Derek Kerton, Autotech Council | Automotive Experts Discuss Data

Hey welcome back everybody, Jeff Frick
here with The Cube. We’re in Milpitas at an interesting event, it’s called the
Auto Tech Council Innovation in Motion Mapping and Navigation event. So a lot
of talk about autonomous vehicles. There’s a lot of elements to autonomous
vehicles. This is just one small piece of it; it’s about mapping and navigation, and
we’re excited to have with us our first guest, to give us a background of
this whole situation, it’s Derek Kerton, and he’s the founder and chairman of the
Auto Tech Council. So first off Derek, welcome. Thank you very much, good to be
here. Absolutely. So for the folks that aren’t familiar, what is the Auto Tech
Council? Auto Tech Council is a sort of a club based in Silicon Valley, where we
have gathered together some of the industry’s largest OEMs, and OEMs mean car
makers you know of, like Ford or Toyota, or Renault from France and the variety of
other ones. They have offices here in Silicon Valley, and their job is to
find innovation. Find that Silicon Valley spark, and take it back and get
into cars eventually, and so what we are able to do is gather them up, put them in
a club, and route a whole bunch of Silicon Valley startups and startups
from other places too in front of them in a sort of parade, and say, these are some
of the interesting technologies of the month. So did they reach out for you? Did
you see an opportunity? Because obviously they’ve all got the the Innovation
Centers here. We were at the Ford launch of their innovation center. You see the
tagline’s all around Palo Alto, and up and down the
peninsula. So you know, they’re all here. So was this something that they really
needed an assist with? Was that an opportunity you saw, or was it- did it come
from more the technology side, to say we need
an avenue in to go talk to Raj at Ford. Well it’s certainly true that they came
on their own. So they spotted Silicon Valley, said this is now relevant to us.
Where historically we were able to do our own R&D, build our stuff in Detroit,
or in Japan, or whatever the case is. All of a sudden these Silicon Valley
technologies are increasingly relevant to us, and in fact disruptive to us. We
better get our finger on that pulse and they came here on their own. At the time
we were already running something called the Telecom Council in Silicon Valley, where
we’re doing a similar thing for phone companies here. So we had a structure in
place that we needed to translate that into the automotive industry, and meet
all those guys and say listen, we can help you. We’re going to be a great tool
in your tool kit to work the Valley. Okay and then specifically what types of
activities do you do with them to execute division? You know it’s
interesting, when we launched this about five years ago,
we’re thinking well we have telecommunication background, we don’t
have the automotive skills, but we have the organizational skills.
What turned out to be the case is there not coming here, the car makers and the
tier 1 vendors that sell to them, they’re not coming here to study break pad
material science and things like that. They’re coming to Silicon Valley to find
the same stuff the phone companies did years ago. Silicon Valley stuff. You
know, how does Facebook work in a car?How do all these sensors that we have in
phones relate to automotive industry? Accelerometers are now much cheaper
because of reaching economies of scale in phones. So how do we use those more
effectively? Hey GPS is you know, reached scale economies. How do we put more GPS
in cars? How do we drive mapping solutions? All these things sound very familiar from that smartphone industry, and in fact
the thing that disrupts them, the thing that they’re here for, that brought them
here and out of out of defensive need to be here, is the fact that the smartphone
itself was that disruptive factor inside the car. Right, right. So you have events
like today; so give us a little story what’s it today- today’s event is called the
Mapping and Navigation event. What are people that who are not here, what’s
what’s happening? Well so every now and then we pick a theme that’s really
relevant or interesting. So today is mapping and navigation. Actually
specifically, today is high definition mapping and sensors, and so there’s been
a battle in the automotive industry for the autonomous driving space. Hey what
will control an autonomous car? Will it be using a map that’s stored in
memory onboard the car? It knows what the world looked like when they mapped it
six months ago say, and it follows along a pre-programmed route inside of that
world, a 3d model world. Or is it a car more like what Tesla’s currently
doing, where it has a range of sensors on it, and the sensors don’t know anything
about the world around the corner; they only know what they’re sensing right
around them, and they drive within that environment. So there’s two competing
ways of modeling a 3D world around an autonomous car, and I think you know,
there was a battle looking backwards which one is going to win, and I think
the industry has come to terms with the fact the answer is both. More of everything,
and so today we’re talking about both and how to fuse those two and make
better self-driving vehicles. So for the outsider looking in right, I’m sure they
go, “Wait the mapping wars are over!” You know Google Maps, what else is there?
Righ, but then I see we’ve got TomTom and I mean a bunch of names that we’ve
seen kind of pre-Google Maps and you know shame on me, I said the same
thing when Google came out with a search engine. I’m like, search engine wars are over.
Who’s Google? Shows you what I know.. Right, is interesting. There’s a lot of
different angles to this beyond just the Google Map that you get on your phone.
What what do you think, MapQuest? What are you doing, have you moved on from a MapQuest? You print it out;
your good. Well that’s the.. Well my wife still prints them out. Yeah there’s people
printing them out somewhere; burning through paper. Listen, the the upshot is that Map Quest is an interesting starting point. I mean first it’s these
maps, folding maps, we have in our car. That’s the best thing we have. Then we
moved to Map Quest era and $5,000 Sat Navs in some cars, and then you
jumped forward to where Google had kind of dominated. They offered it for free,
kicked you know- that was the disruptive factor. One of the things where people
use their smartphones in the car instead of paying $5,000 for that car’s sat-nav, and
that was a long-running error that we have in very recent memory, but the fact
of the matter is when you talk about self-driving cars or autonomous vehicles,
now you need a much higher level of detail than “TURN RIGHT in 400 feet right.”
that’s that’s great for a human who’s driving the car, but for a computer
driving the car, you need to know turn right in 400.005 five feet, and adjust
one quarter inch to the left please. So the level of detail requirement is much higher,
and so companies like TomTom, like a variety of them that are making more
high-level Maps. Nokia’s formed a company called Here; doing a good job. And now
lots of car makers, lots of startups, and there’s crowdsource mapping out
there as well, and the idea is how do we get incredibly granular high detail maps
that we can push into a car so that it has that reference of a 3D world that is
extremely accurate and then the next problem is oh. How do we keep those
things up to date? Because when we mapped- when a car from Nokia
Here, Here is the company now, drives down the street. Does a very high-level
resolution map with all the equipment you see on some of these cars, except for
there was a construction zone when they mapped it, and the construction zone is
now gone. How do we update these things? So these are very important questions if you want
to have the answers correct and in the car stored well for that car to self
drive, and once again, we get back something mentioned just two minutes ago.
The answer is sensor fusion. It’s a map- it’s a mix of high-level maps you’ve got
in the car, and what the sensors are telling you in real-time. So the sensors
are now being used for what’s going on right now,
and the maps are ‘give me a high level of detail from six months ago when this
road was driven.’ It’s interesting, back in the day right, when you had to have the
CD for your on-board mapping. We had to keep that thing updated and you could
actually get to the edge of the CD. It didn’t work anymore. Yeah, the world is flat.
The other thing they are covering here too which feeds into this, is kind of the
whole optical sensors, because kind of LIDAR school of thought,
and then there’s the the biopic cameras school of thought, and again the answers probably
both, right? Well good. That’s a you know, that’s- there’s all these neat little
battles shaping up in the industry, and that’s one of them for sure, which is
LIDAR versus everything else. LIDAR is the gold standard for building, I keep
saying a 3D model, and that’s basically you know, a computer sees the world
differently than your eye. Your eye looks out a window we build a 3D model of what
we’re looking at; how does computer do it? So there’s a variety of ways you can do
it. One is using LIDAR sensors which spin
around. Biggest company in the space is called Velodyne. Been doing it for
years for defense and aviation. It’s been around pointing lasers and
waiting for the signal to come back. So you’re basically using reflected signal
back, and the time difference it takes to get those back, it builds a 3D model of the
objects around that particular sensor array. That is the gold standard for precision.
The problem is, it’s also bloody expensive. So the carmakers said that is
really nice, but I can’t put four $8,000 sensors on each corner of a car, and get
it to market at some price that a consumers willing to pay. Until every
car has one and then you get the mobile phone effect. Yeah but economies of scale at
$8,000, we’re looking at it going, “That’s a little tough.” So there’s a lot of
startups now saying now listen. We’ve got a new version of LIDAR that’s solid-state.
It’s not a spinning thing. It’s actually a silicon chip with MEMS and
stuff on it, that are doing this without the moving parts, and we can drop the
price down to $200 maybe $100 in the future in scale. That starts being
interesting that’s 400 dollars if you put it on all four corners to the car, but
there’s also other people saying listen, cameras are cheap and readily
available. So you look at a company like NVIDIA that has very fast GPUs, saying
listen, our GPUs are able to suck in data from up to 12 cameras at a time, and with
those different stereoscopic views, with different angle views, we can build a 3D
model from cheap cameras. So there’s competing ideas on how you build a model
of the world, and then there’s companies like Bosh, saying well we’re strong in
car and radar, and we can actually refine our radar more and more,
and get 3D models radar. It’s not the good resolution that lidar has, which is
a laser sensor. So there’s all these different sensors, and I think
there the answer is not all of them, because cost comes into play. So a
car maker has to choose. Well we’re going to use cameras and radar, or we’re gonna use
LIDAR and high-definition maps. So they’re going to pick from all these
different things that are used to build high-definition 3D model
of the world around the car. Cost effective and successful, and robust. Can
handle a few of the sensors being covered by snow hopefully, and still
provide a good idea of the world around them, and safely. And so they’re going to
fuse these together and then let their their autonomous driving intelligence
ride on top of that 3D model and drive the car. Right so it’s interesting, you
brought up NVIDIA and what’s really fun I think about the autonomous vehicle, and
self-driving cars and the advances, is it really plays off the kind of Moore’s
Law’s impact on the three tillers of its compute. Massive compute power to
take the data from these sensors. Massive amounts of data; whether it’s in the
pre-programmed map, whether you’re pulling it off the sensors, you’re
pulling off a GPS. Lord knows where. Wi-Fi, waypoints. I’m sure they’re
pulling all kinds of stuff, and then of course you know, storage. You’ve got to
put that stuff, the networking you know gotta worry about latency. Is it on the edge, is
it not on the edge. So there’s really an interesting combination of technologies
all bring to bear on how successful your car navigates that exit ramp. You’re
spot-on, and that’s you’re absolutely right, and that’s one of the reasons I’m
really bullish on self-driving cars. A lot more than in the general industry
analyst is, and you mentioned Moore’s Law and NVIDIA’s taking advantage of that
with their GPUs. So lets wrap everything you said into kind of big answer. Big
data and more and more data? Yes, that’s a huge factor in cars. Not only are cars
going to take advantage of more and more data; high definition maps are way more
data than the MapQuest Maps we printed out. So that’s a massive amount of data
that car needs to use, but then in the flipside, the cars producing massive
amounts of data. I just talked about a whole range of sensors. I talked LIDAR,
radar, cameras ,etc. That’s producing data, and then there’s all the
telemetries of that data. How’s the car running, how’s the engine performing. All those
things. Car makers want that data. So there’s massive amounts of data needing
to flow both ways. Now you can do that at night over Wi-Fi
cheaply, you can do it over an LTE, and we’re looking at 5G radio standards.
Being able to enable more transfer of data between the cars and the cloud. So
that’s incredibly important. Cloud data and then cloud analytics on top of that. Okay now
that we’ve got all this data from the car,
what do we do with it? We know for example that Tesla uses that data sucked
out of cars to do their fleet- their fleet learning. So instead of
teaching the cars how to drive, if I’m a programmer saying if you see this, do
that; they’re taking the information out of
the cars and saying what are the situation these cars are seeing. How did
our autonomous circuitry suggest the car responds, and how did the user override
or control the car in that point, and then they can compare human driving with
their algorithms, and tweak their algorithms based on all that fleet
driving. So it’s a master advantage in sucking data out of cars; massive advantage
pushing data to cars, and you know, we’re here at Kingston SanDisk right now today,
so storage is interesting as well. Storage in the car increasingly
important for these big amounts of data, and fast storage as well.
High Definition maps are beefy beefy maps. So what do you do? Do you have that
in the cloud and constantly stream it down to the car? What if you drive
through a tunnel or you go out of cellular signal? So it makes sense to
have that map data at least for the region you’re in, stored locally on the
car, and easily retrievable in flash memory. That’s dropping in price as well. All
right so the last thing about that, that was a loaded question by the way,
and I love it, and this is the thing I love- this is why I’m bullish and more
crazier than anybody else about the self-driving car space.
You mentioned Moore’s Law I find Moore’s Law exciting. Used to not be relevant to
the automotive industry. They used to build- we talked about
briefly about brake pad technology material science. Like what kind of
asbestos do we use, and how do we I would dissipate the heat more quickly? That’s
science, physics, important R&D. Does not take advantage of Moore’s Law. So cars
been moving along with laws of thermodynamics. Getting more miles per
gallon. Great stuff out of Detroit, out of Tokyo, out of Europe, out of Munich, but
Moore’s Law not entirely relevant. All of a sudden,
since very recently, Moore’s law starting to apply to cars. So they’ve always had
ECU computers, but they’re getting more compute put in the car. Tesla has the
NVIDIA processors built into the car. Many cars having stronger central
compute systems put in. Okay so all of a sudden now, Moore’s Law is making cars
more able to do things that they we need them to do. We’re talking about
autonomous vehicles. Couldn’t happen without a huge central processing inside
of cars. So Moore’s Law? Applying now what it did before. So cars will move quicker
than we thought. Next important point is that there’s other
exponential laws in technology. If people look up these other cool things, Kryder’s
Law. So Kryder’s Law is a law about storage, and the rapidly expanding
performance of storage. So for every dollar spent, how many megabytes or gigabytes of
storage do you get? Well guess what. Turns out that’s also exponential, and your
question talked about isn’t Big Data important? Sure
it is. That’s why we could put so much into the cloud, and so much locally into
the car. Huge, Kryder’s law. Next one is Metcalfe’s
law. Metcalfe’s law is law of networking, and it states, basically in it’s roughest
form, the value of a network is valued to the square of the number of nodes in the
network. So if I connect my car, great. That’s that’s awesome but who does
it talk to? Nobody. You connect your car, now we can have two cars you can talk
together and provide some amount of element of car to car communications and
some some safety elements. Tell me the network is now connected; I have a smart
city? All of a sudden the value keeps shooting up and up and up. So all of
these things are exponential factors and they’re all of a sudden at play in the
automotive industry. So anybody who looks back in the past and says well, you know,
the pace of innovation here has been pretty steep. It’s been like this. I
expect in the future it will carry on, and in 10 years, we’ll have self-driving
cars. You can’t look back at the slope of the curve, and think that’s a slope
going forward. Especially with these exponential laws at play. So the slope
ahead is distinctly steeper. And you left out my favorite law
which is a Amara’s law, which is you know, we underestimate in the short term- or
overestimate the short term, and underestimate the long term. That’s all
about that slow. Yeah, all about the slope. So Derek, we could go on for probably like an hour and a half. I know I could. But you gotta
go. You got to go into your event. So thanks for taking minute out of your
busy day. Really enjoyed the conversation, and look forward to our next one. My
pleasure, thanks. All right, Jeff Frick here with The Cube. We’re at the Western Digital
headquarters in Milpitas at the Auto Tech Council Innovation in Motion
Mapping and Navigation event. Thanks for watching.