced talk sensing, or we don't have standards for RL learning of arms. And so in order to, it's not just the open source software that needs to be there. It's these common representations of knowledge. so other people can train on data that you collected, and you can build that corpus like JPG is incredibly important for robotic perception. And everything else is missing. So
let me ask you, Helen, and surreal. I mean, both of you were talking about collaboration have sort of accelerator models or colocation models. And Peter you as well with playground you're putting founders together, they're learning from each other. I mean, are there sort of lessons learned that are being shared from those sorts of groups? Yeah, there's
definitely synergy when you put a lot of companies solving different problems, but working on similar pieces of technology, they can do design reviews for each other, they can share customer sales experiences, I think, all beneficial to the group. And and
Helens startups are mostly in San Francisco, is that correct?
not accurately. So we'll actually invest across the US that's something that we've become more open to recently, I think there was an article in The Economist earlier in the week that maybe the peak
Valley Yes, that's definitely
something that US and other investors are becoming more open and even looking outside of the valley more aggressively.
Absolutely. And it's real. I mean, you've been a heavy investor, particularly in China and Asia more broadly, and around the world. I mean, what are you seeing? I mean, do you believe in sort of this peak Valley hypothesis? Or is the robotics industry still really centered here?
Yeah. So so we have a you know, fairly large office in Shenzhen. And what's been great was to additionally see the same thing as a Helen as you have pores of, of technology is going together the can definitely stay abreast of the latest latest technologies, and they are still not economies of scale on on on volumes and whatnot. But we also have some, some alumni from, you know, four or five years ago that now have fleets of hundreds of robots out there. And so, so really helps to, to figure out the path and it's not, you know, the technologies, great next things, events, but I think the next frontier is definitely to find what what customers really want, which is a common issue with technology is looking for a problem and robots, dicks or a technology
as we understand it, not really an industry and it's really important to confront it to reality as fast as possible. I want I want to follow up on that project. Specifically, who are we designing robots for you think about customers? I mean, who ultimately are the customers? This is enterprises? Are we talking consumers? Is it you know, into the supply chain? And anyone who wants to sort of engage on that, you know, feel free to stand up? But who are the customers for these sorts of technologies today?
Yeah, well, you can definitely divide them between consumers and robotics which everybody has seen or heard off today and the interacts and are trying to interact with you. And the first one I guess the first real consumer robotics company was it was iRobot in and having a Roomba around but over the past few years the thing has been you know the up a bit you probably have seen some some humanoid robots, educational robots have been coming up and rubbers and don't look like robot like for example nine button there they are, they are kind of you know, scooters that running around on their own and do surveillance now, so. So as a consumer part, but where the most exciting things are happening, obviously, on the, on the b2b side, and where we, we hear the most about it is in in manufacturing, and any anything that robots will, you know, take over jobs and whatnot. And it's definitely true. And, you know, it's, it's a, it's actually a good thing because we're wishing the bigger opportunities are, are in the free these, the, the, the, the door and dangerous and dirty, you know, jobs essentially, that can be replaced by robots. And you see, I mean, this is for Helen, you have a company here in San Francisco that's doing the robotic hamburgers. You know, this is sort of, like a b2b to see kind of model. I mean, does that something that you're seeing as well in the market? Or is that sort of like a one off play? No,
we actually did something slightly similar that we can't disclose yet. But I think that's a model of type of robot that is a long ways away from being a personal consumer thing that you have in your house. But in the meantime, you can set up a retail location that can delight lots of consumers. Yeah,
one of the biggest questions I have about robotics. So we're talking about representations we're sharing data is this a winner takes all market, you know, when you look at kind of the robotics technologies, perception is machine learning. So the bigger the data sets, the better you do, you know, way mo has driven more than a million miles or something along this line? Is that something that you're worried about? Is it a winner takes all? Or is it possible to have 50 different stacks? Well,
I mean, I think over time collecting data to do interesting things, collecting skills for robots is going to be incredibly valuable. There is no robot out store iRobot app is intelligence and experience the robot has. And so being able to use that knowledge across heterogeneous devices and make it repeatable, I think, is a is a real, real valuable thing to have as a lot of tooling that needs to get built before you can do that. But at the limit, to be able to technology transmitted the various robots and have them wake up smart to do new things is, you know, it's it's plausible, just need to do some, some underlying work to make it real. And Claire, you've also thought a lot about this sort of topic as well. You know, do you see it as a winner takes all market from the Nvidia perspective?
Um, so that's a very good question. I,
I think so. When I'm French, obviously. So when just in case you didn't knew it, he's French. I'm not way
it's like, this is the French invasion.
But when I when I arrived here, I was extremely, as a as an engineer, I was blown away by technologies, the power of dreaming about technology, not product technology, about becoming a be having a startup of a building, like the best robot server or the with the best technology over and and
and then what that that's the question I never asked myself. So then for them for what and and the more I saw, I worked at Google I worked at Uber I worked in different companies and and the more the more I worked on these difficult topic The more I I realized that technology by yourself comms like delivering technology to to market comes at a price and comes with a responsibility. And for instance, for self driving car, there is something that that's really annoys me, which is all these companies are competing to get the best self driving stuck ever. But but the ultimate question is that if there is really one stock much safer than all the other one, wouldn't all the self driving car after some stuck, why would i would say why would anybody be mandatory to all of this and stuck instead of like having dangerous tax on the, on the market. And, and I think this is this is a reason reason why I joined Nvidia. Because Nvidia is a platform provider, we provide platform and building blocks for other to read on put on top of it. So every every technologies that we develop it, the goal is not to capitalize on it, and to keep it for our self is really to enable those are companies to build on top of it. And for some very complex stuff. Because as robotics, I don't think there is really an option and I think he would be actually pretty sad if only one company in the market would have all the knowledge all the power and would not have would not he would not let any third party entity to have a look at the technology and to criticize and to make sure that the safety is in accordance with our economy or with with aware I mean, basically, we would have a third party able to look at it. So I think it's very important that we democratize as much as possible
this technology, not because it's cool, because because we need multiple actors to make the right decisions, a single entity cannot make the decision alone. And
on the democratization me. That segues into I think, an important question, obviously, automation, in many cases, cost jobs, whether it's hamburger flippers or car drivers, or pick and pack in an Amazon store, you know, is there gonna be another kind of performance apps are political dimension and going on here, you know, regulations or challenges as you sort of build these technologies? Yeah,
look, I think the robots destroy occupations, not jobs, I think there's plenty of opportunity for automation to continue to create wealth does tend to concentrate it. So we have to do you think about that. But, you know, I think there are extraordinary opportunities for agriculture, extraordinary opportunities for manufacturing and the future retail that don't have the humans to do the work. And I think there's an opportunity for robots default to fill that gap. And I'm not concerned about 90%, the jobs are going away in two years, because it's nonsense. And I think that transition will be more measured. But we have to be concentration of wealth is one of the biggest problems you need to solve. Yes, I think robots
also can improve lots of lives. One of the great things about creator is that the $6 burger, you don't need to work in tech to pay $6 for hamburger.
Mm hmm. You know, I think that's actually just just to give up on that when you look at the end burger robot, or actually, any robot, you I think you have to see it as
more of a something that creates jobs. It's a it's a company on its own, in a way. And you could imagine that, you know, we've all the efforts are made on the on the pricing as
you don't take technology gets gets cheaper anyway. And faster and better, the cost for building those machines is becoming just just relevance, which means that everybody could online or at least could release one and they can be very powerful, I think like we are, we've invested in a company called avid bots, which does a industrial cleaning for airports, and for shopping malls and whatnot in today, they are relying on the networks of distributors. And you can imagine that this thing just rolls around and clean stuff. And so you know, you and I, tomorrow could just just buy one and create your own job this way, and create a new market and etc. And you can think about this with every single robot essentially, is just a business on wheels. Sure. Well, we're almost out of time. So a quick classic robotics panel lightning round, how long until we see autonomous cars on the road, and let's say, let's say 10 to 15% on a majority. But let's say, you know, a decent size of the number of cars you see on the road would be level five, sort of fully autonomous cars. 30 to 50 years, 30 to 50 years, Helen
predictions, I'll say, 10, and you will be in China and you will be a top down decision
will be a Belt and Road. Yeah, pollution entirely out of China. Exactly. Clear. All right.
Do you mean you mean self driving cars on the street passenger cars with us? Yeah. All right. I I agree with serial 10 in China, in China
in the US. Yeah.
All right, everyone. Thank you so much for joining the panel. I hope you had a good time.