null Episode 18: The Future of Mobility
Double Take podcast

Episode 18: The Future of Mobility

Double Take podcast Audio Thematic Equity Multi-Asset Index
May 2021
Episode 18: The Future of Mobility

Our latest episode of Double Take explores the future mobility revolution and the innovative technology behind how we get there.

Rafe Lewis: Hello, and welcome to episode 18 of Double Take, an investment podcast that delves into big beefy themes impacting the public markets today, and more importantly, tomorrow. I'm your cohost, Rafe Lewis.

Jack Encarnacao: And I'm your other cohost, Jack Encarnacao. Today's big beefy theme, mobility, or more precisely, the technology that makes mobility possible and the thorny questions and obstacles raised by the march toward, for example, autonomous automobile traffic, pilot-free passenger aircraft, autonomous mobile robots, ADAS, and so forth. Our conversation as usual will offer two takes, one from Rob Zeuthen, a 30 year veteran small and mid cap investor who also serves as the lead portfolio manager of Mellon's Mobility Fund, as well as Sam Anthony, chief technology officer of Perceptive Automata, a US-based startup working with major auto manufacturers on developing software and systems to allow mobile objects like cars to better recognize and predict the movements and intentions of things like pedestrians and bicyclists and other hazards.

Rafe: Yeah, no small feat here in Boston where we live in infamy. But I don't know about you, Jack, but after a year of sitting in my house amid a pandemic, lockdowns, just general fear, I'm a pretty darn ready for a mobility revolution. So let's turn to Rob Zeuthen. Rob, welcome.

Rob Zeuthen: Thank you for the invitation. It's good to be with you.

Rafe: Oh, it's great to have you. So, Rob, set the table for our listeners and what's happening broadly in the field of mobility. For all the dreams that we, or certainly I had as a kid of flying cars, jet packs, individual land speeders like on Star Wars, it feels like not a heck of a lot has changed for the everyday consumer when it comes to how we get around. I mean, the distance between the Model T Ford and today's car, it feels like it's not that massive, right? I mean, they still travel on four wheels made of rubber on a road, right? So are we really on the verge of something revolutionary here?

Rob: Yes, we are. First I'd like to say that your dreams are still valid, but they're still years away when you think of fully autonomous vehicles or flying cars. But the good news I can tell you today is that we've seen lots of innovation in just the last few years that really bring improved access to transportation to all those who need it, and as importantly, more safe transportation for everybody.

Rob: With the mobility strategy, the entire team across the Mellon Equity Research Platform focuses on innovation that can really deliver improved outcomes over the next several years. And so when you say autonomous, we think a little bit more in terms of adaptive innovation, what many people call ADAS. And you're seeing ADAS penetration significantly increase in many of today's vehicles. And this is not going to stop anytime too soon.

Rob: We really need that innovation because over 1.3 million people are lost per year, and it just doesn't have to be that way. Over 90% of auto fatalities are due to human error. And so even small increments of progress can have outsized impact on really improving and saving lives.

Rafe: Rob, can you just give us a couple of examples of what ADAS is that's out there in people's cars today?

Rob: Well, there's many different forms, and there's going to be more forms in the near future, but you can think of ADAS as any combination of taking control of steering a vehicle, breaking the vehicle, or accelerating the vehicles. Those are the earlier forms of ADAS, and it can be delivered automatically where the car takes control, or where it can work in conjunction with the driver.

Jack: So Rob, where do you think we should expect to see the first tangible, widespread changes in mobility, something that feels like a step change in how we get around?

Rob: I think we've already seen that to a degree with ride-sharing companies, which, as you know, are broadly available, not just in the US, but in many regions around the world. And ride-sharing companies are an important innovation because it provides ready access to transportation. And it's not just the access, it's the safety as well.

Rob: It can help people get home late at night safely, whether it's because someone has a concern about their own wellbeing or perhaps because they shouldn't be driving and this is a way for them to get a safe ride home.

Rob: So when you think of those two different dynamics, it can help you see how impactful ride-sharing can be to really protecting and improving people's lives. You should know that, at least here in the US, over 10,000 people die per year because of drunk driving, and ride-sharing is just one of those many ways that we can really save people from that tragedy.

Rafe: Yeah. So, Rob, it's funny you bring up the ride-sharing companies, because when I think about what happened there, they came on the scene, they had all this innovation, and people went running for it. And all of a sudden, it was sprouting on every corner and every street in America, and then it started around the world, and regulators and lawmakers really had to sprint to play catch-up to the new reality and to try to figure out a way to fold ride-sharing in after the cow was already out of the barn.

Rafe: So I guess what I wonder is for this next phase of mobility and some of the innovations that are coming down the pike here, are the lawmakers and the regulators ready here in the US and abroad? Are they coming up with things now to enable this to happen? Or are they going to be playing catch-up and there could be some wild cards out there?

Rob: So Rafe, I expect it to be a repeat where regulators are playing catch up for two reasons. The first is innovation continues to accelerate across a wide spectrum of applications, which makes it more difficult to craft a consistent regulatory framework. The second challenge is that unlike ride sharing, which still relies on a human driver to deliver the service, autonomous is much more complex where a driver is relinquishing control to a computer to control the vehicle. And because of that technical complexity and all of the potential legal ramifications that might come around it, it will require a more comprehensive regulatory framework that could take several years to really form. The good news however, is that recently two senators have proposed new legislation in the United States that would empower the NHTSA to exempt more autonomous vehicles and what this would help do is facilitate more rapid deployment of these vehicles that can help accelerate the US's potential leadership in this space and field these vehicles to make sure that they're tested properly to deliver the convenience and the safety for consumers.

Jack: Rob, President Biden has unveiled a rather ambitious infrastructure spending plan. Has a lot of money in there for things like electric vehicle charging stations, green energy generation, but correct me if I'm wrong. I didn't really see anything in that plan that would help usher in a new era of smart mobility like tomorrow. Do you think this is ultimately going to fall entirely on the private sector to usher in this revolution? Do we need new dedicated infrastructure, like highway lanes for autonomous semis for example, semi-tractor trailers if this is really going to happen?

Rob: I think there are two different paths to innovation around mobility. The first is moving to electric powertrains, which I very much think the Biden plan supports. The second track is more around autonomous vehicles. And I do think there will be more reliance upon the private sector to advance that innovation. But regardless as to what happens to either track, both will really rely upon improved infrastructure, not just for electric charging, but also for what we think of as smart infrastructure, like V2X. Examples like that would include smart traffic lights and smart ropes that would help provide an extra layer of security and safety around today's transportation system that would reduce the reliance on a vehicle having to take all the steps it might need to take to provide a safer, more autonomous future.

Jack: And to be clear Rob, V2X meaning?

Rob: Vehicle to everything. So that means that a vehicle would be able to communicate not just with other vehicles, but with infrastructure, with lights, with roads, with pedestrians, with cyclists to make sure that everyone has situational awareness that leads to safer outcomes.

Rafe: Hey, Rob, just thinking geographically for a moment here, where should we expect to see this mobility revolution kick into high gear first? I mean, would it be here in the good old USA? Is it Europe? Or is it Asia where there's so much innovation going on? Or maybe it's in the Middle East where it might just be easier to get this done because they don't have snow or curvy roads?

Rob: It's a great question. And as we sit here today, electric vehicles, which by and large bring some of the highest levels of autonomous or safety related features and services, are seeing the fastest adoption in Asia, especially in China and in Europe. And so that's where we really expect the innovation to be adopted first. And those regions have also provided stimulus over the last year to really help accelerate that adoption cycle. But thanks to the Biden administration, that is a sign that the US is starting to think about realizing its potential as well.

Jack: Where do we need to see much more investment in your opinion, Rob, as far as research and development to kind of smooth the path here and accelerate the developments we're talking about?

Rob: Well, humans are very difficult to model and that is the gating factor to autonomous systems. As we sit here today, today's ADAS systems can identify objects, but we still have a long way to go in terms of understanding human intent. These systems can recognize when pedestrians or cyclists may be on a curb or in a crosswalk, but they don't necessarily understand well what the intent of that pedestrian or cyclist is. And so that is an area where we expect much more investment in the next several years to help promote a better performance of these systems.

Rafe: And a little spoiler alert here, the second half of this podcast folks is with Sam Anthony. He's the chief technology officer of Perceptive Automatica and he's going to be talking quite a bit about kinds of human intention, how to read it and things like that. But before we get there Rob, one or two more questions if it's okay. I can't help it. I already kind of talked about the land speeders and flying cars and stuff like that. So let me just throw one more at you. In the public markets today or very soon, we have companies that are going public either in these blank check vehicles or otherwise that purport to be flying taxi companies. There are some pretty far out mobility concepts out there that equity investors can get involved in. So without going into the merits or lack of merits of any one of these particular companies, what can you say broadly about this phenomenon and whether institutional investors like us who are going to be putting risk capital into that kind of thing yet?

Rob: Well, I think it will still be quite a period of time before we invest in several of these longer term ideas. In order for us to deliver performance for our clients, we need companies where we can vet the technology, the management teams, understand the regulatory framework and the liability associated with that, and then really have confidence that there will be consumer adoption. And while a lot of these applications look exciting and futuristic and filled with potential, we really need to see all of those elements work together before businesses can be successful. And so given the complexity of what I just talked about, we think it will still be several years for that to transpire and thanks to the strength of the Mellon Equity Research platform, my co-managers and I really have the ability to leverage our firms global capabilities and deep domain expertise to understand which companies have that reward versus those that have the risk.

Jack: And Rob, the mobility strategy has been in place for how long now?

Rob: Well, it's been in place for over three years.

Jack: Okay. So take us back three years as we wrap up with you here, sort of how you looked at mobility as we crafted a strategy around this theme and where we are today. Are we where you've expected us to be further ahead, further behind?

Rob: I'd say in many respects, we're pacing ahead. Our team uses an innovation framework called CASE, which stands for conductivity, autonomous, sharing, and electrification. And much of today's conversation has been around autonomous, which is one key element of the strategy. That's still filled with promise, but as we talked about, there's still a lot of work to go. I'd say the other elements of CASE around conductivity, thanks to 5G, sharing which is ride sharing and electrification are growing rapidly. And it's not just the innovation that has propelled the fundamentals of these companies. It's also because of COVID, which has accelerated these trends as well. And when you think about all the stimulus that's been announced worldwide to help create a greener recovery, it's provided further propellant to these different elements of the mobility strategy.

Jack: Well, Rob, we look forward to seeing it develop on pace and thanks very much for your time today on double-take all the best with the strategy. And we'll talk to you again real soon I'm sure because things change very quickly in the mobility world.

Rob: Thank you.

Rafe: Welcome back to part two of Double-Take. And we're going to be doing a deeper dive now into this mobility theme. And to do so we're going to bring up Sam Anthony, who's the Chief Technology Officer of Perceptive Automata. Sam, welcome to Double-Take.

Sam Anthony: Thanks for having me.

Jack: So Sam, we've spoken a few times now, as I've done some investigative projects, trying to better understand certain nuances into the mobility theme. And so, it's really great to get you on the pod finally. And we heard from Rob Zeuthen from Mellon, who laid out a pretty compelling view of the brave new mobility future. So now, maybe you could take us into the weeds. It sounds like we're going to need an awful lot of artificial intelligence, machine learning and really fast communications hardware and software to make this all happen.

Jack: So, could you give us a sense for, I guess, where the low hanging fruit is here, and where it's going to be really tough to get computerized vehicles to a place where they operate as safely and hopefully much more safely than humans behind the wheel?

Sam: Yeah, I think that... I mean, the first thing I would say is that, that low hanging fruit is not super easy to come by in this world. One of the aspects of autonomous vehicles that I find very interesting is that there tends to be, within the field, a very good understanding of just how hard a problem this is, that doesn't necessarily get out to wider understanding. We're making enormous progress. And it's incredible that this much progress has been made.

Sam: But in terms of the complexity of the problem of let's say getting a level four, so a fully autonomous vehicle, operating on busy urban streets, that's a technological problem the scale of which has very rarely been seen, It's an extraordinary advance along a whole number of levels. And I think, because of that, what you really going to see in terms of how automation is going to become part of mobility, it's going to be kind of bit sliced off at the edges. So, you're going to see applications that are applications like the next generation of adaptive cruise control, something like what are called level three systems, where it can take over a chunk of the driving task for you, in your regular car, while you're on specific well-marked highways.

Sam: Applications like that, we'll increasingly see. We'll increasingly see applications where you're able to use a tele operator, so someone piloting the vehicle remotely. If you can kind of use that and still get the economic value, so something like a sidewalk delivery robot, you'll see those coming out. I think the kind of last thing you're going to see the, the hardest nut to crack, and the one that we specifically pay a lot of attention to, because we think we can kind of knock out some of the harder aspects of it, is having a vehicle that can navigate itself, without tele operation, in a dense urban area, with other drivers and pedestrians, and the kind of social interactions, and the kind of social driving hat humans in particular are good at.

Rafe: Sam. What is tele operation, just to clarify?

Sam: Tele operation is when you have a driver of a vehicle who's remote, so somebody sitting in a screen with, let's say, 10 or so vehicles. And when there's a situation where the vehicle can't navigate itself, they can step in and they can take control of it. And they can actually drive it remotely.

Jack: Wow.

Rafe: Got it. So, this is almost like when you see someone operating a drone or even remote surgery when a surgeon is in LA, the patients in New York, that kind of thing.

Sam: Exactly. And when you look at autonomous vehicles on the roads today, there are certain companies that have vehicles that are operating without safety drivers. In California, in particular, you can get permission to do testing of these vehicles without safety drivers behind the wheel. Pretty much universally, the reason they're able to do that is that they do have tele operations. So, they do have a remote driver who's able to take over those vehicles. For autonomous cars, in kind of the classic sense, which is like, this is a car you get and it drives you anywhere where you want to go on its own, tele operation is a tricky economic proposition because you really...

Sam: It doesn't really buy you much over not having a safety driver in the car, because that remote safety driver has to be really just as attentive as the safety driver in the car would be. For some of these other applications like sidewalk robots, like some of the automated trucking applications, you can get a lot further and say, "Okay. That's okay to have tele operation. We know we're going to have tele operation for the difficult part of the drive, but we think we can get to automation for the easier parts of the route, and kind of unlock some of the economic value, even without the ability for the vehicle to really completely drive itself the way you would need to for an autonomous car in a city."

Jack: In terms, Sam, of those dense urban environments, I know that a fair amount of your work focuses on a car being able to detect the intent of, say, a pedestrian, looking to cross the street. Are they looking to cross, and so this car should slow down? Or maybe they're just standing at that corner because they're waiting for a taxi or something like that. And their body language indicates they're looking at their phone and not looking to cross, even if they're standing right at the sidewalk or the crosswalk, we should say. That's a fascinating thing to me. I mean, can you talk about where we are in achieving that level of intent recognition and discernment?

Sam: Absolutely. Yeah. And that is definitely... It's what we focus on. And the reason we focus on it is it's a huge part of how humans actually drive. And that's not necessarily something that we, as humans, are super aware of. When you're driving a car, if I asked you, "What are the skills that are necessary to drive a car?" You're going to say, "All right. Well, you need to be able to steer it. You need to have control over the motion of the vehicle. You need to know the rules of the road. You need to be able to read street signs. You need to use your signal." And the story we sort of have in our heads about what driving is, is kind of a rule-based story. So, there's a set of rules. There's a set of skills you need to acquire. And then you'll... Then, you can drive.

Sam: And the reason that humans... we think about driving that way is that, that is how we learned to drive. But what that ignores is all of the things that we just naturally are incredibly good at as humans. And one of the sort of biggest categories of knowledge like that, of abilities like that, is social understanding, is the ability to pick up social cues and non-verbal communication from other humans. And it turns out, if you look at what's actually happening in the driving, those kinds of social interactions are a huge part of how you make decisions about how you drive a vehicle in the real world on the road. And so, what we do is we build models that give autonomous vehicles some of that understanding. And in terms of how far we've gotten, well, we have a product, and that product is available to customers who are building out autonomous vehicles.

Sam: And what it does is it will tell you, at a pretty high of fidelity, this person who's standing out here... If you had 500 people in the car with you, and you asked them all to take a vote on whether that person wanted to cross in front of the vehicle, this is what that vote would look like. And that's information that you can then use to change the behavior of the autonomous vehicle, to slow down and let that person cross, or to keep going, even though it's a crosswalk, and they have their toe in the crosswalk, and so you otherwise would have to stop. And so, that's what we're providing. And I would say that, that's what...

Sam: It works. It does that. It doesn't, by itself, get you to an autonomous vehicle that is going to handle all of these situations perfectly. But without it, you can't get to that state of the vehicle. So, it's not... It doesn't get you all the way to "Now, you can deploy your level four vehicle in a city without a safety driver." But if you don't have this ability, if you don't have the understanding of sort of what's in the heads of pedestrians out in the road, you can't get there.

Jack: And how about vehicle to vehicle communication? That is the ability of one vehicle to tell, not what... obstacles on the road, like a cyclist or a walker, are doing and thinking, but other cars. I understand that it gets really complicated when it comes to being able to see through the windshield of a fellow driver, to be able to tell what he or she may be thinking.

Sam: It is. It's interesting. When you look at what people are doing, people don't actually look at other drivers that often because you mostly can't see them. But it turns out that the pose of another car, that the way that, that car is moving, and information, what kind of car it is, how the person is driving, is actually a very rich social signal. So, if you sort of put yourself in the mindset of, you're going through an intersection, you have a green light, there's an opposing car that wants to make a left turn in front of you.

Sam: This is Massachusetts, so they don't have their signal on. But you can still tell very, very clearly that they want to make a left turn. And so, how do you know that? Right? And it's not because you're looking at the person in the car. It's because they are kind of edging into your lane a little bit. And they're just... They're signaling, with the motion of the car, what they want to do. And so we, as social creatures, are incredibly good at picking up cues like that. And so, what Perceptive Automata does is build models that does that same kind of thing.

Rafe: Got it. Okay. That's very interesting. And that's a good segue into kind of something that I'm personally fascinated by, which is some of the ethical dilemmas posed by autonomous mobility. So, here's an example, right? You have a driverless car. It's filled with teenagers. They're speeding along a road, and it's edged on one side by a cliff. And then, on the other side, there's a rock face. So, think kind of like Big Sur in California, right? All of a sudden, an elderly woman pushing a baby carriage, walks into the roadway. The car's on board computer now has this split-second calculation to make, because there's no time to hit the brakes and avoid a calamity, right?

Rafe: So, here's this calculation. Do we drive headlong into the elderly woman and the baby, or do we drive over the cliff and kill the five teenagers in the car? So, my question to you is, are the programmers of these systems effectively playing God? And can a few kind of tech geeks, like us or anyone else in a Silicon Valley office, can they really make these decisions unilaterally?

Sam: So, I love to talk about this stuff. And I love to talk about this stuff, because I think that there's a... It opens up a really interesting way of understanding the engineering problem, because... So, for your example, what I would say is that, if an autonomous vehicle ever got into that situation, the people developing that autonomous vehicle would have made huge mistakes. So, for example, if I'm programming an autonomous vehicle, and I'm building a motion planner, that motion planner should not get in a state where it's going too fast for conditions. And certainly, if you're speeding on a mountain road, and don't have the ability to stop safely for an unexpected obstacle in the road, that's too fast for conditions. So, as an engineer solving the engineering problem of how these vehicles should behave, just by getting to that situation, you've already made some very, very big mistakes.

Sam: And I think, actually, when you look at the kinds of challenging situations that come up, almost always, the answer is, "Well, the vehicles should be designed with a conservative enough motion plan that this does not happen." And once you've done that, then the questionable... How do you handle these very, very tricky, ethical situations? Those ethical situations come up vanishingly rarely. And it's something where, really thinking about the safety engineering for these vehicles, your first, second, third, 100th impulse should be, "How do we do the safety engineering so that the vehicle does not get into these failure modes where there's no good option?"

Sam: And I would say, in terms of testing and understanding the behavior of these vehicles, if we got to the point where the biggest safety problem we had was the overwhelmingly unlikely situation where there are sort of two equally bad, or two complexly bad options, and no other options, these vehicles would perform incredibly well. So, we're not there yet. I'm honestly not even sure we'll kind of ever get there. And really, the question that people should be asking of these designers is, "How do you institute behavior in these vehicles such that they do not end up in these situations, which are these complex moral dilemmas?" Because really, that's a very well-formed safety goal. And that's something that, as an industry, we still have, I would say, a fair ways to go before we're there.

Jack: Sam, could you walk us through the various components necessary to really take us into the next chapter of mobility? Maybe more importantly, are there programs or even pieces of hardware, not yet invented, that we will really need to get this done to its highest and best potential?

Sam: You know? The hardware is good. I think that some of the sensors have been coming down in price dramatically. If you look at LIDAR, the kind of LIDAR that you can get now, for a reasonable amount of money, to put it in a fleet of vehicles, would have been hundreds of thousands of dollars not that long ago. The cameras are good. There's sort of some iterative engineering work around sensors just in terms of getting to the right scale, where they cost the right amount, where you can deploy them on vehicles. I would say that compute power for vehicles is really already there.

Sam: The next generation of just regular passenger cars is very likely to have sufficient or almost sufficient computer in it to run a full autonomous vehicle stack. Fundamentally, this is an incredibly challenging software problem. And I think that it's really... If you look at what is hardest now, what is furthest from being solved now, it all resolves to software, but really to questions of behavior and understanding, so not just a software problem, but a problem that that gets into some pretty deep questions about what sort of intelligence you need to be... to have an autonomous system that operates in the world.

Rafe: Okay. Sam, we've seen a problem very recently here, but one that I feel like is going to have some legs, which is a shortage of some of the key technology components for today's cars. And we have some plants that are kind of idling around the world right now, because there simply aren't some of the components necessary to build the cars, which are increasingly computerized well before we get to the point of kind of full autonomy, et cetera.

Rafe: Is there a supply chain issue here that's going to have to be rectified before we can get to this brave new future? And just thinking kind of nationalistically here for a moment, does the United States need to invest in some kind of manufacturing capability to be able to effectuate the brave mobility future?

Sam: I would say yes, but I would say that, that, if we get the conversion to electric drive trains, and the conversion to most vehicles sold being electric vehicles... If we get that right, the infrastructural questions underlying autonomy are kind of going to get solved along the way. So, I've seen very convincing arguments that the power requirements of the computer that's going into vehicles now, and which is required for autonomy, but which is going into these vehicles anyhow, is such that the fleet fuel economy standards in Europe, and also now they're back in this country, it's...

Sam: You're not going to be able to reach those fleet fuel economy standards on an ongoing basis, unless you really accelerate and focus on converting most vehicles produced to... and most vehicles sold to electric drive train, so having really a primarily electric set of vehicles on offer for sale. If you can accomplish that, those vehicles very much are kind of ready for the autonomous revolution. That's where the supply chain issues are going to come.

Sam: And that's going to be, I think, a pretty substantial retooling of an industry that is not historically known for being terribly nimble about these things. So, I think that, that's huge. And I think that, that's going to be hard. I think that readiness for autonomy is, to some degree... or the readiness for kind of next generation, more autonomous mobility applications is a little bit of a trailing indicator of that.

Jack: Sam, another piece of the mobility question is consumer acceptance of the technology. We've all seen the news stories about reckless early adopters, closing their eyes, letting their car do the driving, and with today's still improving autonomous technology. But what will it take, do you think, for the vast majority of humanity to not only be safe, but also to feel safe in these cars?

Sam: I think that's a hugely important question, and it's a hugely important piece of the puzzle. What you need for acceptance of these vehicles is both clarity around how they work, clarity around how they differ from human drivers, but also you need these vehicles, when they're on the road, to behave in a way that is comprehensible. So, they need to drive in a way that people understand is respectful, and safe, and coherently similar to how a human would drive a car. And if you have that, then when people experience these vehicles, both inside them and around them, they will become routine. They will fade into the background. And acceptance will very much be achievable. On the other hand, if the performance of these vehicles is shrouded in mystery, if there are incidents where these vehicles are not considerate in they're driving, or, God forbid, incidents where these vehicles are causing, or are involved in, accidents that would not have happened if there wasn't an autonomous vehicle involved, the risk to the industry of social approbation and the lack of acceptance is very, very high.

Sam: So, for us, one of the reasons we talk about social driving and social interactions so much is that, without the ability to both understand and kind of participate in social interactions with human road users, you can't get to driving behavior that is sort of recognizable and comforting for the people inside the car and outside of the car. So, it's really having the ability to drive, in the right way for an autonomous vehicle to drive, is the central bit of this acceptance story. And the acceptance story is sort of the absolutely existential one for the industry being able to unlock the potential of this technology.

Rafe: Sam, another big topic going on in the country right now is whether and how much to spend on infrastructure, and even what the definition of infrastructure is in the United States. And one thing that came to my mind is wondering whether you... someone like you believes that there is a large investment in infrastructure necessary to have the true mobility future that we want. So, in other words, you were saying some of these engineering questions are incredibly complicated to get to that highest ultimate level of autonomy, right?

Rafe: And so, I wondered, "Well... But what if you build highways where there are special kind of lanes for autonomous traffic, or there's certain investments made in signaling and other things to create the communications infrastructure so that this is way safer than it would be with today's infrastructure?" I just wonder if you think that's part of it. Is it necessary? Is it sufficient? Is it neither?

Sam: Speaking as somebody who is trying to solve these engineering problems, I don't let us off the hook like that. The challenge of building an autonomous vehicle, I think, really should be defined as building a vehicle that can operate as a human would in human environments. And we know that it is possible to drive safely with the infrastructure that exists because people do it every day. And so, that's really what we're trying to accomplish. When I think about the infrastructure that would lead to a future where mobility is transformed into something that's much more humane, and is less of... and kind of unlocks the ability of people to have a life that's not sort of so constrained by driving everywhere, it's things like building multimodal transit in cities, having environments that are shared environments, things where you could take a rideshare vehicle, or a bike share, or an electric scooter, or walk, and kind of pick them as you choose.

Sam: And in environments like that, the power of autonomous vehicles could be that you have a vehicle that, unlike a human driver, can be in these sort of dense multimodal environments and be much, much safer than a human, not drive too fast, not get frustrated. But I don't think that... I think that building infrastructure specifically for autonomous vehicles is not the... It's moving away from what you can really get with autonomy. And I think it's redefining the engineering problem away from what the engineering problem should be, which is, how do you have vehicles that can drive themselves, that can interact in a human-like way in human environments?

Jack: Interesting. So yeah, don't make it dependent on the environment around the car changing, but make it work for whatever environment it's dropped into, including the current one. Yeah. So, should we expect a kind of transition period then, Sam, where legacy vehicles driven by humans are intermixed with autonomous? That, to me, seems more almost complicated and fraught than the desired end state, where all vehicles are controlled by computer. And the possibility of random chaotic human actions is much lower.

Sam: You're still going to have pedestrians, right? Or I hope you're still going to have pedestrians. I certainly intend to keep walking across streets. And I would encourage everyone else to do so. And so, I think that, when you talk about cities or roads, these are for people. These are the things that we construct for ourselves. And so, I very much imagine that human operated vehicles are going to be... and pedestrians, are going to be part of the story forever. You're not going to move away from bicycles. You're not going to move away from... Hopefully, we could go back in the direction of children walking to school. And so, in that context, I think that relying on a completely autonomous system is not... I don't think it's achievable and I don't think it's the right goal.

Rafe: Sam, the obligatory final question for all guests like you, when the conversation of autonomy and mobility comes up. When do you expect fully autonomous vehicles to be on the roads of the United States?

Sam: So I will say, in my time working in this field, I have gotten incredibly good at avoiding that question. I think-

Rafe: But Sam, my job is to not let you be good at that.

Sam: Here's where I stand on. What I usually say, and really, I think this is what we can know, is that you're going to see something with autonomous vehicles like you have seen with AI per se. Where there is going to be an incredible amount of progress in terms of increasing autonomy in specific domains and for specific use cases. So we talked about sidewalk robots with tele operation. We talked about highway based level three systems, which can take over the driving task as long as the human driver is still there and being attentive. And there's a lot of iterative process in those. The question of when we're going to see fully autonomous robo taxi type vehicles on urban streets or where every car sold in this country is going to be capable of driving itself, any estimate I would make of that, the error bars are going to be far too wide. It could be five years. It could be 50 years.

Sam: But I think that what you want to look at is how these limited domains in these limited applications get rolled out. And that's something there's going to be a lot of progress in specific domains over the next five or six years. It's just, it's not going to look like robo taxis in cities.

Rafe: Okay. So the headline is, Sam says five years now.

Sam: Well, you always say five years. If you look at the history of AI, everyone has always said five years. We're going to have it in five years. So if anyone with expertise in machine learning or artificial intelligence or autonomy ever says five years to you, that's how you know they're not engaging with your question.

Jack: Great takeaway.

Jack: Well Sam, thank you so much for taking the time and joining us here on Double-Take, getting us into the weeds on what it's going to take from an engineering and kind of almost a moral and ethical perspective. You're wrestling with quite a bit there, and doing the important work. So, thanks for sharing some of that big brain with us this week on Double-Take.

Sam: Well, thank you so much for having me.


Mellon Investments Corporation (“Mellon”) is a registered investment advisor and subsidiary of The Bank of New York Mellon Corporation (“BNY Mellon”). Any statements of opinion constitute only current opinions of Mellon, which are subject to change and which Mellon does not undertake to update. This publication or any portion thereof may not be copied or distributed without prior written approval from the firm. Statements are correct as of the date of the material only. This document may not be used for the purpose of an offer or solicitation in any jurisdiction or in any circumstances in which such offer or solicitation is unlawful or not authorized. The information in this publication is for general information only and is not intended to provide specific investment advice or recommendations for any purchase or sale of any specific security. Some information contained herein has been obtained from third party sources that are believed to be reliable, but the information has not been independently verified by Mellon. Mellon makes no representations as to the accuracy or the completeness of such information. No investment strategy or risk management technique can guarantee returns or eliminate risk in any market environment and past performance is no indication of future performance. The indices referred to herein are used for comparative and informational purposes only and have been selected because they are generally considered to be representative of certain markets.  Comparisons to indices as benchmarks have limitations because indices have volatility and other material characteristics that may differ from the portfolio, investment or hedge to which they are compared. The providers of the indices referred to herein are not affiliated with Mellon, do not endorse, sponsor, sell or promote the investment strategies or products mentioned herein and they make no representation regarding the advisability of investing in the products and strategies described herein. Please see for important index licensing information.