Smart Driving Cars episode 402 with guest Alex Roy
Annotations will appear as you listen
0:00
56:59
Brand
Tesla
Tesla is a company that makes electric cars. They are well-known for their advanced technology, especially in self-driving features.
Brand
General Motors
General Motors is a big car company that makes many different types of vehicles. They are known for brands like Chevrolet and Cadillac.
Concept
self-driving
Self-driving cars are vehicles that can drive themselves without needing a person to control them. They use special technology to see their surroundings and make decisions while driving.
Concept
driver assistance
Driver assistance technologies help drivers by making it easier and safer to drive. They can automatically brake the car if it senses an obstacle or help keep the car in its lane.
Brand
Mercedes
Mercedes is a well-known car brand that makes luxury cars. They are famous for their high-quality vehicles and advanced technology.
Concept
autonomous vehicles
Autonomous vehicles are cars that can drive themselves without needing a person to control them. They use technology to see the road and make decisions.
Concept
semi-autonomous
Semi-autonomous cars can help you drive but still need you to pay attention and be ready to take over if something goes wrong. They aren't completely self-driving yet.
Concept
DARPA Grand Challenges
The DARPA Grand Challenges were competitions to see who could create the best self-driving cars. They helped push the technology forward by getting universities and companies involved.
Company
Google is a big tech company that created a self-driving car project. They worked on making cars that can drive themselves without human help.
Term
autopilot
Autopilot is a feature in Tesla cars that helps the car drive itself in certain situations. However, the driver still needs to pay attention and be ready to take control if needed.
Term
driverless
Driverless means a car can drive itself without anyone inside to control it. It uses technology to see the road and make decisions just like a human driver would.
Term
semi-automated driving features
Semi-automated driving features are tools in a car that help the driver, like keeping the car in its lane or adjusting speed. The driver still needs to be in control and watch the road.
Company
Cruise Automation
Cruise Automation is a company that works on making cars that can drive themselves. GM bought this company to help them with self-driving technology.
Car
Audi A4S
The Audi A4 is a car made by Audi, a German company. It's known for being comfortable and having good technology features.
Company
Waymo
Waymo is a company that works on self-driving cars. They started as part of Google and are now known for their advanced technology in making cars drive themselves.
Term
Uber ATG
Uber ATG was a part of Uber that worked on making self-driving cars. They wanted to create cars that could drive people around without a driver, but they had some problems and were sold off.
Company
Aurora
Aurora is a company that focuses on making self-driving technology, especially for trucks. They bought a part of Uber that was working on self-driving cars to help improve their own technology.
Term
RoboTaxi
A RoboTaxi is a self-driving car that you can use like a taxi, but it doesn't have a driver inside. It uses technology to drive itself and take you where you want to go.
Term
automatic safety system
An automatic safety system helps keep you safe while driving by noticing things around your car and warning you or stopping the car if there's a danger, like a person or object behind it.
Term
sensors
Sensors are little devices in cars that help them understand what's around them, like if there are other cars or people nearby. They help keep the car safe while driving.
Concept
cannonball run
The Cannonball Run is a race where drivers try to get from one side of the country to the other as fast as possible. It's not an official race, but it shows how fast cars can go.
Term
AVs
AVs are cars that can drive themselves without anyone controlling them. They use technology to see where they are going and make decisions like a person would.
Car
Jeep Wrangler
The Jeep Wrangler is a tough, off-road vehicle that can handle rough terrains. It's popular for outdoor adventures and is designed to be driven in places where regular cars might struggle.
LIVE
You have found us again. Welcome back to the Smart Driving Cars podcast. I'm Fred Fishkin,
along with the Faculty Chair of Autonomous Vehicle Engineering at Princeton University,
Alan Cornhouser. Hi, Alan. Hey, good morning, Fred. Good morning. Well, it's been a very
busy autumn. The semester is in full swing there at Princeton. And you're back from the
Florida A.V. Conference. And what you heard there prompted the invitation of our guest today,
Alex Roy, general partner and co-founder at NIVC, co-founder of the Narrative Command,
the drive, Autonicast, and Chairman Emeritus at the Moth. Great to see you again, Alex.
Great to see you, Fred. How are you? Hey, Alan.
Yeah, great to have you, Alex. And again, you know, you basically led off the Florida A.V.
Conference. I guess this was the 13th version. I don't know. I think I've been to almost at
least 11 of them, if not while we're 13. You've been the most. But I just thought your presentation
was really, I guess, you know, I'd like to maybe say A.V. 101, the things you should know about A.V.s
in 2025, because we are, you know, at a different time point than we've been, certainly over the past
13 years of A.V. And it was so nice to have Jeff, who I guess he is the godfather of A.V.s in Florida,
because of the legislation he introduced, I guess, not 13 years ago and so on in Florida.
That's a lot of Florida. There'd be a strong participant in this. And I must say, you know,
those of us here in New Jersey, we're still trying to make things happen. We did have a hearing
in the New Jersey Senate on Monday in which a bill is finally coming through in New Jersey,
so that New Jersey might at some point be able to participate in all this,
not to suggest we're at least 13 years behind Florida or whatever. But if please, I'd like you
to go through some of the things, some of the comments you made there on the stage as you let
off the Florida A.V. conference a week ago. Well, I'd love to. Let's pull up my slides,
and we can march through them, because I'm sure we'll have a fun convo, because we can stop at
any of them. The three of us could just tell funny stories. Absolutely. But they are substantive.
So I've always been fascinated by how companies, when you go to their website,
they tell the history of self-driving. And it's always like, well, there was a DARPA challenge
in 2007, and then we showed up. I was like, that's not the history. And so if you look at a cross-section
of all the different companies and put their timelines together, you still don't see a full
picture. And so Wikipedia and other sources are okay, but they're often edited, often by players
in the industry, to remove things that they don't like. And so there was no single point of truth.
So I put this little deck together for the conference. And so I was guided by my theory
about narrative command, which is my filter for investing out of my venture capital fund.
Different companies try to tell their story, but one company in every sector ends up dominating
the narrative of that vertical. And in self-driving, it's Tesla. It doesn't matter what you believe
about Tesla. Or you think about Tesla, they own the language and narrative. And if you ask young
people, they'll think Tesla's self-driving today. Anyway, this is my period of narrative command
I use when I went on ranked companies for a narrative hour. And so there was a wonderful
historian of science named Ludwig Fleck. I learned about him when I was researching the
history of elevators that you and I have discussed in the past, because elevators are very analogous
to self-driving and as are the narratives about elevators in the late 19th century. And so he
said the straight path of progress is a lie because every company that's successful, basically,
I want to call it straightwashing, but they remove all complex elements and failures from
their timelines. And Otis Elevators is a great example. Today, no one knows that Otis himself,
the original founder, when he died, the company was a failure. His sons had to buy it back.
There was lots of twists and turns, bankruptcies and financial issues. And eventually they did
M&A and rolled up the American industry. But that's another show. Anyway, there is no
straight path to progress. There's always issues and companies die and go away. So with that in mind,
put together this little history. So sometime in the 16th century, Da Vinci designed a
self-propelled cart. He never built one, but a few years ago in Italy, an engineer did build
one according to the design and it worked, which is incredible. Bless his heart. In 1925,
there was a remote-controlled car drove through New York City called the Houdina.
And with mixed results, it was quite cool people saw it. But it ended up having a crash and striking
a vehicle full of journalists who were taking pictures of it. And then in 1939, there was the
New York World's Fair. There was the Futurama exhibit put on by General Motors, which showed
the only cool cars driving through city streets. I actually think that maybe the first real milestone
in self-driving was the Knight Rider TV show in 82, because this was the biggest, best and most
popular depiction of an autonomous vehicle, probably in history, maybe to this day. Everybody
who came, I was 11 when it came out, but anyone between the ages of, you know,
eight and 25 probably saw this show. And 100% of people who have now aged into running the
company today saw this show as a kid. And you had voice interaction, you had full autonomous mode,
you had full driver assistance that would prevent you from crashing, even if you wanted to,
you had, you know, connectivity. I mean, you literally everything that we're talking about
was in KIT. And more importantly, there was a bad version of KIT called CAR. And CAR was a failed,
I think it was a failed prototype out of like an earlier program, who somehow broke loose,
and he was out there always trying to, trying to defeat KIT. And so all the scenarios we talked
about, good, good AI, bad AI, they're all in the show. And I think everyone involved is literally
a genius stuff. But then you get into like an academic who, an engineer who is completely
unappreciated today. This is Ernst Dickmans. He, back in 86, I was working with Mercedes and
built several autonomous vehicles that drove on the Autobahn and through cities.
Alan, do you know Dickmans? I had heard of Dickman, yes. And in fact, you know,
out here at Sarnoff Labs, which aren't too far away from here, maybe, you know, a mile and a half,
they actually had a test track in the fifties. But that was, you know, sniffing a wire in terms
of doing the lateral control piece of this thing. And I think there's still remnants of the darn
thing out there at Sarnoff Labs. And yeah, there are these guys out there that really did this,
but you know, it's, and they deserve an enormous amount of credit because of the vision that they
had at the, at that time. Yeah. I mean, this guy, why is it like on the speaking circuit,
getting a fortune? I don't know. Like, it's just, you know, maybe he's too modest. I don't know,
guys are genius. And everybody I know on the industry side of Brian Celeski was the one
who ran StackAV today. He's like, yeah, Dickman is the guy. Moving on. The second most important
depiction of a time in these vehicles in pop culture history, total recall. This is the Johnny
cab. This is a kind of an interesting iteration because it's an autonomous, it's a rubber taxi,
but the driving seat is occupied by an upper torso and head of a, of a driver that is already built
into the car. So this is completely unnecessary. But I guess in the theory behind this is that
you need a human appearance. You need a human similar physical avatar in the seat to for,
you know, I guess passengers who feel more comfortable with a humanoid in the seat, even
though the humanoid itself, the torso is not physically doing the driving. If there's the AI
is just like the facsimile of a human there and not even a good one. But the other part is this
wonderful depiction is that it, the Johnny cab, it, it's really annoying. It's really,
really annoying. And it's interface, voice interface will like talk back and like, you know,
disagree. And so this, it's actually an example of two major design errors in the Johnny cab
product is you don't need to take up the space with a humanoid torso. And you really want to limit
its ability to interact with the driver through with a password group voice interface. And in
this case, the Schwarzenegger had no way to turn it off other than to rip the humanoid torso out
of its, its driving seat. Wonderful, wonderful, wonderful movie. Well, if I can, if I can jump in
yesterday in my class, I had Chen Yi Chen, who was one of my former students who is, I don't know,
the lead designer close to the lead designer with the X-Pang, who last week announced their
whatever version two of the humanoids. And of course, you know, what I'm trying to do is to
get from Chen Yi a, a, one of these guys to actually put in the driver's seat. But of course,
not to drive, but to be the chauffeur, and of course provide chauffeur functions. And so what
kind of chauffeur functions you need to print? How about opening the door for me? How about lending
a hand to help me in and out? And in some sense, you know, the driverless cars, as long as we have
good legs, and as long as we're really mobile is just fine. And we really don't want anybody else,
just like we get in and out of elevators very easily by ourselves, you know, but, you know,
for older people, for, for, let's say, you know, the folks that have had their keys taken away from
them and really need this high quality mobility. Wouldn't it be nice if we had, you know, one of
these humanoids who was then trained to basically assist them in and out of the vehicle, assist them
up the steps. And then of course, if you get pulled over by the cops, I guess the thing, you know,
is able to do with the large language models and all that stuff, at least, you know, talk their
way out of the situation. I don't know. Talk to me here, Alex. Is it, is it a realistic concept?
Or is it just crazy? Well, okay, so I have long said that many of the, so right now, if you get
in a regular taxi, yellow cab in New York or an Uber, you, there's a continuum of experiences
regarding what the human driver does for you. So in a yellow cab in New York City, there's no,
that guy's never getting out of his seat. He's never going to help you ever. Maybe if you go to
the airport, he might, but it's, it's based on their mood with Uber and the rating system, you
know, that encourages the drivers to be much more helpful and interactive for the passengers,
which is great. Although I, there's no, I think it's pretty, I mean, rare for older people to
get groceries in an Uber and then expect the driver to help them. And also you would need
to really help from like door front of the supermarket to the vehicle, which, you know,
I've never seen that occur in an Uber, but there's certainly a higher level of customer experience
expected from an Uber driver in certain scenarios. Now, when I was at Argo, and we talked about,
you know, we talked about two things a lot, which was the fear of job loss to autonomous vehicles,
robotaxis, as well as, you know, how to maximize the customer experience. And I thought I,
five, six years ago, I said, look, it's very obvious to me that if you have an autonomous vehicle,
robotaxis, well, I don't call it robotaxis, an autonomous vehicle, whether it's in a taxi
capacity or an interacting or replacing known forms of transit, that you have an opportunity to
create experiences for users that are incredible. And a great example of this would be there's a
company called Get Papa out of Miami. And it is basically Uber for help for older people.
Like, oh, I need someone to help with my bags, but it's not related to cars. I need an assistant
for the day. I need someone for three hours. I need anything because I can't do it. This
app platform enables that. So I was thinking about my mother at the time, who, you know, was
unable to take care of herself. I thought, wow, if I could get an AV and I had Get Papa as an
option and the Get Papa app and the Waymo app were merged or the Argo app were merged, I could
summon an AV. The AV would go somewhere, get the assistant, come to my mother, help her for the day,
and then the AV would take the assistant back to wherever the assistant came from.
AV goes back to Depot. And so that unlocked incredible opportunities to help people
that have no option today. It would be wildly expensive to hire an assistant by the hour
and then coordinate an AV and all the pickup and drop stuff. So there is absolutely merit
to the notion that an AV doesn't necessarily have to do away with a human or a humanoid to
accomplish tasks to create products that don't exist right now. So I first saw the humanoid,
potential for humanoid application or ribotaxi. I think it center eight years ago I was in a
conference. I forget what city it was. It was in Tel Aviv where a startup had just built a humanoid
that only could drive. It was just sat in the seat. It could stand up and kind of do some
basic stuff, but it sat in the seat and just did driving. And I asked the guy, what else could
he do? He's like, well, it could get out and help open a door. I don't know what happened to that
company. So yes, there's absolutely merit to a humanoid maybe just being in the seat
and getting out and helping. Maybe the humanoid is getting into a current human centered vehicle
and driving using only sensors on the humanoid. It's certainly possible. It's not optimal,
but it's possible. And then it could perform a task of helping with doors and packages.
Now, whether there is a business case like a financial model where that makes sense,
I don't know. Has humanoid become cheaper? Yeah, sure. But the question is going to become
who resolved the driving task? The humanoid in any vehicle? The humanoid deferring to a vehicle
just at the driving? Because the humanoid, it's not free. So anyway, that's for another show.
Yeah, but so, you know, why I come to this where what struck me when you presented this and of
course, what brought it to mind for me is I just came back from the ITN America annual conference
and Catherine Freud have been running I think for 30 years and basically they provide high quality
affordable mobility to seniors who don't drive. And how they do that is through volunteers.
And the volunteers really end up doing two things, one driving and the other one is helping
these seniors. I mean, at her dinner last night, she had one of her riders is 102 years old.
And actually Catherine is going to give her a ride to church on Sunday. And so there's this high
quality mobility requirement for those. And in fact, if we have a Waymo or a Tesla or a
Zoos that can drive, the problem is there isn't the person to help these seniors who really then
need just an arm to help them. And in fact, you know, what I was telling Chen Yi yesterday,
look, you guys instead of, you know, having to do fist bumps and so on,
why don't you just have this darn thing give me an arm? You know, who cares whether or not
you can play soccer or baseball or something. Boy, if you can have one of these things help
a senior be able to be mobile in a vehicle that can drive itself, it doesn't have to drive it.
We got stuff that can drive these things. But what we don't have is the human capability to
help the person so that they can then, you know, go out and actually acquire and do the ride.
And so I'm like going nuts in terms of these. I told Chen Yi, you do that. Plus,
I have a sister who's 88, who's exactly as you describe your mother. She needs somebody there
during the day. Okay. And this thing could also do that during the day and help her into the car
and just sit there in the front seat so that, you know, when the policeman pulls a damn thing
over at least with the large language models, can they sit there and discuss with the policemen,
get themselves, you know, let me go. I'm just fine. I don't know. It's just, you brought this
up and I just loved it because I think there is for a certain segment of the population,
certain needs, not me going to the airport. Okay. I can throw my stuff into the back seat and
whatever and so on. But the first segment of the population, I think can make a real difference.
And that segment is not necessarily small. And there, I think there is a business case.
I totally agree with you. If I'll throw one more in there and then let's keep going because we're
on the clock. Yeah. Yeah. No, I know. I know. Otherwise, there'll be days on this. I mean,
there are so many things you have in this presentation that just, whatever, go ahead.
I want you to get going too, Alex. You're taking my seven-year-old to school?
Or, you know, I'd like, I'd like, I don't want to send her in a Waymo alone, but let's keep going.
All right. So now I was going to say, Joe, though, as important, and this is another discussion,
is the social aspect beyond the physical help for older adults, the social aspect,
the interaction is as important, maybe more. But go ahead, Alex. I think actually there'll
be a lot of jobs created for people just who are in an AV or like, you know, no driver in a bus.
You're still a human in the bus who's like your community, like a connector. But let's keep going.
Let's keep going. Yeah. So now the top one here is really important of this.
All right. So real, real history. 1995, Navlab out of Carnegie Mellon sends a vehicle from Pittsburgh
to San Diego and something called No Hands Across America. It is not a fully autonomous. It's still
got a guidance seat. So it's semi-autonomous. But this is a real thing. I mean, this is the,
this is a huge milestone. Nobody talks about it because none of the current AV companies
can take credit for it. They did it. They did it too. I mean, it actually, you know, amazing. 1995.
Yes, go. So in 2004, you had DARPA began the Grand Challenges. So there was an off-road one in 04.
There were several more in 07, 08, where they went across the desert and then through the DARPA
Urban Challenge. And teams from all over the country representing all the big universities
went and competed and was from that group of teams. Including us. And I like the president team.
Right. Go ahead. I was about to say that legendary. So 2009, Google launched the self-driving car
project having fired many, a majority, many, if not most of the people from a lot of the top teams.
And that was the birth of really, that was the birth of Waymo. Now 2012, something no one talks
about. HB7027, this is the first state-level preemption law. We're passed by our friend,
Jeff Brand, this is how I met him. And this was a law which basically enabled, made it
preempted local municipalities from regulating testing and deployment of self-driving vehicles
in the state of Florida. And this unlocked enormous value for the state because many companies
came down there to test trucks and then LiDAR. And, you know, this is a really breakthrough piece
legislation. And we need things like this. We really need this harmonization of AV
legislation on a national level. So what Jeff did for the state, the federal government could do
for the country. We need common sense regulation that flattens the landscape so everyone can
compete and, and deploy. In 2013, this is a big thing, Elon Musk and Sergey Brin and Larry Page
got together and had a meeting. It's unclear exactly what was said there. But allegedly,
allegedly, Elon Musk was like, yeah, you know, this self-driving thing you're working on at
Google is pretty interesting. I'm thinking about something like that. And maybe there's
something we can do together. And anyway, it didn't happen. And like a year and a half later,
Elon Musk announced his autopilot driver assistance, which was completely different
choice from what Google decided years earlier, which was we will never do driver assistance.
We are getting our goal is driverless out of the gate. We don't believe in semi-automated
driving features because of, you know, attention and cognition decline as you're
managing automation. That's another show. So two years later, Google does his first driverless
ride at public road, which was, you know, I don't even, we could spend two hours talking about that.
But another thing happened, which is rarely discussed today, Chris Ermson, one of the
OG people at Google Self-Driving Car Project in there with, you know, Brian Sileski and Thrun
and Lenodowski. Ermson gives an interview where he says, oh, my son's 11. And my goal is that he
never needs a driver's license since 2015, which means that six years later, autonomous vehicles
would be sufficiently ubiquitous that his son won't need a license. So unfortunately, he was
misquoted and it became my son won't need a driver's license. But whether it's the fault
of his communications people or Ermson misspoke, this was a catastrophically bad thing to say.
And it's set for literally the world, expectations on AVs, which even today,
10 years later have not been manifested, except in certain cities and successfully in certain
cities, but not at the scale sufficient for any parent to say, my child won't need a driver's
license. And this mismatch of the expectations set by Ermson's quote, and hundreds of billions
invested in the sector in many companies, which would never execute this, is a catastrophe,
a financial catastrophe and a cultural catastrophe, because of all the misalignments of money and time
and safety that have occurred since. Also in 2015, Uber, aware of what Tesla and Google are doing,
they decided to launch their own self-driving project called the Advanced Technologies Group,
and they raided Carnegie Mellon and hired a lot of people at the very top, including a lot of
professors, which basically gutted not just engineering talent, but I believe some moral
talent. Alan, you've often been a moral, a moral light in discussions of these things,
and there's not a lot of moral talent in the senior leadership of the companies of the late,
you know, late 20 teens. So, and then of course, a bald guy named Alex Roy went across country
2015 in a Tesla. This is the first Tesla autopilot cannonball, which is not, well, it's significant
in my mind. But what's really interesting is that even though it's just some automated drive,
in the spirit of a nav lab drive from 20 years earlier, it was taken to be like a milestone
by some people in self-driving, which it wasn't. Again, a mismatch of reality and actual technology.
But it was a little bit of a milestone because you were basically using off-the-shelf technology
to do what you did, whereas, hey, not to take anything away from what was done earlier,
that was a one-off for which, you know, people were soldering and whatever and, you know,
whatever, and that's a whole, that's bringing it to a different light, isn't it? I mean,
that was useful. I mean, yes, I mean, that was a certainly a lesson. I mean, I've done,
actually, we shouldn't jump the gun here. It was very educational for me because it actually
may be optimistic about it. Prior to that, I'm like, this can never work. I'm like,
this will work someday. So, 2016, Joshua Brown, a veteran in Florida, had a Tesla and it, he was
on autopilot, allegedly watching a movie and wasn't paying attention. And the vehicle went
underneath a white box truck and he was decapitated. This is a big thing and even largely forgotten
today. But this... It should not be forgotten because it was a monumental one. Yes. And many
people thought it was self-driving. This proved that it wasn't. Many people thought that Tesla's
approach self-driving was wrong, which is certainly not optimal then. It's gotten better today, but,
you know, they've taken some risks and it's in the design. And so, this is something that's
long forgotten and buried. But it's not the only disaster that occurred because GM invested in
cruise. GM wanted to get in on self-driving and for a billion dollars bought cruise automation,
which was a company in San Francisco that was attaching sensor hardware to Audi A4s. I was
on the waiting list for this product and GM bought them and said, well, let's just make
that a self-driving car division. And then, soon after that, Google rebranded the self-driving car
project and called it Waymo. In 2017, I think it's a fun milestone, the first transportation
focused technology that your capital firm was founded. And it's founded by Riley Brennan,
who's now become a friend to all of us. And it's really important, not just because of the fund,
but because his newsletter to this day is one of the most important on earth. If you're in
I read two. In 2018, second really big public disaster, Elaine Hertzberg was a homeless woman
in Tempe, Arizona. And she was hit and killed by an Uber ATG test vehicle. And we could spend
a whole hour talking about it. This was the first time, well, not the first time, this was a major
milestone because in this scenario, everything that could have gone wrong did, but it wasn't a
technical problem. This was an organizational cultural problem at Uber ATG. It was the result
of a cascading series of potentially okay, but suboptimal decisions, which compounded over time
to lead to this tragedy. And I don't really think the full book has been written about this yet.
And that led to the shutdown of the ATG unit and its acquisition by Aurora, which Chris
was running for time and is today an AV trucking company. In 2020, big thing, Waymo launches
its RoboTaxi service, which is today the only driverless RoboTaxi outside of China that you can
get in and ride around. And it's a stellar product. I ride almost every weekend in Scottsdale where I
live. Now in 2023, an unnamed person in San Francisco was struck by a human-driven vehicle,
which fled the scene, and then struck a second time by a cruise driverless vehicle,
and then which dragged them. And then cruise was not fully transparent with the city and
government officials looking into the matter. This led to the shutdown of crews and the loss of
tens of billions of dollars. Again, another example of a cascading series of internal decisions,
combined with potentially some technology issues. It's unclear if there were really
the details here, but that company's evaporated. And then we have today.
The technology issue that I put on that is that they didn't have a camera in the front bumper
that nobody ever thought that you would start from stop with a body in front of your car. And
none of these sensors really could see what's right in front of the bumper. I ran over my dog
once in my driveway, and I heard a noise under my car, and I thought I had dropped my transmission.
I opened up door and looked underneath, and it was my dog. I never looked in the front to see
if my dog was lying in there. A woman two miles from here killed a workman in her driveway. Why?
Because she put the car in reverse, thinking it was going to be going forward. He was back there.
She crushed him against the garage door. And darn it, if the system isn't there,
sensing the object for which whatever camera she had in her car didn't have the automatic,
hey, there's somebody there. Don't go into reverse or don't do that. Or in a cruise case,
I don't think the system knew that she was there. But again, that's not a technology problem,
because a person at cruise said, we don't need a camera there. And recently in Tesla,
they began putting cameras in the front low to prevent that. I certainly hope Waymo has.
I'm afraid to ask them. They must. I hope they did that the next day on October 3, 2023.
I know Waymo has sensors in front. I still don't know which one it is. I'll look tonight
when I get back to the house. I mean, I don't care whether Ryder, Schmider, or Ryder.
Let's keep going because we don't know what we don't know. Yes, go ahead.
Oh, we got to keep going. All right. So 2025, my team performed a Tesla
cannonball run using full self-driving across country safely. We've done it three times now,
but only one fully successfully because of some weather issues. And Waymo began,
this is a big one. People don't grasp how hard this is. You could build an autonomous vehicle,
but to scale operations across many cities globally is not just expensive. It's
operationally incredibly difficult. And Waymo is doing that right now. And then the last line
I've got for today is, Tesla launches the urban taxi service. Now, theirs is not a driverless
service, but it is important because Waymo Tesla are currently the market leaders.
In Tesla's case, it's theoretical, but it's going to happen in some form.
Camera will work in some way. It might, it's going to have performance differences from
a multi-sensor stack, but it's going to happen. And so on some timeline, it's like arguing the
difference between types of jet engines in 1955, like it doesn't matter. It just doesn't matter,
which brings us to my next point, the present. Nobody knows when they get on a plane or when
they buy a plane ticket. No one knows, unless they're a pilot themselves, whether or not the plane
has a flight envelope protection system designed by Airbus or some variant of it, like, you know,
you're in a Boeing. And most people don't even know if they're getting on a Boeing or an Airbus.
They don't care. They buy a plane ticket. They get to their destination. But this is really
important because the two systems, the Boeing and Airbus forms of flying automation and safety
are very different. And I think there are people, certainly after the recent Boeing crashes, who
only want to fly on an Airbus, but don't know enough. They don't even know what to look for,
what to ask when they're buying their plane tickets. And at the end of the day, aviation safety is
certainly in the United States is so good. It's the safest form of travel that nobody really
cares until there's a crash. And they start asking questions. This is really relevant in other
countries where crash rates are much higher than the United States. But fundamentally,
no one cares in the mature, the mature technologies, what's behind the curtain, which is why debates
about end to end neural nets versus rule based systems or anything in between also don't matter.
The neural net argument is made by Tesla fans or neural nets for driving AV. Whereas rules based,
which at one time was the only way to design an AV is no longer binary. There's a thing
of design choices in AI, which leads you to an AV. Nobody cares. But investors and analysts want
to fight over the different approaches. Nobody cares because eventually these things will just
work. And the only question will be the people running the company know how to build a financial
model and raise money to build vehicles and scale them and offer them at affordable prices
for the majority of people. That's the only thing they would care about. But no one really talks,
you talk about it, but most people don't talk about it. So in my opinion, there's really only been
three autonomous AVs or equivalents in history. And I have this thing called Roy's razor, which is
yourself. It's like Occam's razor for AVs. And it basically is this. Can the user fall asleep
in the backseat of the vehicle while it's functioning? Yes or no? Yes or no. And at Tesla
today? No, you can't. A Waymo today? Yes. That's Roy's razor. So history has examples of this,
a horse and cart. If you had someone who could deal with a horse and you were sitting in the car,
you could sleep. That's an autonomous vehicle, self-driving. If you get a human-driven taxi
ever anywhere in the world, it's self-driving. In fact, in India, the signs at taxi stands say
self-driving cars available. This is a word game that people play. And then finally, of course,
Waymo RoboTax, you can fall asleep in the back. So operations are everything. And in the future,
AVs will be ubiquitous. Doesn't mean 100%, but they will be ubiquitous. It's inevitable.
The only question is whether leaders of these companies will make moral choices and build
really good ones on a pace to deployment, which does the most good for the most people,
or if they're going to rush out suboptimal solutions and cut corners. And we've seen some
companies shut down because they have done that. I think safety is a moving target and
different things to different people. So I think that right now, we're moving from a culturally
accepted level of safety and just driving daily, which is not very good in this country. It's not
that safe, but people kind of live with it. We're moving into a quote unquote safety,
like people try to figure out what is acceptably safe for an AV. I think they're being held to
unrealistically high standards, but that's a different show. And eventually, different safety
narratives will be fully cooked around AVs, and they'll be culturally accepted at some perceived
level of safety, and then we'll go back to just safety. And so right now, Waymo is putting out
tons of data about their safety, which is very good, and also probably indecipherable to the
average person. They're the only ones putting out data at that level from, you know, robotaxies.
And Tesla's not being transparent. And yet there are people who think Tesla owns this and they're
safe now. So you see, there's two different lenses upon which people are looking at these
companies. Eventually, they're just going to blur. My seven-year-old daughter gets to Waymo.
She's like, oh, Mr. Robotaxies, so good. We get in my Tesla and I activate FSD, take my hands off
the wheel. And she says, why are you behind the seat, daddy? The car's a better driver than you.
She thinks they're both safe. And she thinks that I'm an idiot for being in the driver's seat of
my Tesla. That's the future. She's seven years away, eight years away from driving a car. And so
at that point, I don't know what's going to happen. So on the safety end, Alex, if we could spend just
a minute on the safety end, because the safety end is, to me, I consider safety to be a constraint,
not an objective. Because if we are really trying to solve the safety problem, then there are ways
to solve the safety problem. You just put speed limiters. You put whatever, you know, breathalyzers
and all that stuff in cars. But in terms of the perception, it has to be safe. And I must admit
that at the New Jersey Senate hearings, state Senate hearings on Monday, Waymo did make a
presentation. Matt Walsh made it. He made a beautiful presentation, basically just, I mean,
nailing. We're safe, okay? He sort of suggested that their objective is to, you know, reduce
lives. And that'd be the objective of the technology. But it's really the first time
that I think I've heard Waymo make a real public statement saying, we're safe. But we're doing
this maybe for a lot of other reasons, because people need mobility. I mean, your daughter is not
taking this because it's safe. She wants to get something. She'd like to get there. Maybe she
loves getting there with you, but she really wants to get someplace and sees that the hurdle for it to
be safe enough in her mind is, hey, this is fine. I just want to get to where I'm going and I'm
fine about it. I mean, how do you handle this subtle difference between having safety be the
objective and safety being a constraint? People don't take airplanes because they're safe.
People take airplanes because they want to get someplace fast and whatever.
And honestly, I think this solves itself because people flew planes at scale through into the
70s when truck, you know, fatalities and plane crashes were horrific. Nobody cared. Nobody cared.
Yeah, we flew it in Lockheed Electric. Lockheed Electric crashed more than they
knew. Anyway, yeah, it's ridiculous. We're going to race through these because I'm getting
short on time. Yeah, I know. I know. But this is yes, we can do a whole episode about every slide.
I know. I mean, this is why this is so darn good. I mean, it really is go. Sorry. The end of case,
you know, these, these lot of folks consults like, Oh, the future mobility is connected a ton. I'm
a shared electric. And the truth is it, it will be all those things. It doesn't need to be all those
things in one product. It can be one, two or three of those things. It can also be none of those
things. And so this has become like a narrative funnel that like restricting our ability to think
about what these products should be. And honestly, I wish this would just go away. I wish it would
just be used by like a company building a case type vehicle. And so everyone else can just move on
and build, you know, pedal assistance unicycles, which is not it, which isn't any of these things.
And yet is it really cheap and great products? I mean, the thought that Waymo or Tesla or
anybody would put a vehicle out there without it being connected and without them having the
opportunity to see exactly what the hell is going on with that damn thing all the time is like
ludicrous because that's that's trivial. It's trivial on the end. It's table stakes.
You know, let's keep going. Let's keep going. Anyway, the SE level, the SE levels are been a
disaster for all this discourse sector disaster. And everyone is I always like I know some people
involved and they were well intentioned nice people. But they did no good last time and they
probably should just stop. And I'll say like, I'm not saying I have a solution. But in my opinion,
all vehicles should be called one of two things. They should be called human driven,
or let's call assisted by a human with some aid aspect assisted or geotonomous. It's it's
driverless in some places. And then the geotonomous vehicles have a map on the door, like that looks
it's a map of where you are. And it just like it just has a little color thing for where it's
driverless. Yeah, let me keep going out. The thought that it could be everywhere to me drives me
nuts. Or you really drive down a riverbed the way they show I can take a Jeep in my Jeep.
Really? I mean, cut it out. There are so many places that could do so damn much good. Oh my
goodness, without being everywhere. What's so great about everywhere? Sorry, but it's toilet.
The fun thing about the word autonomous is that actually, an autonomous vehicle would be one
of things for itself. And that would be a way more robot taxi. Today we have an autonomous vehicle.
But they're not they're not they're told they're told when to go when we're to go.
But hear me out. You decide that themselves. I'm getting to that. So if you want to buy
an autonomy vehicle, an autonomy vehicle, an autonomy vehicle is one that can go anywhere
that's human driven. And the perfect example is a Jeep Wrangler is an autonomy vehicle.
But a ton of us vehicle is always a rubber taxi, which actually can go fewer places. So
pick your poison. Yeah, absolutely. Yeah, my, my, you know, almost done here. I think all vehicle
of the future will be dual mode. All vehicles of the future will have driverless capability
inside some domain and have some form of driver assistance. But in the future, it'll be it'll
I think driver augmentation. So because the vehicle has will have amazing sensors and AI,
and it'll still have you can still have a steering wheel. And so you drive it in all
kinds of places, it just won't let you crash at a level of safety far better than current
driver assistance. Or it might have a joystick instead of a steering wheel. So I mean, damn
it, if you have a joystick and whatever, I mean, it's kind of that guys, they'll all have
elevators still have ways that you can control it in case the damn thing go ahead and never.
Frank, I wish that elevators still had like the old handle with like the wheel.
Just for fun, but it would still stop at the floor correctly, which is the old wheels and
handles didn't do well. Yeah, it could be a little assistance. Yeah. Yes. And then my final
slide is just look, a lot of people forget this. A lot of I mean, I'm sorry to say this,
I want to say the very selfish and cynical people who are invested in many of the companies in
autonomy would have you believe that Tesla is going to win everything. Everyone else is garbage.
And that's it. But the reality is it doesn't matter who or how many American companies are
successful. What matters is that there are only two countries on earth that two countries
that dominate this development. And this is China, the United States, there are some European
companies, but wave is an example on this mobile. But fundamentally, we need to be exporting the
best, safest autonomy in the world to the world. This was, we've done this before with cars and
aircraft. We need to do this again now. And we have to stop having debates about worth of stupid
debates about neural nets versus rules versus this. None of that matters. We need to be building
great affordable products, vehicles enabled with AI that are really great, that are affordable,
that can be deployed at scale and licensed and sold at scale. Because without that, the United
States is just not going to, it's not going to be good for us. In the end, they're mobility vehicles
and we need mobility to basically improve our quality of life. And the way that many people
are going to get that is through the affordability aspect, because the reason their lives are not
as good as they are is because they can't get to places affordably. And at least that's my
current pitch. Those that have plenty of money and don't care about the affordability have a ball
too. But it looks so insane. I grew up in New York. I never considered myself. I was never
like a political person. I never get engaged in political debate. But a fundamental aspect
for capitalism is that it's a really expensive to invent something the first time and put the more
of them you make, the cheaper they get and then everybody has one. And so TVs, cars, that scale,
that scale, that scale. Because it became cheap enough. And we also add that when Uber ATG shuts
down and cruise automation shuts down, these are disappointments. However, it doesn't mean the money
was wasted because everyone who's there, their lives will be defined by what they learned there,
good or bad. And it will take those lessons to the next place they go. So capital and I look
through the lens as an adventure capitalist today, capital allocation, it matters. You want to bet
on the winners. But in the long term, everyone wins because money was invested in a noble idea
that can be manifested in the physical world. Look, I also say that you might learn more when
you fail than when you succeed. Oh, always. You know, and so, you know,
Princeton rejected me. Thanks, Alan. Princeton rejected me. And now here I am.
And no, no, no, no. Well, no, I mean, look, hey, I failed. You know, hey, I wrote my first
Trenton Transit study to bring automation to improve mobility in Trenton. It's dated March 1975.
So I have had more than 50 years of failure in bringing any automation to improve mobility
in Mercer County, Trenton, New Jersey. So I mean, you know, and I'm finally decided, damn it,
you know, I want to fail at least one more time. Damn it. Well, Alex, I know you've got to run soon.
Yeah. I do have to write one more thing. You can't on my American autonomy slide on the bottom right
corner was a tiny little white box. Let me bring it back again so you can see it. And here it is.
You see in the bottom right. Yeah. Yeah. Yeah. Yeah. Yeah. All right. Do you know what that is?
What? That's the symbol of Colossus, the AI for the movie Colossus to form in projects.
And I left it there with a little nugget for anyone to recognize it. And we should do a whole
show about that another time. All right, friends. Alex, thank you so much. Sorry to keep you and
whatever. It's always fun. Enjoy. And let's keep it all going. Hey, we are done with this stuff.
We'll have we'll have more wonderful times. Anyway, thank you. Great. Thank you. Love you. Bye.
Thank you. Well, a couple of other headlines and from the smart driving cars newsletter, Alan.
Elon Musk is saying Tesla owners could be paid to make their vehicles available to create an
enormous shared AI inference capacity. Adam Jonas was quoted in seeking alpha on this.
I mean, that's it's I think that's just very interesting, especially, you know,
the concept that he might pay $100 to $200 to you to to basically use your your
Nvidia chipset that happens to be in in your car as a compute engine for the time you're not using
it. And oh my goodness. Now, of course, who pays for the electricity that gets to be needs to be
used while he's sitting there doing computations on it. So there's an expense side and maybe there's
a, you know, somebody has to cough up the electricity. But the prices and Jesus, as I've
suggested, hey, if he he lets me use FSD, and I let him use my my compute when I'm not using it,
it might be a pretty fair trade. We have to discuss, you know, the electricity that is consumed
while it's my car is just sitting there running his compute cycles. But my goodness,
might have a deal and so on. So very, I mean, that's just a whole new twist on this thing.
Now, you know, I happen to think that the price of FSD is is a pretty good price. If in fact,
you're using it to give rides. But this makes it even more interesting. Whatever. Another twist,
I didn't anticipate that one. Musk has also made some comments that came from the shareholders
meeting, in which he said that Tesla drivers, within the next month or two, perhaps will be
able to text and drive while using FSD. And I'm not sure how we're going to feel about that.
I don't know. I don't think I don't think it meets Alex's test of being able to get in the back
seat. No, it doesn't meet Alex's test to be able to get in the back seat. And what I would like it
to do is for Tesla and NHTSA to have an agreement that one, me sitting there looking out, having the
eye tracker with my eyes looking out on the road ahead is the safety condition that's important.
I mean, I actually, I actually think it's not because the, to me, in evaluating FSD,
the user interface that he has on the screen that shows me what the car's intentions are over
the next three seconds, four seconds, and what the car thinks it sees are way, or as important as
me looking out the front end at the road ahead. I think I should be looking at the road ahead,
checking that out. Do the two of them make sense? And where it's intending to go in the next three
to four seconds, is that pretty much given the, my gaze out the front, where I would want the car
to go. And as long as those things are, are copacetic together, then I'm good. So therefore,
my gaze should be going in front. Now, somebody might say he should do projection on the screen
to the D. I mean, I just hate that stuff, but whatever, because I don't, me focusing on what's
on the windshield is not really seeing what's on the road ahead, or at least maybe that's just my
visual perception. But the looking down to see what it is that the vehicle is seeing,
and what its intentions are over the next X seconds are really important. And I only get that by
looking at the screen and deciding whether it's not. So if all of a sudden the thing goes nuts,
which is the real thing we're afraid of, we don't know when this thing's going to go nuts.
Okay. I mean, that's the hallucination part of the whole AI black box structure is the real
unknown and using that kind of structure to generate the controls here. And so,
but it gives you a hint, a very strong hit is what it intends to do. So all of a sudden,
if the road is curved and left ahead and you want to be going left, you look at this thing and it
intends to go turn right, then grab the wheel, then hit the brakes. Okay. But you have to
realize you have to somehow know it's trying to tell you this and just to leave that information
there and not have it in your mental state, I think is inappropriate. I'm surprised that NITSA
allows, you know, thinks that this is great. Hey, have an eye tracker to make sure that
my eyes aren't a road ahead. No, road ahead, what it intends to do, what it sees, does it match
and have this go on as the interface between these two things.
Interesting. Alan, before we wrap, any other headlines or updates to share?
Well, no, except to repeat, we're going to do another one of these podcasts with
Catherine Freud of ITN America. I mean, what they do in terms of providing high quality,
affordable mobility to seniors is just marvelous. And they've been doing it for 30 years and the
information they have as to, you know, why people do the, get the rides, what benefits they get from
it and so on. It's just fundamentally valuable and I think really important and I think is
important to the whole so-called robo taxi. If we're out there giving rides to me, I think the
value of those rides are because they're going to be high quality and affordable. And if they're
high quality and affordable, then of course you get more on the demand curve such that your demand
goes like this. And of course then through scale, the costs should even go further down.
And of course then this is, this is a real, really good dynamic in terms of
what all this is about is to provide mobility. It's not about the technology. Nobody cares about
that. Really, they don't. It's just, you know, for the one selfie, look mine, whatever. But
I don't think anybody cares. I agree with Alex on that.
Right. And you can find more on the great work Catherine is doing at itnamerica.org.
Hey, we want to thank Alex Roy for spending time with us and sharing that terrific presentation.
You can find us at smartdrivingcar.com and my tech reports are at textination.com.
Thank you for watching or listening. Please stay safe.
I'd like to also piggyback on Alex. Alex has been around for a while on all this and has
spent the significant efforts and welcome to him. Thank you.
About this episode
Fred Fishkin and Alan Cornhouser welcome Alex Roy to discuss the evolution of autonomous vehicles (AVs) and the narratives shaping the industry. Roy shares insights from his recent presentation at the Florida A.V. Conference, emphasizing the importance of understanding the historical context of AV development. The conversation dives into the influence of pop culture on public perception, the challenges of legislation, and the future of mobility, especially for seniors. Roy argues for a focus on creating accessible and affordable AV solutions while addressing safety and operational challenges.
Original notes
0:00 Open and welcome to Alex Roy
2:40 Alex Roy presentation on the Past, Present & Futureof Self Driving
4:54 Alex presents The True History of Self Driving from DaVinci to today
10:47 Could in vehicle humanoid robots be helpful
19:50 from DARPA on
31:30 In 2025 Waymo scales operations and Tesla robotaxislaunch
34:00 Roy’s Razor -the test of what really is self-driving
40:30 SAE Levels should be done away with
44:55 A focus onaffordability?
48:20 Tesla owners could be paid for use of vehicle computepower for AI
50:25 Elon Musk issaying drivers using FSD may soon be able to text and drive
54:55 Congratulations on ITN America’s 30 years of providingrides
Smart Driving Cars Podcast
Fred Fishkin/Alain Kornhauser
0:00
56:59