Credit: AI Trends
Five years ago at the Code Conference, self-driving cars seemed as though they were just around the corner: Google unveiled the project that would later become Waymo, and Uber’s then-CEO Travis Kalanick stirred controversy when he talked about the benefits of replacing human drivers. But in 2019, autonomous vehicle prototypes are a rarity in most cities outside of San Francisco, and humans are still vital to companies like Uber and its first-to-IPO rival Lyft.
That’s because self-driving is a really, really hard technological problem, Ford CTO Ken Washington said on the latest episode of Recode Decode with Kara Swisher. But, very slowly, beginning in 2021, you’re going to start seeing cars with no one in the driver’s seat.
“You may see some earlier ones in 2020, but we believe in taking the time to work with the cities,” Washington said. “If you just put a bunch of autonomous vehicles in the city without designing it to make life better in that city, you’re gonna have an analogous problem to what happened when Ubers first started showing up. People hated them because they’re camping out on the corners, and it made congestion worse, it created additional pollution.”
Ford is currently testing its self-driving cars (still with humans in the front seat as a precaution) in Miami, Washington, Dearborn, Pittsburgh, and multiple places in California. Washington explained that, in order to be ready for regular consumers, these “robo-cars” need to have a pre-existing 3-D scan of every street they might drive on.
“This is not your navigation map, the kind of map that you would use on your cellphone that you pull out and you do a Google Map or an Apple Map,” he said. “It’s actually shooting light beams out in three dimensions off of the roof of the vehicle … [and] capturing these points and creating a 3-D image of what the world looks like.”
“If you don’t have that part of the map, you’re relying on, in real-time, detecting everything that might happen, and that’s just too hard of a problem,” Washington added, before taking a dig at Tesla’s so-called Autopilot features. “That’s why these vehicles that don’t have LIDAR, that don’t have advanced radar, that haven’t captured a 3-D map, are not self-driving vehicles. Let me just really emphasize that. They’re consumer vehicles with really good driver-assist technology.”
Q&A interview follows:
Kara Swisher: How are you doing?
Ken Washington: I’m great.
Now Ken, you’re gonna tell us how cool all this stuff is now that we talked about the disaster that’s coming. Let’s talk a little bit about … I wrote a column last week, which did get a crazy amount of commentary. Thousands and thousands of comments on the New York Times. People were for it, people were really, really for it, and what I did in the column is that I really don’t want to own a car again. I wrote a piece in the Wall Street Journal 25 years ago saying, “You will have mobile phones. You will not have landlines. You will not be wired. It will be all be wireless.” It was a very good prediction, and then I said, “Now, you’re never gonna own a car,” and, “It will be as quaint as owning a horse.” I think that’s the expression I used.
I was trying to get a discussion going. I mean, obviously you wanna talk about where this is going, but I do truly believe that we’re on the cusp of this because of self-driving and AI and the stuff that are gonna go into transportation. So let’s talk a little bit about cars first, and then we’ll get into the other things that companies like Ford and others are doing, where AI does benefit a company. So let’s talk about where we are with autonomous vehicles right now and where AI fits in.
Well first, I just wanna say, I think you definitely got people’s attention with the article.
I think the response was a reflection of the fact that people love cars, and some people hate cars, but everyone needs to move around. What autonomous vehicles has done for us is given us the potential and the promise of a new way of moving around, and a new way of creating mobility, and a new way of solving real pain points in cities. What I loved about your article was it really shined a light on the fact that in urban centers where you really need a different model, that now this new model is beginning to emerge.
Right. I wasn’t talking about car ownership. I was talking about — and not — I was talking about car ownership, not car driving. I will continue to drive and move in mobile vehicles depending on … but it was the car ownership and it’s the idea of ownership. Just the way we owned entertainment before, now we really don’t. Just the way we owned records, we owned this. It’s the same concept, that this is one that’s moving in, and especially in urban areas where I think 90 percent of the population’s going to be in a megalopolis over the next 25 years, like 90 percent. So those other 10 percent can have their cars, it’s fine, but it’s just what happens when those population …
So talk about where we are and how AI fits into the idea of where we are with autonomous. Let’s start with autonomous vehicles. We’re right in several stages, right? Three, four, five, we’re in three now, which is?
Well, let me just clarify that.
It’s a term that’s often really misused and misunderstood. When you’re talking about autonomous vehicles, particularly in urban cities where you predict 90 percent of people are gonna live. What you really have to think about is an autonomous vehicle that can truly sense the environment completely and can truly take the human out of the loop. So that’s a level-four autonomous vehicle.
Even that level-four autonomous vehicle that can operate in this urban environment, it’s gonna have some boundaries around it. It’s gonna have to have the ability to operate in a city that it’s seen before, so it has to have been mapped. There are gonna be weather restrictions on it, at least with today’s technology, and there are gonna speed restrictions, because the sensors aren’t perfect and hopefully no one in this room believes that the AI part of the problem has been completely solved. You can’t even review résumés with AI and not screw it up.
Getting the AI right in a vehicle is really hard. It involves a lot of testing and a lot of validation, a lot of data gathering, but most importantly, the AI that we put into our autonomous vehicles is not just machine learning. You don’t just throw a bunch of data and teach a deep neural network how to drive. You build a lot of sophisticated algorithms around that machine learning, and then you also do a lot of complex integration with the vehicle itself.
Read the source article in recode.