Driving hands-free across America - feat. Alex Roy

How far can a vehicle drive itself without human intervention?

In this episode of FISITA's Beyond the Car podcast, we talk about just that: long-distance, hands-free driving.

Motivated by decades of record-breaking road trip experience and enthusiasm, and inspired by the idea of driving from west coast to east coast with zero human intervention, our guest on this episode joins us to talk about his recent trip from Los Angeles to New York City in a Tesla in Full Self-Drive mode.

As you'll hear, Alex Roy is a huge fan of human driving, and of automated vehicle technology. He's also very competitive, curious, analytical, and methodical - all the things you need to be to sit in a car that drives itself 3,081 miles from Redondo Beach in LA to midtown Manhattan in NYC - that’s 4,958km, in a time of 58 hours and 22 minutes, all without human intervention.

We discuss the background to the challenge and the history of coast to coast driving, and Alex tells some great stories from the trip he completed in January 2026.

He also shares valuable insight into automated driving, having advised at a strategic level right at the top of an AV company, namely Argo AI, and from the perspective of a regular automated vehicle user.

Connect with Alex Roy

New Industry Venture Capital: NIVC

LinkedIn: linkedin.com/in/alexroy

Follow FISITA's 'Beyond the Car' podcast on LinkedIn

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

You can listen to FISITA's 'Beyond the Car' podcast wherever you get your podcasts.

For more information about FISITA, head to FISITA.com - and to learn more about the 'Beyond the Car' podcast, you can contact us at info@fisita.com.

Thank you to Martyn Strong for production.

Transcript

SPEAKER_01 0:08

How far can a vehicle drive itself without human intervention? Welcome to Fistita's Beyond the Car podcast. I'm Martin Carl, CTO at Fistita, and in this episode we'll be talking about just that long distance hands-free driving. Motivated by decades of record-breaking road trip experience and enthusiasm, and inspired by the idea of driving from west coast to east coast with zero human intervention, our guest in this episode joins us to talk about his recent trip from Los Angeles to New York City in a Tesla in full self-drive mode. As you'll hear, Alex Roy is a huge fan of human driving and of automated vehicle technology. He's also very competitive, curious, analytical, and methodical. All the things you need to be to sit in a car that drives itself 3,081 miles from Redondo Beach in LA to Midtown Manhattan in NYC. That's 4,958 kilometres in a time of 58 hours and 22 minutes. You're going to hear about the background to the challenge, Alex Roy's vast coast-to-coast driving experience, and some great anecdotes from the trip that he completed in January of this year. But you're also going to get valuable insight into automated driving from someone who advised at a strategic level right at the top of an A V company, namely Argo AI, and who uses automated vehicles every week. I'll be back at the end of the show with some closing thoughts. Here's my conversation with Alex Roy. So welcome to the Beyond the Car podcast, Alex. I should say to our audience that our background overlaps a little, not much, but it does a little. I was doing some work for Argo AI at the time when you were director of special operations at the company. And you came into that role because the CEO at the time, Brian Sileski, wanted someone with a fresh, independent, and critical eye as an advisor. You're also an author and co-host of the excellent Autonocast podcast and a general partner at New Industry Venture Capital. And critically, founder of the Human Driving Association, which we'll talk about in a moment. And the reason that's critical is that you're also a cannonball run record holder. So welcome to the show. Let's set the scene. Where are you right now?

SPEAKER_00 2:15

Uh I'm in Scottsdale, Arizona, where I've lived for a few years. And I first came to visit when I was doing some competitive intelligence for Argo to sniff around the Waymo and Cruise deployments, if anyone remembers Cruise. And uh I liked it here. And so this is where I live.

SPEAKER_01 2:30

So I said we know each other from Argo days. Shed a little light on to the work that you were doing there at the time.

SPEAKER_00 2:36

So uh, you know, back in 2018, uh I was running the Atonicast and I was an editor at large at thedrive.com cars that I co-founded, and uh I wrote a column about the fatal crash of the an Uber ATG-equipped uh test vehicle, which was a Volvo XC90. And the crash occurred in Tempe, Arizona, not far from where I currently live. So I wrote a column speculating on the cause of the crash and uh which killed a woman named Elaine Hertzberg. And my my thesis was that it was not a single technical error or even multiple technical errors, but a fundamentally human chain of cascading decisions which led to this inevitable result. And I was I was immediately contacted by Uber, uh, who asked me to come visit them at Pittsburgh at their headquarters, and they offered me a job. Uh, and I and I spoke to Dar about it at the time. But it was there were several great people there, but it was not a company that I wanted to work for. I thought the problems were really cultural. And and so across the street was Argo's headquarters, and I had interviewed the CEO for the Atonicast, and I liked him very much, and I thought that he was a very ethical and terrific leader. And so we met and he said, Look, how would you defend a company of you know this size and scale? You know, they'd raised at that point over a billion dollars. How would you protect the company from the types of the decisions that led to the fatal crash at Uber? How can we prevent those? What what's the list? And I made him a list and he said, okay, if you want a job, that's your job. You report directly to me, and you will be separate from the other teams because they're already answering that question. I need someone to answer that question independently. And I was there for four years.

SPEAKER_01 4:22

So well, and the four years came to an end, as we know with the demise of Argo. You know more than most about what went wrong there, but I was always deeply impressed with the technology, the cautious approach and the idea of rolling out when we're ready, rather than the more popular idea of moving fast and breaking things. Based on the technology alone, do you think that Argo could still be competitive in the AV world now?

SPEAKER_00 4:46

All right. So uh, you know, Argo was always probably the most methodical and cautious of companies, but Waymo also was methodical and cautious and was around for many years prior and was you know much better funded. I mean, Argo ultimately had raised almost uh almost$4 billion. So Argo fundamentally, its problem was that it was telling the truth in an ocean of companies that were exaggerating their timelines and capabilities. But if Argo had survived, Argo would still be in a difficult position because even if Argo could have caught up in terms of operationally and technically, you know, Waymo is still very well funded, very, very well funded. And Waymo is you know doing most of the things that pretty much everything Argo is doing well, or Waymo is doing them, it continues to do them well. So you have a broader threat, um, which is the perception of Tesla as you know an overnight arriver at driverless operations, persists. Uh, and Tesla possesses what I call narrative command, which is that many people want to believe they're they're correct, and to end vision only neural nets can work, and that they can build vehicles at scale more cheaply than anyone else that will be driverless. And as long as that perception persists, and if everyone, including Waymo, is at risk. Now, I believe Waymo will be persistent and you know, survivor and successful ultimately, but if Argo existed today, it would be subsidiary to Waymo in perception, in the same way that Waymo is subsidiary to Tesla in perception. And that is a tough place to be. That being said, the other dimension that people rarely talk about in the autonomous vehicle world is if companies A, B, and C all possess equivalent self-autonomous capability that the stack can drive safely and they possess um uh you know sufficient funding to compete in the market, you then have the the the other two wars to fight, which is uh who has the best set of strategic partnerships in the right markets. Because you need to, you know, looking at an Uber demand map, you look, you can see the top 10 cities in the United States. In the AV world, your strategy must, you must possess or at least uh command and get to profitability in uh key markets in New York City, Miami, San Francisco, Los Angeles, you know, Boston. And you have to get in early. Why? Because an AV company that deploys a an owned fleet must have real estate for the vehicles, it must have partnerships with the key stakeholders in the communities that are non-elected and persistent, and then it must have relationships that uh with both political parties that can transcend an electoral loss because you want to make sure you, if you invest a several hundred million dollars in a city, you're gonna be there 50 or 100 years, you're gonna be there forever. And you cannot be at risk of losing that city because there's a new election and such and such, you know, puts in a permitting regime, which then kills your business overnight. And this has happened before in the scooter business of San Francisco. So the strategic and political dynamic of deploying a Navy fleet uh at scale in urban centers, uh, absent what I just described, is in at risk. Argo, you know, very wisely went to Miami when no one else did. And I and Waymo just arrived there, just the last recent months. So that would have been interesting. And so you can see Tesla is out there now, but they're only operating in a couple of places. And Waymo has announced dozens of cities. So Waymo understands the strategic political game very, very, very clearly. Uh and then, of course, the other dimension that Cruz did not get, and Waymo does get, is the operational depth of bench, which is again two axes. Depth, how many people do you have per vehicle? Can you invert that safely? And then quality. Like, no matter what you're doing, is the ride experience actually reliably, consistently good? And so this is like a six-axis game, and the media is still usually talking about one, which is does it drive at all? And then maybe, and when do we pull the drivers? Like, completely oblivious to like the audience that you have on this podcast of engineers are solving everything else except a strategic political dynamic, which is really the one time the policy team better be good.

SPEAKER_01 9:12

Well, there's the policy team and then there's the technology. And maybe just a few words. We'll we won't talk too much more about Argo, but since uh we've been there for the last few minutes, what are the key ways you think that AV technology has evolved in those few short years since the end of Argo and where we are now?

SPEAKER_00 9:30

Well, I mean, the conversation has moved from can it work? Because we've seen that it can. We know like Waymo is deployed driverlessly. I ride them several times a week. It's a terrific product. And we know we can operate in complex scenarios. It does so routinely in San Francisco and Los Angeles and here in Scottsdale. Um, although complexity in Scottsdale is if it if it rains, it's really twice a year. Otherwise, it's a pretty simple domain. Uh, but the the the two other evolutions are first, um I mean, we have seen extraordinary evolution of Tesla's FSD, and you know, which is vision only. And I know I I'm not convinced that the metrics they've released publicly are sufficient to demonstrate safety uh equivalent to Waymo, but I am convinced that that they will get there. Uh, you know, recently I drove cross-country, or I did not drive cross-country in my Tesla um on FSD 100% of the time, uh, you know, uh 3,000 miles through through snow and rain, and it was it was flawless. It was great. And so Tesla will get there, but it that's how long it takes is another issue. So vision only can work, but there will be for a time, and it could be many years, some delta between the best vision can do and the best a multi-sensor stack can do. Uh so the war, whatever war there is in AV, is can Tesla or the vision only companies get to a level of safety equivalent to multi-sensor at its best, while the multi-sensor companies have to drive the costs down of the hardware such that they can deploy them in passenger and privately owned vehicles. And one of those two, someone's gonna, whichever side accomplishes their very different goal first, is going to be a very powerful company. But again, Waymo and Tesla are both likely to be permanent players. Uh, it's everyone else that we don't know. Now, the other thing, the the elephant in the room in the evolution of the technology, has been that, you know, it we could say pulling the driver is not a technical uh you know achievement, but it is because it means the technology is sufficiently safe to pull the operator, a driver, or supervisor. But the caveat is this the entire sector made a catastrophic mistake seven, eight, nine years ago, which is which is everyone knew that there would need to be in-car monitors or supervisors for a time, and that when they were pulled out, there would be remote supervision. And then eventually remote monitoring, because I think that they're a little different. Eventually the ratio will flip. So it's you know, one monitor for dozens of cars, who's only looking where there's an issue. But the industry, the entire industry, well, actually, not Tesla, but everyone else, Waymo, Argo, Cruz, everyone on earth allowed by omission of explanation of what remote operations would look like, allowed the perception to grow in the AI hype cycle that when in-vehicle supervisors were pulled, there would be no one on the other side. Right. Right. And so recently, uh, you know, Waymo has been very unfairly accused of keeping the big secret that they've got people in the Philippines, remote teleoperations, operations, as and now it's being cast as if that is a technical failure and that it's that none of it will ever work, and that there are remote drivers. Now, remote drivers and teleoperation and monitoring supervision are all very different. But the failure to explain it back in the day, and and and by not explaining it, perpetuate, you know, building the AI, you know, myth, um, has is is they're paying for it now.

SPEAKER_01 13:08

So you mentioned driving across the country 3,000 miles. Uh, let's talk about that now. That has been known as the cannonball run, and you've uh you've set the cannonball record. Just talk about what that let's talk a few couple of minutes on the history of cannonball and how it came to be, and then your ride in 2006, which led to the film that you produced as well on your drive.

SPEAKER_00 13:32

So back in 1915, uh the American car companies, Cadillac, Stutz, Gartner, uh, Ford, uh, you know, they uh they hired uh a guy named Erwin Baker, uh, to he was in a motorcycle racer uh as a to drive cross country and set records as fast as possible. Why? Because you know, you have a new product category, personally owned internal combustion cars that replace horse ownership or rental and train tickets. And they wanted to demonstrate the capabilities, you know, the the unlock, the freedom unlock. So he goes cross-country, no gas stations, no interstates. And so he's buying fuel at hardware stores and pharmacies. And every incremental improvement in fuel economy, in gas tank size, in vehicle weight, in road surfacing conditions, every little improvement generated incremental time uh cuts. So his original records weeks, went to two weeks, went to one week, went to five days, and he set hundreds of records. And so uh, in time, you know, uh companies like uh, you know, the car companies and even Ford debated verticalizing and building their own uh gas station networks. They did not, but Coca-Cola saw how much coke they were selling in pharmacies and hardware stores and began um you know investing in and partnering to build that gas station network, which is why there's so many Coca-Cola antiques today. And so, you know, 60, 70 years later, Brock Yates from Car and Driver Magazine said, Well, you know, why don't we just do a race? We'll name it after Irwin Baker. His nickname was Cannonball, we'll call it the Cannonball, you know, the the the Irwin Baker Sea to Shining Sea Memorial Trophy Dash, uh Cannonball Baker Dash. And so, and that became the Cannonball Run movie, which inspired all the rallies today. So in 2005 or 2002, you know, I was aware of this, some of this history, and I wanted to do and break this record. But for me, it was always like a quantitative and technical challenge because anyone can hit the gas puddle and drive fast.

SPEAKER_01 15:20

All of this is meticulously documented in your 2019 film, Apex, The Secret Race Across America. Let's fast forward 20 years from 2006 to 2026 and your recent hands-free drive across America following the same route.

SPEAKER_00 15:33

Yeah. The path to the modern cannonball with Tesla and FSD is really inspired from uh the Nurburg Ring. You know, every OEM sends cars to Nurberg Ring, even a family car like goes to the Nurberg ring and sets time because performance specs are now part of marketing every vehicle. So, what would be the Nurberg ring of autonomy? You would need a long route called 3,000 miles long that is urban and rural and highway. It's fast and slow, it's complex and simple, and it has vastly differing weather, altitude, and lighting conditions. And I'm like, wait a second, why not just take go across in a Tesla? You know, Musk promised, you know, many years ago, coast to coast drive uh in a Tesla would be possible driverlessly. Let's do it. So I prepared my Model S and I installed everything from my old Cannonball BMW in the Tesla. Okay. And um and began a year ago uh trying to do this. And we made four runs with four successive generations of Tesla FSD software and two versions of 12, one version of 13, and then uh FST 14223, which just came out. And so and I brought two Tesla owners with me because a key aspect of this is you know, if is having someone in the car who's trained the way a Waymo or Argo safety operators trained, they will monitor the vehicle's you know function and and take over to prevent a crash of the vehicle is going to make a mistake. You can't let it the scenario play out. So I brought two friends, um, Paul Fam and uh Warren Honor, who's uh an Oracle executive, Tesla owners, and we went across a few weeks ago just before a major storm. Now, in the previous drives, we'd had disengagements as high as 32 across the drive. And then the second drive, it was you know 21, and the third drive, not good at all. And so I was not optimistic, but I'm like, we have to try. And so we did go. And the only I mean, the miracle of the drive is that it functioned even in snow. Because previously FSD would disengage in rain, and that was a big surprise. And then the other uh not surprise, but it was an incident, is that the vehicle does you know does not have self-cleaning cameras. Yeah, and so if you want to deploy an AV, you need to be operate, need to operate in four seasons, 24-7. And so, you know, I learned from previous drives that if you light rain and mist on the cameras, the system will disengage. On this drive, it didn't. And we were we're shocked, shocked, until we got to supercharging locations in the eastern half of the country.

SPEAKER_01 18:05

Well, we'll talk about that in a moment. I I want to talk about the uh the kind of challenges that you face, but what were the rules that you set yourself? What were the kind of guidelines or ballpark figures that you set to say, I've managed this or I've achieved this?

SPEAKER_00 18:21

So, in order to declare a record, you have to have a clear set of rules that people can understand so people can follow you if they want to duplicate or beat what you did. And so in our case, it was um there would there would not be any human interventions to prevent a crash. Okay, and if there were, then the vehicle uh we would be disqualified from claiming a record. And so uh, and if the vehicle had what you call an involuntary disengagement, which means it forcibly just shuts it off because it can't resolve the scenario, there's no record. And so, you know, we would classify all the disengagements in a very granular way, because you might have, you might have the system might just make a navigational error and we would let it play out. Um, it might also do something illegal, which we would not let it play out, which would disqualify from our record. And so if you look at the way like Waymo or Argo classify disengagements, each disengagement type then has a team like all over it trying to resolve that. Um, and so we just duplicated that method because it's the safe method, it's also the honest method. And so uh we had uh on our drive only one incident which it would classify as driver error, which is I was telling a story. I was telling a story about how I failed Argo safety driver training. And as the car was doing something, I just grabbed the wheel to demo, and I'm like, guys, and so uh, you know, we're we were fully transparent about it. The vehicle did nothing wrong, uh, nor uh nor did I intend to take over. So we classify that as human input error. And the funny thing is, Elon Musk's said, you know, all at a certain level of technological progress, all human input is error. I disagree with that, but not completely, because in our case, we had many instances where I changed the route on the GPS. The car did its job perfectly. Every time I tried to optimize the navigation in the car, it costs us time. If we had sent the car empty across country and just had people at the charging stations to plug it in, the car would have gotten there eight hours faster. Really?

unknown 20:30

Yeah.

SPEAKER_00 20:31

And so Musk is right. Like at if if you all our input was error. The only exception in the future will be if you want to drive for pleasure, in which case all input is correct and everything the car does is error. Uh, but if you're driving, you know, for as a commuter, eventually, yeah, you won't need a steering wheel. And why would you want one unless you have a dual mode, dual mode car?

SPEAKER_01 20:54

So we talked about the rules there.

SPEAKER_00 20:55

Describe the vehicle and the technology you had on board. So uh I have a 2024 Model S. You know, we have uh three iPhones, uh, one running on an app called Highway Radar, the other ones run ones runs Waze. The third uh we would use to manage social media and monitor a competing team that was going at the same time. Okay, right. And um the iPads were were for entertainment in the back. And uh the car also had a uh a FLIR, a Telodyne FLIR thermal camera mounted on the front, um, which we didn't use so much on this drive, uh, but was critical in prior drives because we were very concerned about animals. And um, other than that, a Starlink mini mounted uh to the roof, and um that might be it. Because Tesla FSD maxes out at 85 miles an hour. And even if you set it to its fastest mode, which is called Mad Max Mode, it defaults to about between 79 to 81 miles an hour, which in most places is legal in the United States. Um, and you're in the flow of traffic in places where the limit is 75. So that was not much of an issue. Now, a very interesting thing I rare people rarely talk about is Tesla FSD 14224 is the current one, allows you to shift between five different driving modes. The driving modes have a different delta between the speed limit and where how fast it will drive. And also it's lane-changing behaviors and aggression. And so I routinely shift modes dynamically, minute to minute, based on conditions, because I'm just familiar, I'm just familiar with how it works. And there is probably some safety delta there that we do not have that the public does not have data for, but I'm sure Tesla has it. Um and that I think the future of all personally owned vehicles that are capable of higher levels of automation will eventually have to have these modes. Because if you set it to the most aggressive and fastest mode, which some people will love, um if that's the default and you can't adjust it, many consumers will not like to ride in that car. And so that's a very important UX uh aspect for the future.

SPEAKER_01 23:01

You mentioned supercharges, and I wanted to ask you about how you planned the route. Did you plan the route around charging? And did you work out when you might need to charge in advance?

SPEAKER_00 23:11

So interesting. You know, on the original cannonball, the gas one, we I went out and measured the fuel uh transfer rate at different pumps and stations because it's essential. It can be a difference of you know, 30 minutes could become 60 if they're you know going to the wrong stations. So, you know, Tesla is interesting because the superchargers, number one, that they exist, you know, is that was a key catalyst of the growth of the Tesla as a company. And they began, uh, they were always called superchargers, but the speeds have increased. The original ones were, I think, 120 kilowatt chargers 10 years ago. And now the default is 250 kilowatt. And they have some that are 325. Yeah, but they also have some that are 72, and they're all called superchargers. If one selects a destination, you know, in L you're in LA and you select New York as your destination, the the Tesla will pre-map the entire route with uh charging stops. And you have two options fastest or best amenities. And the amenities are bathrooms and bathrooms and food. And so uh the fastest is gonna have more charging stops of shorter duration. And best amenities is I don't know, it's a mixed bag. It's fewer stops. Actually, there's a that's sorry, there are three options. There's fewest stops, amenities, and fastest. And so a fewest stops is fewest stops, and they're just longer. And then there's the middle one. And so we just left uh on uh using the fastest, um, the fastest route in most chargers. So because of the nature of the Tesla battery pack, that means you're arriving in the you know nine to twelve percent range, you're charging to 45 to 50 percent, and you leave. Uh that you know, that that's the fastest way to go cross country. Now, as we were going, you know, Tesla has a real-time and three-hour forward weather prediction uh map uh overlay of the GPS. So we observed this storm coming uh on the East Coast. We were heading west to east, and we could see two storm fronts, and our route was passing directly between them. And as the storm was evolving, and I feared we would be stuck in one of these storm fronts and fail the entire drive, I began changing the route. And as I changed the route, it would dynamically change all the charging after the fork I had selected. And that uh cost us that was not apparent to me as I was doing it, but it it cost us a lot of time and led us to a very hilarious and unfortunate incident. Which was what? Can you tell us? We were the the navigation threaded us uh down a mountain road that was not on the highway. It was still listed as a highway east of Pittsburgh, but it was not a major highway. You know, it was not two lanes in each direction, it was one lane in each direction, divided road. And we're going downhill, and we have you know brand new Michelin all seasons, and the there's snow on the road, but it's not yet ice, it's slush, and uh the car is shimming a little bit, and its car is going 50 miles an hour. And I I lower the mode to sloth mode, the slowest FSD mode, and it's still going 50 miles an hour. So we we choose to pull over, and um, I select a random, you know, two o'clock in the morning, like a tractor supply shop uh in Pennsylvania, and the car slows down. I'm I I can't believe it slowed down safely and pulls over into the snow in this parking lot. We're surrounded by tractors for sale and black Ford Economine vans with blacked-out windows with engines running. And we get out, we get out to stretch our legs, and there's this smell of burning chemicals. And my co-driver Warren says, We probably should get back in the car and leave right now. And a man emerges from the darkness and says, Hey, bro, are we trespassing? And I'm like, Are we trespassing? And Warren's like, get in the car, we go. So we get in the car, and Warren says, This is a mobile meth lab parking lot, like breaking bad. This is really bad. We have to go. Engage FSD now. So I engage FSD, it's hoping the car will turn around and go back up the road. And instead, it advances through the snow at one mile an hour in between the meth lab burn the meth lab trucks. And but we can't disengage because we're here for an autonomy record, and we just advances and then makes the left and goes back on the road and we we escape safely. So yeah, there are still interface options and like functionalities for autonomy for personally owned autonomy that don't exist. One of them would be pull over now, bathroom break, emergency bathroom break now. Like, you know, you know, uh, because I mean if you had a steering wheel, you could just do it. But if you have if you have a vehicle, a personally owned AV with no steering wheel, you have to pee. It's like stop now, emergency stop. And uh the other one would be like emergency turnaround, because you need to be specific. Like, we're headed towards something I can't do the car doesn't identify as bad, like a volcano is erupting, turn around now. So that was the first of two incidents. Well, in this case, the FSD worked perfectly. You just didn't know who our neighbors were. So you mentioned two stories there, Alex. What was the other one? Well, the other incident points to uh shortcomings in the current hard uh Tesla Hardware's tweet. So if you want to, and again, a navy fleet operated or personally owned has a run in four seasons. And uh for many years, Waymo and Argo and everyone else has clean self-cleaning sensors, whether you know it's a wiper blade or air or something. And so if you want to cross country, you've got to charge your Tesla. And you know, in good conditions, the Tesla will back into a supercharger or advance into a supercharger and uh perfectly. Uh, but in rough conditions, uh, if the backup camera is obscured, it won't. And so we learned very quickly as they went into the weather that the car wouldn't back it all the way into the charger. So the uh the solution, you know, it would just declare itself arrived and you'd be six feet away. And so one of us would get out and then stand in the stall and we'd re-engage the system, telling it to park, and it would move over at one stall, and we'd have to do this three a couple of times till it parked. But in one specific incident in Breezewood, Pennsylvania, there's a truck stop. And the way it is set up is the Tesla supercharger is in the middle. And the truck stop is a like a like a record player. It's a single road that spirals out, and it's a one-way. You go, you go into the middle, and then you drive out. So we we arrive in the Tesla, we trick it into parking away from the chargers. One of us would get out, clean the camera, and then we'd engage it and would park in the stall. So my code driver, Paul Fam, gets out, he cleans the camera, he's outside the vehicle, and we tapped engage. And the car did not back into the stall. It advanced back into the flow of traffic, following the trucks back onto the interstate. And we need to charge, and we've now abandoned our code driver. So he's yelling and waving, please come back. But we're here for a record, so we can't disengage manually. So we we we we go in the map and start selecting POIs that are in the truck stop so we can go back and get them. Anything we selected, including the location we were currently in, the ETA was an hour and 41 minutes because the next exit is 50 miles east and FSD is obeying traffic law. And so we just called, we said, bro, just find shelter, we'll be back for you. So we drove uh 40 minutes east, 40 minutes west, charged at another location where we had to clean the camera and trick it again, and we went to get him. And miraculously, as we arrived, we really feared the car would not get him, but it did.

SPEAKER_01 30:40

So your cross-country time of 58 hours, 22 minutes includes 90 minutes or so of rescuing one of your co-pilots.

SPEAKER_00 30:47

Yeah, more because you had to add an extra charging stop. You know, the human-driven Tesla cannonball record is around 42 hours, and this is 58. So that difference is a function of two things human error and temperature forcing additional charges. And the human error is worth at least four or five hours. So this record will be broken by someone when it gets once the weather gets better, but who does not introduce human error into the process. And I estimate they will cut they will cut probably eight to ten hours off the time.

SPEAKER_01 31:26

So let's talk about driver monitoring because obviously you are a uh a driver who's not driving the vehicle. And I guess this gets into terminology a little bit as well. Of uh, you know, we've already during this call used automated and autonomous, and somewhere there's self-driving, right? Um the the vehicle is doing the driving, but you're still the operator. You're not asleep in the back, you're being monitored by the driver monitoring system, so you have to stay alert, you have to be ready to take over at any time, right?

SPEAKER_00 31:56

So, yes, I mean, this is according to SE levels of automation. Uh this is a level two, remains a level two vehicle, a very high capability level two vehicle. Uh, and so there's a human in the loop, you got to sit in the seat, there's a seat sensor to measure your weight, there's a torque sensor on the steering wheel, and there is a camera uh, you know, DMS with a camera above the rearview mirror. Now, it is so from a classification standpoint, this is a level two vehicle. Now, from a capability standpoint, it feels very like a level four vehicle. Uh, and I know from experience, because I drive my Tesla every day, that I have a certain pair of sunglasses that will successfully block the camera from observing my eyes. However, it can still observe my head pose in the daytime. The Tesla DMS camera will allow you to have your head pose down towards the camera. If you're wearing sunglasses, that will block your gaze. At night, the Tesla DMS will not allow your head pose to go down. It will not. And so they know from their data that um people are doing this over time, uh, and that it's in the daytime it's mostly safe, and at night it's mostly not. Uh, because there's no other explanation for why they would the constraint would be different in darkness than in daytime. And so I'm uh amazed how good the system is. And based on my personal experience in it, I will I may look at my phone and I may do things in the daytime, depending on context, complexity, bystanders, traffic. Uh, I suspect there are a lot of people who have no idea about the risk they're taking who are doing much, much worse all the time.

SPEAKER_01 33:37

Driving coast to coast is a significant human challenge, as we've discussed, and uh it takes a certain kind of person with a unique set of skills, personality, and mentality. I'd say very few can do it. When you set your record in a combustion engine car, you are very deliberate about the choice of car and the equipment on board. Fast forward from 2006 to January 2026, when you did this in a specific Tesla running a specific version of its FSD software. Can you envisage doing this in another brand of car or a different operating system?

SPEAKER_00 34:08

Yeah, I would love to be able to do it in a different car. It just so happens that Tesla is the only OEM today that offers a system this good outside of China. Uh, and I don't know which of the Chinese OEMs offer a system this capable. Uh, I suspect there's maybe someone. At some point, someone, another vendor will, it could be Wave, uh, Mobileye, or maybe it's Waymo, will offer such a system. They'll license it to an OEM. You know, for Wave and Mobile, it'll be fairly easy because they already operate on hardware that is conformal to current passenger car design, which is beautiful. Well, not always, but it it'll work. Uh, Waymo's job, because you know their stack is very advanced, is to minimize the hardware form factor to fit into passenger cars. They have different tasks ahead of them. I am absolutely convinced that there'll be autonomy available. Driverless capability will be standard on 100% of vehicles in the future. It won't all happen in the same place at the same time for everyone, but that's that's the future. My dream is that vehicles are dual mode. So we'll have steering wheels and driverless capability, so we can send the cars out, you know, empty, or we could sleep in the back while it's driving around, or we could drive them inside a safety envelope that is boundaried by what the autonomy stack determines is safe. Because I'd like to, if ideally at Porsche 911, you know, I can nap on the way to the track, I go around the track, and behind the track afterwards is a country road. I want to drive it the way I want to drive it. But I would still like to have the autonomy stack replace the traditional ADAS. So as I'm approaching the boundary, the system might prevent me from having a crash. But inside the boundary of allowable inputs, um, my outputs are awesome, awesomely fun. And I think that you know, driverless um capability must be developed first, after which we can begin looking inside the safety envelope for how can we push these boundaries out to make sure human driving is great? Because people will pay for that.

SPEAKER_01 36:08

I'll pay for it. Yeah, you talked about sleeping in the back. When do you think someone would be able to do the trip while asleep in the back? Or even you mentioned earlier, when do you think we might be able to send the vehicle without anybody on board at all?

SPEAKER_00 36:21

Uh I think Waymo could send a person cross-country now. Uh, I think they absolutely could. Uh in maybe in bad weather, but certainly in good weather, I think Waymo could do it now. That they haven't is probably a function of it's just the ROI isn't there on the marketing. Like, do they have to? Like, how many people care about the Nurbergering of autonomy? How many people truly care about the Nurburgering? It's a subset of sports car purchasers, but it's it's the default setting for the marketers. In Robotaxi, um, it's not necessary. And now the good, here's the good news. My daughter's been in a Waymo, she's seven, and she's been in my Tesla with me. She already believes these are safe. By the time she's old enough to ride in one alone, they will be. And that was what happened with aviation. And it's happening in autonomy and automation at one-tenth a timescale. The trust is baked into the next generation of passengers. And so I hope that the engineers out there, but especially the marketers, will focus on making the experience great. Because by and large, I think the safety debate is going to solve itself. And it's solved by cultural acceptance, which is we're approaching that already.

SPEAKER_01 37:32

And on that, a great moment to wrap up. Alex Roy, thank you for being a guest on the Beyond the Car podcast. So, can a car drive itself hands-free across America? It seems the answer is yes, but that comes with certain caveats. Alex completed the drive in a 2024 Tesla Model S on full self-drive mode, a system that Tesla calls FSD supervised, and notes on its website that currently enabled features require active driver supervision and do not make the vehicle autonomous. That means it requires a human driver to be ready to take back control at any time. This was an independent project, and while Alex himself acknowledges flaws in execution, the completion of the challenge raises interesting points for discussion and highlights incredible capabilities as well as notable vulnerabilities in the state-of-the-art technology. What stood out for me was the impact that humans made on the challenge and Alex's view that without humans on board, the drive would likely have been shorter by 8 to 10 hours. That of course raises the question, so what? But as Alex said, like the Nurbo Green test for a family saloon car, the point of this was to set a benchmark time, whatever that might be and however it might be used. And Alex ends on a really positive note about the future of automated driving, as seen through the eyes of a child. And when I was a kid, automated vehicles were still the stuff of science fiction. Alex's younger daughter rides in them as if they were the norm everywhere. It seems we're only limited by our imagination, our technical ambition and achievement. Thank you to Alex Roy for being a guest on the show, and thank you to Martin Strong for producing this episode. You can learn more about Alex Roy's company, New Industry Venture Capital, at nivc.us. You can learn more about Fisita at Fisita.com. And if you'd like to be involved in this podcast or any other Fasita activities, reach out and get in touch. And finally, thank you for listening to this episode of Fisita's Beyond the Car podcast.