with Abe GhabraMore Info
Tom goes to Las Vegas to talk with Abe Ghabra, Managing Director at autonomous vehicle firm Aptiv, and ride in a self-driving Lyft. Safety First for Automated Driving White Paper: https://www.aptiv.com/docs/default-source/white-papers/safety-first-for-automated-driving-aptiv-white-paper.pdf One-Year Las Vegas milestone: https://www.aptiv.com/media/article/one-year-in-aptiv-marks-50-000-public-robotaxi-rides Delphi Research Report "Transportation 2050" http://delphigroup.com/wp-content/uploads/2018/06/Transportation-2050-Delphi-Group.pdf
TK: In this episode of Foresight Radio: driverless cars. How the AI-driven revolution in autonomous vehicles would change so much about our urban landscapes, safety, and the way we think about transportation. Foresight Radio is brought to you by our good friends at Wasabi. Find out how Wasabi is changing the rules of the games when it comes to cloud data storage at Wasabi.com. In my last book, Revealing the Invisible, I wrote a great deal about the progress made in the driverless car industry. One of those companies that I profiled was nuTonomy. nuTonomy was recently purchased by a company called Aptiv. Aptiv was formally known as Delphi Automotive. I know this sounds really complex, but the bottom line here is that a lot of players are converging to build this driverless platform. To find out more about it, I met with Abe Ghabra, their Managing Director of the Las Vegas Customer Technology Center. Aptiv launched a fleet of about 30 self-driving cars that, in conjunction with Lyft, are providing rides to Lyft users in the Las Vegas area. Now when we talk about driverless, the first thing that's important to understand is that there are actually five levels that are used to categorize a degree to which a car is driverless. At the first level, there's no automation whatsoever. At the fifth level, the passengers become basically cargo. When I got into the BMW 5 Series that Aptiv is using as its driverless car in Las Vegas, Abe told me a bit about the capability of this particular automobile. Abe: Level four is driver out of the loop. Once it's engaged, which just happened, he will do nothing. TK: Once we hear that tone, that means the car is taking over. Abe: The car just made a decision to move us to the left turn because we told the car that we want to go to the Las Vegas welcome sign, which is all the way to the other side of the Strip. All David did was enter the destination before we picked you up. The moment he engages in autonomous mode, he's doing nothing. He's not signaling. He's not touching the steering wheel, not applying breaks or throttle. TK: In case you missed that, our driver's name was Dave. Naturally, I wanted to ask Dave a question about what it felt like to drive a driverless car, but as you'll see, that was not the right thing to do. Even in a driverless car, the driver plays a pretty important role at this point. Dave, as you drive the car what did you find to be most surprising in your experience? Abe: Sorry. I should have mentioned that at the start. David is not allowed speak with you. David's entire focus is on the road. TK: Every driverless car being used by Aptiv today in its trial with Lyft actually has two drivers. They call them co-pilots. In this case, Dave was doing the actual driving, meaning he was sitting behind the steering wheel in case he had to take control what's called a hand off. Evan was the other co-pilot, so I turned to Evan to ask him a bit about what it felt like when he was in the driver's seat. Can we even call it the driver's seat anymore? Evan, can I ask you that same question? What has it been about driving the vehicle or co-piloting the vehicle that surprised you? Evan: We've been over 50,000 rides with Lyft since last May, and the amount that they’ve come and become more humanlike since then. TK: There's a term you hear used frequently when you're in a driverless car, this notion of it being humanlike. Until you've actually been in one, it's difficult to characterize what that means. In general terms, it means the car exhibits a behavior. It's the kind of behavior you would expect from a well-qualified driver. However, that behavior does change from culture to culture, geography to geography. In other words, I would not drive the same way in Boston than I would drive in Athens or in Rome. You adopt certain behaviors based on the contextual understanding you have of how a driver should behave in that situation. Driverless cars aren't quite there yet, but this notion of it being humanlike is one of the things I hear every time I speak to anyone who has been involved with driverless cars for some period of time. They have the ability to learn and they express that learning in the same way that a human being would in terms of behaviors that are sometimes unexpected, but typically, which you will look back on and say, "That's exactly what the car or a driver should have done in that situation." Evan: One of the things our safety drivers as far as like operating the vehicle is you're paying attention even more than if you're driving a normal car because you have to be vigilant of the other human element of all the other drivers out there. TK: We just heard that quick beep there. What was that telling us? Abe: It's basically telling us there's a crosswalk with engaged pedestrians in it, so this is a queue for our safety operator to keep an eye on the road, especially in a busy crosswalk area. TK: When I got into the BMW driverless car, I pretty much knew what to expect having done this before. However, I was really curious as to how the typical Lyft passenger would react to getting into a car that was in fact in some ways driving itself. Abe: It’s been interesting here. As I’ve mentioned, we've been operating here for over a year now – actually, over a year and a half if you include CES when we first debut these cars and did a pilot with Lyft. You see some people that are inquisitive, that they are curious. Does this car really drive itself? Passengers are usually delighted. The main feedback we get is - it feels like a safe professional driver is operating the car, which was our entire goal. Our Lyft passenger rating speaks for itself. It's a near perfect 4.95 out of five star average over those 50,000 rides. TK: Of course, this is all happening in Las Vegas where the tourists are not just from around the world, but are not always in the best frame of mind. Abe: It's interesting because obviously, Vegas tourists are not always in the sharpest state of mind. [Laughter] TK: Good point. Abe: The Lyft app actually not only ask you to opt in one time before you start using the service and you have to agree to the terms and conditions, but when you do hail that specific ride, you have to pick the driverless or the self-driving car option. You have to go through several motions throughout the app experience in order to end up with a self-driving car, and yes. [Laughter] We have people who get in the car and they're completely surprised by the fact that they are in a self-driving car. TK: Have you ever had someone get in the car and decided to get back out again because they're just too freaked out by it? Abe: Not that was explicitly made available to us by Lyft, so that kind of thing, unless I canvass all of our drivers and ask that specific question, I wouldn't be able to know that because it wouldn't register as a ride. TK: Interestingly enough, this is not the first time they’ve gone driverless. In fact, during the mid-1900s, elevators were driven by operators but the word driverless elevator is as far back as the early 1900s. There was just one problem. Nobody would use them. Given the choice between the stairs and a lonely automated elevator, the elevator would remain empty. It was until the middle of the 20th century that the tipping point finally came along for the driverless elevator. In 1945, the Elevator Operators' Union in New York City called a strike, which pretty much shut down every elevator in the entire city, effectively shutting down the city itself. Enraged renters and landlords demanded action. The strike was devastating to their economy and it costs an estimated $100 million to the city. Suddenly, there was an economic incentive to go back to the automatic driverless elevator. Over the next decade, there was a massive effort to build trust in what were called automatic elevators, which resulted in the elimination of tens of thousands of elevator operators/drivers jobs. A few of us today will step in on an elevator and even casually think about the way it operates, how safe it is or what the risks are. Now granted, autonomous is a world apart from automatic, but the fundamental issue with the adoption of a driverless Lyft in either case is not so much the technology. Elevators already have the technology to be driverless for 50 years. It's about trusting a machine to do something as well as a human. In a word, it's about perception. If you're still doubtful, perhaps you're one of the few people who have a fear of elevators, for example. Well, you may be justifying your fear. Twenty-seven people die yearly as a result of faulty automatic elevators. However, you might also be interested in learning that according to the Center for Disease Control's National Center for Health Statistics, a whopping 1,600 die from falling downstairs every year. I'll save you the math. That means you're 60 times as likely to have a fatal accident taking the stairs. Unfortunately, numbers alone don't often change perception. What I find interesting about that is it's not just the drivers in the cars that need to adopt a new perspective on what driverless means and how our behaviors adapt to it and whether we perceive it as safe or not, but it's also the pedestrians. If you think about this, depending on where you are, if you’re in New York City or if you're in an urban area that's very congested, when you step off the curve, there's an expectation you have as to what the driver will do and probably that is based on your experience, probably it’s based on what's happening in that moment. I'll admit that there are times when I do cross the street without having the right of way, without having a green light that tells me I can cross it, but I know the driver will stop. Think about that for a minute. How does that change with the driverless car? Do I expect it to be differential to me or am I going to be more careful about it? We saw a bit of this as we were driving down Las Vegas Boulevard. Abe: Two years ago, this car, if it would have seen a pedestrian crossing in the middle of the highway, it would have stopped not knowing what to do. It would wait until the pedestrian cleared the road. As of CES of this year, this car would actually behave just like you or I would because a lot of times, a pedestrian, to be safe, they're in the middle of crossing the road but they're going to wait for you to clear before they go. Now if you stop, you're creating this awkward moment, so we taught the car how to read that kind of situation and if it's safe to keep going and allow the pedestrian to go behind us to do just that. TK: Sure enough, just as Abe finished describing the behavior of the autonomous vehicle with the pedestrian, we had a pedestrian step off the curve and right in front of the car on Las Vegas Boulevard, and it wasn't on a crosswalk. Abe: Here you go. We have a situation where we have the green light but the pedestrian ran right in the middle of Las Vegas Boulevard. TK: The car did slow down. Abe: The car slowed down just like you would expect. TK: There's something else which is interesting in the context of the example of the pedestrian that we just gave, which is this notion of how drivers communicate with each other. We use facial gestures, hand gestures, even body language in many cases to signal our intentions and to read the intentions of other drivers. At some point, autonomous vehicles will have to do the same thing. Abe: We're not doing in this particular car. In fact, this car has some stock gesture recognition controls that allow us to operate the HMI that we can use. It’s available now. We're not using it for the autonomous vehicle system yet. We do have an investment in the company called Affectiva. There's a license into that technology. Also, our advance safety and user experience group, which is a separate business unit with Aptiv, is working on that. TK: As you experience a driverless car, one of the things that you notice is the enormous amounts of data that are being collected. Now Aptiv actually has a visual display that the driver or passenger can look at which shows you a rough approximation of all the various sensors and what it is that they're collecting in the moment. Here's what's critical about that. It is far too easy to hold on to this data as a competitive point of leverage. However, in order for the industry to scale, it's mandatory that to some degree that data be shared across otherwise competitive entities. Now, use the NTSB and air traffic as an example here. If there's an incident with an airliner, that data is shared across all manufacturers and all airlines. The reason is that it creates a much safer experience overall for the passenger and for the airline. The same thing applies to driverless vehicles. Aptiv has taken a leading role here. In fact, they published a safety white paper in conjunction with a variety of other manufacturers of driverless automobiles, including BMW and Audi. In order for the industry to scale and to scale safely, much of the data that they're collecting needs be shared with both academia and with software developers in order to create a platform that is as safe and as reliable as possible. Abe: Given that the two startups I mentioned earlier, Ottomatika and nuTonomy, that formed Aptiv’s autonomous mobility came out of the academic space, it's important to them to give back, especially to the academic community but the public at large when it comes to our research and development. We came out with the first release of new scenes about - I want to say a year, a year and a half ago. Then we came out with the last release a couple of months ago here. I think it was a thousand scenarios capturing LiDAR, RADAR, and vision data from real-life autonomous vehicles that were on public roads. It's important to help advance the technology at large. It's important for people in the academia to have access to that type of technology because it's not easy or cheap for them to do so. Whatever we can do to help the industry as a whole advance, then we're willing to do that. Another recent example of that is the safety white paper that we published alongside 10 other partners. We were one of the five partners to be involved in that and it grew during development to 11. Really, what that aims to do is take advantage of some the work that’s been done on the regulatory side, to bring that to the industry and say, "How do we get ahead? While we're developing this technology, how do we develop a common set of standards when it comes to designing, testing, and operating autonomous vehicles safely, with an emphasis on safety?” TK: One of the reasons that the driverless industry is progressing as fast as it is has to do with the way that driverless cars learn. They don't learn just based on their own individual experiences the way a driver or a human driver, but they learn collectively. Every car's experiences uploaded daily, scrubbed and analyzed, and then download back onto the entire fleet. Any incident experienced by any single car, then becomes part of the collective memory and collective behavior of the entire fleet. As you extrapolate this now to not just individual city fleets, but the country fleets and global fleets, you can imagine the incredible power of that learning and the accelerated speed with which it can happen. Male: Wasabi Hot Cloud Storage is proud sponsor of Foresight Radio and their mission to help rethink the future of how we work, live, and play. At Wasabi, we're helping you take control of the future with your ability to affordably store and leverage data will determine success or failure. Our low cost, high speed, fully secured cloud storage blows away Amazon, Google, or Microsoft. There are no hidden fees for Egress or API requests. See for yourself with the free trial of Wasabi.com. TK: These things generate a lot of data. Abe: They do. TK: On a course of a given day, I know - Abe: Terabytes of data every day. TK: Do you upload that every day? Is it all local storage? Abe: We have a data center right here at our Las Vegas Technical Center. We upload all the data every day. We put it through a development lifecycle. Basically, it's feedback. We're giving back to our developers who are mainly in Singapore, Boston, and Pittsburgh. They look at it. They iterate the software. They test it in simulation. Then they send it to the different sites where we test it on private closed course test sites. We have one at each of our locations. Then once it's tested and it passes the test drive, then we deploy it to our public road fleet. Once it passes that, then it goes into the Lyft vehicles. TK: How often does an iteration of that take? Does it happen in a monthly basis? Abe: Usually about three to four weeks, yes. We usually have about roughly monthly releases to the Lyft software that’s picking up numbers in public. TK: One of the neat things about AI is that as the intelligence is gathered, the learning actually accelerates. Are you noticing anything? Abe: Yes, absolutely. Two years ago, I was giving a demo with a reporter and there was bus in front of us that pulled over to pick up passengers. We got stuck behind the bus for a few minutes while it was picking up the passengers at CES, and then traffic, that's not ideal. This year, same exact scenario, when the bus started slowing down and signaling that it's about to pull over, our vehicle shifted to the middle lane, pass the bus, and got on the right lane immediately without hesitating, which is what I would expect from a professional safe driver to do. Lots of examples like that, but that's been the evolution of the technology both on the software and hardware side, but also on the AI side. TK: One of the things that's especially disconcerting if you ride a lot in Ubers or Lyfts or cabs is that every driver has a different behavior. Abe: Yes, absolutely. TK: Sometimes, it takes a little bit of getting used to, especially New York City, cabbies I found [Laughter] like to use the accelerator like an on/off switch. It can get almost nauseating. Is there a way that you can adjust the ride to suit a certain set of preferences? Abe: I would expect that to be the case in the future. Right now, we're aiming to build just the safest, most experienced driver you could have. TK: Tell me about timeline. Where do you think we are in terms of mass deployment and how would you define mass deployment? What are some of the goals that you're looking at? Abe: Really two definitions for mass deployment. One is with the mobility fleets, so incorporating our technology with transportation providers like Lyft. We're doing that now, obviously with the safety driver but we expect to remove the safety driver in the next couple of years and run the operations through our command center, where we have the ability to do passenger support and teleoperations. We expect to have a driverless robo-taxi service in 2022 for mobility fleets. Now, that's going to scale over the next few decades. We think that by 2030, a quarter of the miles driven will be autonomous. By 2050, we expect that more than 90% of road related injuries and fatalities will be eliminated. TK: Which is extraordinary given the fact that we lose about 1.3 million people globally each year. Abe: Exactly. We think that by 2060, as a result of all these efforts, governments will start looking into whether or not human driven vehicles should be allowed on the road. TK: Before that, it might even start to develop some economic pressure because of insurers who no longer want to insure human drivers, given the risk involved in doing that. Abe: Right. TK: Those last two points are really important ones to keep in mind. Much of what will cause a tipping point to occur with driverless cars has probably less to do with the technology, which is already at Level four and Level five in some cases, and much more to do with the economics of transportation. From a commercial standpoint, you can see what Abe is calling mobility fleets, these robo-taxis will be extraordinarily attractive, especially for someone in an urban setting but they're also probably attractive to insurers and to the government. The reason is simply economics. From an insurer standpoint, the risk involved with a human driver may simply exceed the appetite an insurer has relative to the risks of a robo-car. From a governmental standpoint, the safety issues here are absolutely extraordinary. I mentioned in speaking with Abe that we lose 1.3 million people a year globally to automobile incidents and by the way, what about the elderly, the infirmed, the disabled, and the younger children? I have little doubt, for example, that in 20 years, we'll look back on what we used to do today putting our teenagers in cars with other teenagers and wonder what in the world could we possibly have been thinking? The level of risk has serious economic consequences from a public health standpoint, from an insurance standpoint and critically, from the individual standpoint. All of this creates enormous momentum eventually to move as a society to driverless because it simply makes more sense from an economic and a safety standpoint. Abe: Then on the passenger-owned vehicle side of things, we think long-term, having an accessible, efficient, affordable, and convenient mobility fleet is going to disincentivize people from wanting to have that car payment and the insurance, fueling, and everything else. For those that do want it, it's going to lag a little bit behind because of the economies of scale. The curve to passenger-owned adoption will be slightly slower than the mobility fleet. TK: We certainly embrace a lot of the conveniences that today are included in automobiles such the adaptive systems, the automatic braking, what have you. We all push back on those a great deal. Abe: It makes driving so much easier. TK: That's right. Abe: A long road trip that would normally take four or five hours, when the car is doing 80% of the job, it feels like a two, three-hour road trip, right? TK: Did you catch the point that Abe was just making about owned driverless cars? Eventually, there's no need for ownership anymore. One of the more fascinating aspects of what the transportation industry will look like, especially from the consumer standpoint, is that there will be enormous disincentive for us to own our own cars. Today, a car is one of the worst applications of capital because the car sits idle for the majority of the day. If we increase that utilization to 80% or 90%, it also means that not only are we being more effective with the resources that we have, but the number of cars in the road drops precipitously. In fact, a report which you can download on the Foresight Radio webpage shows how by 2050, we could very well end up with less than 10% of the automobiles that we have on the road today because of the incredible increase in the utilization. Part of that is that the notion of ownership almost disappears except as an anecdotal curiosity. You may own an automobile because you're a collector. By the way, if you take this to the extreme, ultimately, some people are claiming that cars will own themselves through the use of technology like blockchain, another topic for another podcast. I was, however, curious if there were other use cases that Aptiv had come across that perhaps we weren't thinking about just yet. Abe: A couple of weeks ago, we partner together to do a new program with the National Federation of the Blind, where that's a use case that wasn’t central to our approach when we first started out but it's pretty self-evident that it's such a big value to those that don’t have the ability to see. We had these Braille cards that we develop to educate people on the technology and how the car drives itself. Being able to give mobility and make it available to those groups like the blind or the elderly is certainly a big advantage and a big positive to being in this industry. TK: The elderly, younger kids. I'd rather put my 16-year-old in a driverless car than put him in a car with another 17 or 18-year-old, right? Abe: Yes. TK: Would you say that, by and large, the car is a conservative driver? Abe: I think it used to be rather too conservative maybe a couple of years ago. TK: Which is why I asked because there are some cases where I have a Boston driver and Boston, you're too conservative, you can get in a lot of trouble. Abe: I would say as of actually seven or eight months, this car is balancing risk and aggressiveness versus conservative behavior. I'll give you an example where you have to be aggressive to be safe. That's exactly what our car would do. A few months ago, we're doing a similar demo here and a car was just pulling out of a driveway in front of us, and we were in a two or three-lane road. The car hesitated to see if this car is going to pull up in front of us or if it's going to stay. The car bolted right on the front of us and for us to avoid being hit by that car, we actually had to accelerate and move out of the way, and that's exactly what the car did, completely an autonomous one. Normally, if you think about that, you might think, “Hey, this is not a conservative behavior. This is more aggressive behavior,” but really, the bottom line is it was the safety. That's what we teach the car. TK: Right. It's making decisions that are based on risk assessment. Abe: Exactly. TK: As our ride came to a close, I asked Abe about one of the classic conundrums that comes up very frequently at events where folks talk about autonomous vehicle. It's called the trolley problem. You may have heard of it or some version of it. Imagine yourself in a situation where you have the ability to change the tracks that a trolley is on. On the one set of tracks, the trolley is barreling towards five people that are on the tracks and they most certainly will all be killed or severely injured. On the other track is one elderly woman. You got the option to pull the level and divert the trolley from the track with five people to the track with one person. What do you do? Well, the answer to that is that we're putting ourselves a bit of an artificial situation by limiting ourselves to just those two choices. It's entirely conceivable that with autonomous vehicles with smart infrastructure, the ability of vehicles to speak with each other and to speak with the roadways and the various systems that they interact with, that that conundrum may never actually occur. Here's Abe take on that. Abe: We think that in the future when road systems are smart and vehicles are autonomous, you're never going to have that problem to begin with. If you do, our car is going to prioritize safety and it's going to do whatever it can to avoid any kind of injury, fatalities on the road. Each scenario is different. I can't give you really a blanket answer to that, but it's the same approach we were just discussing. What is the safest, smartest thing to do for that vehicle if that vehicle was an experienced, safe professional driver? That would be the choice. TK: Abe, thanks very much for taking us on this road trip. Quite an experience. Abe: Yes, good to have you. TK: As I exited the vehicle and took a look at the car driving away, it occurred to me that what we're developing is not just the technology, but if I may, a new species that we're co-inhabiting the world with. These autonomous devices and vehicles will, in many ways, create not just a higher level of safety, but will give us the luxury and the liberty to experience the world in an entirely different way. While I get that saying it's a new species may sound a bit hyperbolic, I’d encourage you to experience a driverless car and to see how it changes the way you think about this notion of how machines can exhibit behaviors. In many ways, what we've done is proven that we can create machines that do in fact learn in much the same way that human beings do. While there are areas where I don’t fear machines will be able to displace human judgment, our ability to be empathetic and the very human nature of how we communicate with each other, there are certainly some areas where intelligence behavior can become very much a part of the devices and the vehicles that we share this planet with. To find out more about driverless technologies and Aptiv, including the Aptiv white paper on safety and Delphi’s research report, Transportation 2050, just check out the links in the Foresight Radio homepage at Foresightradio.com. Thanks again to our sponsors for this episode of Foresight Radio, Wasabi. Take a look at how Wasabi is changing the rules of the game for cloud storage at Wasabi.com. Thank you for listening. If you enjoyed this podcast, please be sure to subscribe to Foresight Radio and to share it with your friends and colleagues. This is Tom Koulopoulos. I look forward to joining you again soon for another episode of Foresight Radio, where we explore the future for how we will work, live, and play in the 21st Century.