And welcome back to the roadshow live stage of the 2017 Detroit Auto Show.
My name is Tim Stevens.
I'm editor-in-chief of Roadshow and I'm here with Danny Shapiro, senior director of automotive at Nvidia.
You and I both doing the double CES then straight up here to Detroit.
How are you feeling?
Great we've had so much fun at CS in Las Vegas now here in Detroit it's really a pleasure to be here.
And the Nvidia's been pretty progressive in terms of adding more digital horsepower.
I keep saying the car is giving more processing power for these autonomous systems.
But I want to talk about the demonstration That you guys were giving at CES first, really giving the opportunity for a lot of people to get inside an autonomous car for the first time and sit in the back seat with nobody in the driver's seat and have that car drive them around.
What were some of the reactions you got from that demonstration [CROSSTALK]?
So basically we took one of our test vehicles, we outfitted with our AI super computer.
We call it Drive PX2.
It takes sensor inputs.
It basically makes sense of what's going on around the car, and then we control the vehicle.
So, we modified a vehicle, brought it out to Las Vegas, and we set up a course.
We recognized there's a lot of different conditions that drivers face, and so we built a course that had lane markings for part of it.
Some areas where no lane markings We had a dirt section.
There's areas that it was grass and it was landscaping features on the side like somebody might have on their driveway.
And then we also set up cones and construction detours and we'd roll a flashing construction airway and roll it out.
And so we wanted to show the power of deep learning and the ability to use artificial intelligence to train the vehicle.
Essentially, we just had four days before the show to train it.
The cars showed up in Vegas.
It didn't know anything other than had the base layer of software.
But we trained it.
We didn't program it.
So we put people in the back seat, you're absolutely right.
And we were then, we had a remote switch to start the car.
And it's driving and people were amazed, absolutely.
It's an incredible experience, right, looking at The steering wheel of a car and watching it drive, and there's nobody there.
I've been looking at a few cars that can do that.
I don't wanna say it's old hat, but I'm getting used to it.
But definitely, that first experience, the first time you let a car take you somewhere, and the first time you Really put your face in this technology.
Obviously that was a controlled course.
But definitely that's opening a lot of people's eyes to think, and a lot of people are skeptical of this technology.
So demonstrations like that are great.
So I'm glad to see more people getting in.
And we were talking before about how more and more people are having that opportunity and ultimately I think it really is About getting people the chance to try that out, right?
To change their minds?
And so, deep learning is important.
AI is a very, very important thing.
Learning is an important thing.
And we're reading about so many millions of miles being covered by these autonomous cars to figure out different situations and that kind of thing.
But ultimately, these cars have got to get better about how they learn.
And that's part of this process as well, right?
Not just throwing more processing power And not destroying millions of miles, changing the way they learn and learn from us right?
Absolutely, what we're doing of course is a car.
Safety is first and foremost.
If something goes wrong, there can be a bad outcome.
So our focus is absolutely on safety and using artificial intelligence to drive Much better than a human could.
I think there's a combination of a car learning from humans.
But it's not like learning for you and I. It's not like my 14 year old daughter who is going to get her permit soon and it's not going to learn from her.
But instead, it's learning from professional drivers and also we're also doing simulation in a lab.
So we can train a car.
How to drive and how to react appropriately to all different types of scenarios and potential hazards.
And from that then we're going to do a massive amount of testing.
And that testing will happen in on-road testing with our fleets of drivers and our customers' drivers.
But we'll also use simulations to test.
So we've Developed an AI system that can recognize people jaywalking or dogs running in the street.
We can run simulation tests to create those scenarios rather than actually putting human life or dogs in harm's way.
And then we can validate that before we put it out on a public road.
I think one thing that was great, just yesterday at the Volvo stand The first key is to the Volvo in the Drive Me program, where I hand it over to a family in Sweden.
And so that program has started, where 100 XC90s, on the road to the public.
They're driven by the Nvidia Drive PX 2. So our AI super computers in those cars Taking all that sensor data and then having those cars drive themselves.
And that is a really exciting announcement.
That's the first time that these autonomous cars have been put in the hands of actual consumers.
I don't think four or five years ago anybody would've thought that by 2017 we would have fully autonomous cars on the road.
Limited circumstances, only in Germany only in Gotham, or excuse me, only in Sweden, only in [INAUDIBLE] But ultimately that's really exciting still.
Yeah, I think the key thing here is this isn't testing the technology.
We're not gonna do that on customers.
But this is testing how they interact.
It's a research project from the sense of observing what does that family do in the car?
How do they Get more time back from normally would be spent on their commutes.
And how do other cars interact with the self-driving car.
I think it's a really valuable and important program that they have developed.
It is also a lot of learning that can be made from the mobile industry, from other technologies, and that kind of thing as well.
Obviously, teaching AI is important, but also learning things as far as Device engineering goes in mobile chips and that sort of thing, which is something that you guys can bring to the table better than anybody else.
Nvidia's been making mobile chip sets for a long time and ultimately graphic card chip sets, too.
But how does that sort of consumer technology then transfer over to the automotive industry.
I think what we're seeing is a compressing of time cycles, to development cycles of
Consumer electronics move much more quickly than the automotive industry.
And I think we're seeing in then automotive industry, not necessarily go on exact same timeline.
But compress that.
I think it's critical, of course, in the automotive space.
That we go through the testing and validation that is so important to the auto industry.
And so I don't think you'll even see these two different industries in lock step.
We go through an incredible amount of validation on the product.
But again I think the key part is we're able to leverage the massive amounts of computing and the concept of the software being updated over the air, to make the product better.
So we're doing this with Tesla, we're doing this for our other customers so that Once you purchase the car, the car continues to get better and better and better over time through a software update.
And that's really not something that consumers are used to.
You're lucky if your car gets an update once a year at this point, or I don't know if I should say lucky, maybe lucky is the wrong word.
But it's rare that you get more than one update a year.
But as you were talking about cars getting smarter and also more secure, that's got to be something that happens more often.
And that's something that I think the automotive industry could learn a lot from the [INAUDIBLE] industry.
I think the security thing
Point that you raise is key.
And many cases the cars that are on the road today were not the computing systems were not architected with security in mind.
They were incremental approach where more and more start getting added to the cars and now all of a sudden now they're connected.
What we focus on is building a secure system from the start and then building a car around it.
And so encryption And authentication and other computing techniques that are used in the data center to ensure the data's secure by what we focused on.
So let's talk about the actual hardware of this system.
I've been in a lot of prototype autonomous cars and you pop the trunk and it's just a mess in there.
There's computers everywhere.
But how discrete and how compact are we talking about getting these systems down to?
Our current drive PX2, it basically has the power of 150 of your Mac Books here.
It's about the size of a license plate today, and this is what over 80 different automakers and tier one suppliers and other startups are using in their self-driving car efforts.
That was what was in our test vehicles.
We also put one in an Audi 27.
And we were doing similar tests in Las Vegas.
We just introduced our next generation that will come out at the end of 2017.
And so we're shrinking down from a license plate to a module a little bigger than a deck of cards.
And what do you have to worry about as far as cooling goes?
Because obviously, the deployment in a car is very different than the deployment Here on a laptop.
We're talking about circumstances where you might be up in northern Alaska, let's say, very, very cold.
Or in a very hot desert environment.
It's gotta work across this range-
Of temperatures, how does that effect?
Everything we do, it needs to be automotive grade.
The processors we put in PCs or we put in mobile devices
Are consumer grade.
So we have an entirely new developed process over the last decade to make them auto grade.
To ensure that at those freezing temperatures in Alaska to summers in Arizona, your car will start and everything will power up.
I think we Probably experienced situations maybe even here in Detroit.
If you're trying to use your phone outside eventually after a couple minutes it's just gonna shut down because it's too cold.
We ensure that our products are auto grade from a temperature perspective, dust, shock, vibration.
All of these.
Different characteristics go into building our solutions for drive techs computers.
And a lot of gamers are very familiar with the idea of upgrading their graphics chips, especially as we're moving to a VR world and a lot of people are their old GSX cards and popping in new ones.
Is that something we might have received in the automotive world?
Might I some day pop the trunk in my car and pull out the autonomous brain in my car and pop in a new one to add new functionality?
It's possible, I think that the notion of having modular components in the car is nothing new.
And many of our customers, Audi and the Volkswagen group and others have a whole modular design Whether you do it yourself, whether you do it at the dealership, I think those are all potential options.
I think at the same time when we're talking about entertainment systems one thing,-
We're talking about the self-driving car, human lives are at stake here.
And the testing, and validation I think as we talked about is critical So I'm not sure that we would want to be messing around with the electronics of a self-driving [CROSSTALK].
I think you could put stickers on the side to identify what kind of the graphics horsepower there is in your car?
That would be pretty sweet.
I think there's a new marketing opportunity for you, Danny.
So maybe you can take that back with you and we'll see how it goes.
Danny, thanks so much.
We had a panel earlier this week as well, which it was great to speak with you then.
Awesome updates from Nvidia, and we're really excited to see how this technology comes to fruition and comes to the market.
It's great to see