March 13, 2025 • 53 mins
Alex Roy and his co-pilots recently made two historic coast to coast trips - California to New York - in his brand new Model S. Their mission: Measure the efficiency and safety of the Tesla FSD system. How close was Tesla to mastering full autonomy? Roy and Team use two key metrics. The first is the total number of disengagements – whether voluntary or involuntary. The second: How quickly could the system recover after a disengagement? Roy, who previously set the Electric Car Cannonball record in his Tesla, has now turned his focus to autonomous driving. His central question: How soon will cars be able to drive themselves coast to coast allowing us to take a nap in the back seat? I first met Alex Roy in 2018 at a private industry event south of Munich. He's one of a kind - colorful, smart and professional. You will enjoy his vivid first-hand account of the cannonball runs, both the triumphs and the reversals - like the time the team encountered a good old-fashioned midwestern thunderstorm in Indiana. How exactly did The Model S and the FSD system respond to that crush of water? Alex Roy, partner, New Industry Venture Capital on the Driving With Dunne podcast.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
We're top executives and crazy entrepreneurs gathered to talk about
the future of electric vehicles. This is the Driving with
Done podcast.
Speaker 2 (00:12):
Hello and welcome to the Driving with Don podcasts. I'm
your host, Michael Dunn. Now you ever wonder what it
would be like to just jump in your Tesla and
take off across the country in full self driving mode.
That is, no hands, no feet, the car drives the
car all by itself. Hmm wouldn't that be exciting, scary, thrilling, terrifying. Well,
(00:38):
guess what someone else has already been there and done
that twice. His name is Alex Roy. He's the co
founder of New Industry Venture Capital, and he and his
co pilots recently made two trips coast to coast California,
New York in his brand new Model Less. Why what
were they all about? Their mission measure just how good
(01:00):
Tesla's full self driving capabilities are today in twenty twenty five. Now,
Roy is no stranger to testing the limits of vehicles.
He is the record holder for the electric car cannonball
run that he achieved in his own Tesla several years ago.
Now he's turned his attention and his expertise to autonomous driving.
(01:24):
His question, how soon will we be able to ride
in the backseat of a car, sleeping, take a nap,
and have the car deliver us coast to coast. I
first met Alex Roy in twenty eighteen at a private
industry event south of Munich. He's one of a kind, colorful,
smart and professional. I just know you'll enjoy his vivid
(01:44):
first hand account of these cannonball runs, both the triumphs
and the reversals. Reversals. Yeah, like the time the team
encountered a good old fashioned Midwestern downpour thunderstorm. It was
in Indiana. I wasn't there, but I picture rains coming
down like a cow pissing on a flat rock. How
(02:04):
exactly did the model esque respond to that water deluge?
Let's find out with Alex Roy, partner New Industry Venture
Capital on the Driving with dun podcast. Alex Roy, the
(02:32):
one and only. Thank you so much for joining me
today on the Driving with Done podcast.
Speaker 3 (02:36):
Great to see you, Michael, Love Coming and Shell love
your show. Love you Love Coming.
Speaker 2 (02:40):
On the first day back from way across the Pacific
in a country called New zeal Zealand, Yeah, what was
going on over there?
Speaker 3 (02:49):
Well, my my friend, and well I think he's your
friend too. Rob Conybeer of Shastadventures loves going out in
New Zealand. He organized a group of American investors to
go down and meet with local investors and local startups
UH to familiarize you know, Americans with the tech ecosystem
and meet some members of the government. We met the
(03:11):
Prime Minister and the head of Immigration and Education and
a lot of interesting folks UH to encourage investment and
and to witness an announcement of a new UH, new
policy which makes it easier for foreign investors to come
invest and potentially by real estate and get you know, visus.
(03:32):
So it was my first time down there.
Speaker 2 (03:35):
Any sightings of evs or a v's in New Zealand.
Speaker 3 (03:38):
Not a single AV in the country I'm aware of,
but a lot of a lot of hybrids every cab
as a prius, I wrote it a B y D
which is cool and UH was just blown away by
the culture, the people, the governance UH. And I didn't
know there was a startup ecosystem down there as healthy as.
Speaker 2 (03:56):
We knew, not at all.
Speaker 3 (04:00):
What a fantastic place. And Rob put a great, a
great event together fantastic. I can't wait to go back.
Speaker 2 (04:06):
We're here to talk about autonomous vehicles and in particular
you're cross country run. But one more point on New
Zealand before we leave. You mentioned riding in a BYD.
Impressions highs and lows.
Speaker 3 (04:20):
Very nice, really well made. The interior was groovy in
a way that only kind of renault like modern renaults
are groovy. Maybe it had a little flavor of like
cicro en, like the what is it the C four
which has the goofy the goofy exterior rubber panels. The
(04:40):
interior of the BYD. The rear passenger door release was
is a curved. It's like a curve. It's a crescent
which rotates around the mid range and tweeter. Assembly for
the rear passenger.
Speaker 2 (04:58):
Is good or not good, I don't know.
Speaker 3 (05:01):
Huh uh. It's probably easily ripped off the door if
one it gets feisty. But I thought it was clever.
I thought it was clever, It was cute, and uh
it was. It was a good car. The driver loved it.
He absolutely loved that car. So my first time into
b Idea, I was impressed.
Speaker 2 (05:18):
Now bigger than Ford, bigger than Nissan bigger than Honda
chasing GM. Wow.
Speaker 3 (05:25):
Uh you know, I think the reputation is probably well
earned from everything I've seen and heard, And of course
I'm going to have some interesting announcements I think later
this year around a Chinese vehicle. But that's for another show,
not this episode.
Speaker 2 (05:41):
Okay, you're right. This episode take us back to the beginning.
Coast to coast time and place. You're it's it's the
Tesla FSD cannonball run across the United States. Where did
you start?
Speaker 3 (05:57):
We start? We started, well, we've done a couple of time.
So the first time we started at the Red Ball
Garage in midtown Manhattan and ended at the Portofino Inn
in Redondo Beach, California. On the second drive, we did
it in reverse order, same locations. The point of it
is that there are very well, first of all, there
are very few people who are doing really authoritative reviews
(06:23):
of driver assistance, and then there are basically no reviews
of driverless vehicles like robotaxis. And so since Tesla full
self driving software is the world's most anticipated and popular
such package, we thought that someone should take Tesla and
(06:46):
their claims of imminent readiness to go driverless. Let's take them,
and let's take those claims, and let's actually put some
metrics behind that. And I had not seen anyone do
that in a meaningful way. So since the nurber ring
is has been in the metric your Nerve Burgering time
is the metric for every human driven car that is
(07:06):
remotely fun to drive. Everyone takes their cars that come
up with a metric. I thought, what is the what
is analogous to that for autonomous vehicles? And so there
we be two. If it's a personally owned vehicle capable
of driverless operation, then it has to be the cannonball
run route New York to LA Or and back. And
(07:28):
then you would take that route because it's it's something
anyone can do, and you would drive it over and
over with every software release or hardware upgrade and watch
the metrics evolve over time, and then you could begin
to learn the rate of improvement of the system performance,
whether it's software hardware changes in a domain which is
clearly recognizable and also very famous. And so a few
(07:52):
months ago Joel Johnson and C Fox and I friends
from Porsche and my consulting life, decided we were going
to go do this.
Speaker 2 (08:01):
Three of you guys, one vehicle.
Speaker 3 (08:04):
Two of Only two of us have gone at a time,
although we'll probably go three of us on the next drive.
And we would take my brand new Tessa Model S
and we would fill it with every possible piece of
hardware after market we could to measure everything. Everything, We
everything everything, and then on every subsequent drive we would
(08:24):
add or subtract something to improve our data gathering.
Speaker 2 (08:27):
Everything related to autonomous driving or everything related to everything
like EV range times what.
Speaker 3 (08:35):
Well, not really interested in the EV range because there
are people who only do that. They and if you
just do that, you will get more, you'll get better
information than a catch all drive. We are interested in
the automation, autonomy and rate of improvement towards driverless operation.
So the first drive, which was in mid December of
(08:55):
twenty twenty four, we got Fleer Teledine Flair, which is
you know, big supply of thermal cameras and imaging equipment
to sponsor us in the form of a pre production
prototype Flear camera, which will be I think it's going
on an OEM vehicle this year or next year.
Speaker 2 (09:12):
How do you spell that?
Speaker 3 (09:13):
Fl I R and I have a lot of experience
with FLEAR because I used a Flear in two thousand
and five and six on my BMW to set the
original Cannonball run record, and that made it a lot
safer to drive at one hundred and fifty miles an
hour in the dark and spot animals or police. But
(09:33):
in this case, since Tesla is a camera only system,
we thought it'd be it would improve our safety to
have a Fleer camera mounted on the front of the
car with the display mounted on top of the dash.
That would show us thermal imagery, which is something that
Tesla cannot do, and that would be a backstop. Will
(09:54):
the Flear potentially see something that the Tesla's you know,
passive traditional camera do not see, and by recording using
a GoPro over our shoulders facing forward, in the event
of an incident of any kind it could be a
fantom breaking incident or heaven forbid, a near miss or
even a collision, then we could correlate what the Fleer
(10:16):
sees to what the Tesla sees and understand whether the
Tesla might have performed better if it had a let's
just say, a different or multi modal sensor package. And
so so this Flear.
Speaker 2 (10:30):
Does it give you more distance and sees through the
dark or what's its secret?
Speaker 3 (10:35):
It can see temperature. You know, it's a thermal thermal sensor,
so it can see at night. It will see well
beyond the headlights of the car, well beyond in the daytime.
You know, it's it's pretty much the same. But what
it will do, which was fascinating on the second drive,
is it can see through you know, fog rain, you
(10:58):
can see through obscurations and so. And it also has
built in even the pre production unit has bounty boxes
and classification of objects. So when we were looking out
the window and couldn't see very far and even the
Flear display was obscured, it would still show bounding boxes
and classifications even on the basically missed it over just
(11:21):
gray background. So it's a very impressive unit. And I
you know, I'm not paid by Flear, but I wish
to me this is the magic sensor that every car
should have because they're very inexpensive.
Speaker 2 (11:33):
Where do you mount it? And how big is it?
Speaker 3 (11:35):
It's a pencil cam. It's like the size of this
little pencil in my head that I don't know. I
think it's like a couple of inches long. I had
it mounted to my front license plate bracket and it
runs a cable to the pasture compartment and connects to
a screen. I think they're they're looking at like eight
ass applications, but honestly, I think it's a great it's
a great technology. I'd like to see it used. And
(11:58):
so in in December we went on our first drive
and that was kind of just a benchmarking drive, and
on that one we were measure looking for two metrics,
because the metrics are not time for that's for something else,
(12:19):
traditional canniball records. And in the future, when driverless operation
is capable is possible with two or more companies, then
you would put them head to head, like how fast
can they go across? But for our purposes, we want
to measure the rate of improvement towards driverless operation, meaning
unsupervised operation, and to do that you need to gather
(12:40):
data on two metrics. The first one is number of disengagements.
The goal is zero. It has to be otherwise it's
not driverless, and you need in vehicle supervision. Potentially you
could have remote supervision, but at highway speeds not realistic
with the current state of the art, or at least
not in the eight not at eighty miles an hour.
Speaker 2 (13:02):
So to put some color on this unsupervised, I think
it was you to coin the term like if you're
sleeping in the backseat of the car, that qualifies.
Speaker 3 (13:10):
Otherwise you're correct, correct, because I mean it is. Every
autonomous vehicle will have a customer service function and some
remote operations function, and they're the goal of the companies
is to reduce that supervision or operations function, the remote
one to never but periodic and something that can be
done safely, which means that there has been low latency,
(13:34):
low latency and you know, a high ratio of vehicles
managed to per remote operator. So if there's if there's
a if it's one to one and someone's got a
steering wheel in an office, you're going you're going out
of business. That's not good. That doesn't scale. And so
that's the first thing is number of disengagements. And then
the second metric is percent system engagement rate. So over
(13:56):
your entire time driving, what percentage of that time is
the system engaged? Because if you have a dis engagement
and currently you're going to have a bunch, how long
in between until you can re engage the system. And
so this is really important because if you have a disengagement,
you want it to be if possible seconds seconds. And
(14:19):
in the event that these vehicles become unsupervised and but
still have a remote operations function that's necessary, well, they
want to minimize that too, because the shorter the time
the remote operators engage with a vehicle in motion, the
better and safer. So those those are the numbers.
Speaker 2 (14:37):
Can I jump into a couple of questions, Yeah, yeah,
when we talk about remote operator in Tesla's case, that
does that exist?
Speaker 3 (14:46):
So Tesla has long stated that they're just going to
be They're gonna be autonomous, and they have avoided the
topic of any remote operations. Customer service function however, uh,
there is strong indication that they're hiring people for an
operations team, and whether or not they admit it, customer
(15:07):
service is requires a remote operations function because eventually the vehicle,
every autonomous vehicle, like every technology, will have an issue MM.
Speaker 2 (15:18):
So that's a great point. So thing go one without
exception is going to have an issue.
Speaker 3 (15:23):
With exception, and especially if if the cyber taxis don't
have steering wheels and that thing accounts as a scenario
cannot resolve. Someone's got to get that thing moving again,
whether or not that's a passenger inside. So who's doing that?
Because a passenger can't do it if there's no steering
wheel path, and even if there was way, MOO doesn't
want you getting behind the wheel and moving the car.
(15:45):
So remote operations is a customer service function. There's no
product made in the history of modernity that doesn't have
a customer service function of some kind. And so, uh,
we'll see what tesla how tesla resolves at later. Right now,
the owners of the vehicles are performing this function.
Speaker 2 (16:03):
You are You didn't know it, but you are the
remote operator inside the camp.
Speaker 3 (16:07):
Yeah, tester, supervisor, whatever you want to call it.
Speaker 2 (16:10):
Yeah, all right. So you're measuring two things, number of
disengagements and the length of time between getting disengaged and
getting re engaged. Yeah, okay, So what does what is
a disengagement any example? What does one look like for
someone who hasn't been in your seat?
Speaker 3 (16:28):
What does it look at the highest level, there are
two types of disengagements, involuntary and volunteery. Involuntary would be
that the system, for whatever reason can no longer drive
and it then says, you know, take over. In a
perfect world, the system would know that would warn you
(16:49):
of a necessary takeover, you know, minutes ahead of the driver,
having the supervisor happy to do so. When you know,
when you are not engaged in the driving task, you
begin to zone out. It's very hard not to, especially
over many hours. In our case. You know, we're looking
we're talking in access of forty hours behind the wheel,
and so studies have shown that it may take up
(17:11):
to forty seconds. Is how long it may take someone
behind the seat to cognitively re engage safely and take
over if you're sitting right there in your task. Is
just that it could be milliseconds. Okay, So that's an
involuntary disengagement is no bueno. You don't want it, but
they happen. And then the other one. The voluntary disengagement
(17:37):
can happen for a variety of reasons. It could be
that the vehicle is performing a maneuver that's illegal and
we don't want it to proceed because the liability falls
on us if we're pulled over. It could be for
the vehicle approaching an unsafe scenario and it is and
it's unclear that the vehicle is going to resolve it safely.
(17:58):
An example would be that a truck is entering into
your lane, uh, and it's doing so very closely, and
the vehicle, your vehicle, the Tesla, does not appear to
be breaking in time and colision might be imminent. So
everybody has a pucker factor, it's a little different on
that one.
Speaker 2 (18:17):
And then how are you on to scale a one
to ten on that puck that truck.
Speaker 3 (18:23):
I'm pretty good. I'm pretty good because I've used so
many driver resistant systems, have been in Sony astonomouses.
Speaker 2 (18:28):
You come to trust it more than a first time
or would Yeah.
Speaker 3 (18:32):
I mean this is the the devil's The devil's math
of testing or supervising test vehicles is you know, what
is your pucker factor? Because if you if you disengage
every time you think the vehicle might not get it,
you're gonna have hundreds of disengagement and but if you
never do, you will eventually have a crash. So you
(18:53):
I like, I mean, I'm always lean towards how we
let the vehicle go as far as it can until
it's imminent and then take over. Well, it's important also
to note that not everyone is equally capable of taking
over and resolving those scenarios well or safely.
Speaker 2 (19:10):
And your average driver in America probably wouldn't want to
have anything to do with it because it's just it
can be a little bit terrifying that truck boots over
some carbs moving into your lane. How much do I
trust this vehicle.
Speaker 3 (19:22):
That is correct? It can be hair raising. I have
good news on that score when we get to the
actual results. Another type of voluntary disengagement would be navigation
related where the GPS indicates that the vehicle is going
to go here, but the vehicle, whatever reason, doesn't actually
follow what's on screen that the nav system and the
(19:42):
vehicle like motion planner don't decide that they're going to disagree,
and the vehicle makes a left turn instead of a right.
So which point you have to choose. Do you let
it play out and see if it makes the U
turn or do you just let it keep going in
the wrong direction? And I like to you know, if
it's a quick turnaround, well, I think I like to
(20:03):
see if the vehicle can resolve it. And then the
last sip of disengagement would be for charging. So on
the first drive, Tesla FSD twelve five six four was
the software version we used. That that version was not
capable of navigating into and out of Tesla supercharger stalls,
(20:26):
so we had to manually take a voluntary takeover for
each of those those stops.
Speaker 2 (20:31):
Why do you think that is if you were just
pulling into a regular parking lot, would have been able
to do so? Or is there something specific about the
supercharger station.
Speaker 3 (20:39):
I think it's specific to the supercharger stations because Tesla
has had parking functionality for some time, which is with
mixed mixed results, but they've never they did not They
never claimed supercharger capability with FSD version twelve. They did
for version thirteen, which we use in the second drive.
Speaker 2 (21:00):
You saw a difference.
Speaker 3 (21:01):
Oh yes, So on the first drive we had a
total of thirty one disengagements, of which twenty one were
for charging. Is that right? Thirty two disengagements with twenty
one for charging.
Speaker 2 (21:14):
Total you're talking total disengagements.
Speaker 3 (21:16):
Yeah, between two thousand between New York and La.
Speaker 2 (21:20):
Two thousands. Phenomenal.
Speaker 3 (21:21):
Yeah, if you take away the charging, yeah, it's quite good. Yeah,
less than a dozen disengagements of all types and and
on a second drive, the numbers were radically improved. So yeah,
the vehicle was capable of pulling into a charge is
successfully almost every time. It's instead of twenty one disengagements,
(21:43):
we had eight for charging, which is extraordinary. And the
other metrics.
Speaker 2 (21:49):
So we're out. You're rolling down Highway eight or ten
or forty whichever one, going east to west, and you
you're you're how much of this is the caught you saying,
I'm driving along the car chooses a super charge station
or do you guys choose it?
Speaker 3 (22:05):
So when you Tesla has an extraordinarily good GPS and
a charging planner, So when you are in New York
at the start line and you select your finish line,
it will prepopulate all the chargers across the whole way, country,
the whole way with as you know, charging times that
give you enough margin that you're safe. We you know,
(22:26):
I like to play with the NAV while in motion,
hoping to optimize for a shorter overall trip. It's just
it's a long trip and if you go based on
the system, it will add it's hours longer than it
needs to be. So we will add extra charging stops
sometimes with shorter charges to see if we can't game numbers.
Speaker 2 (22:49):
Shave off some time. Okay, not to go too deep
on this, but it's fascinating. It's real world practical stuff. Food.
What do you did you pack? Food? Did you stop
for a deep Uh?
Speaker 3 (23:00):
Well, we try. One tries to bring healthy food because
it's a it's a long trip and i'm you know,
getting up there, and my cardiologist is strict, so yeah,
I mean we you know, see Fox, she's a former
professional athlete and very much into fitness, so she insisted
on a very healthy, healthy meal plan. So no fried foods.
(23:24):
We're not buying you know, crap at the stops. You're
packing you know, nuts, fruits, and some vegetable snacks, a
lot of water. I want to stay away from coffee
and energy drinks because they they're not just unhealthy, they
also you can affect your cardiovascular system, which can change
your mood and your reaction times. And you don't want
(23:44):
you want to have a as much as as you
can a kind of flat cognitive you know, kind of
behavioral envelope, because you have to be monitoring the system
at all times, and especially when you have involuntary disengagement
warnings can occur at any time. When they do, you
don't want to be panicked or jerk the wheel. And
you also don't want to jerk the wheel when everything's
(24:05):
going well. It's very easy to accidentally disengage Tesla f SD.
Speaker 2 (24:11):
That's right. So I'm driving down the highway here and
they'll hey, put your hand back on the wheel. And
if I tuget it too hard, I'm out right right.
Speaker 3 (24:20):
You want to prevent that, And I'm at a point
I think of experience, it no longer happens to me. Johnson,
Joel Johnson is a very experienced driver and has quite
a bit of Tesla. Time's a He's a tall man
who's very fit, and so even let the weight of
his arm just landing on the wheels. Yeah, it's enough.
(24:42):
So he that happened a couple of times. We don't
count those. No, No, that's not systemat.
Speaker 2 (24:48):
That does that's a not counter. Don't count that one. Okay,
so we're cruising. Remind us again to say that again.
The numb total number on the most recent drive was x,
of which hel you were charging.
Speaker 3 (25:01):
All right, So on the most recent drive we had
fourteen disengagements, down from I think it's thirty two on
the first drive, and of which four were involuntary disengagements,
which is actually higher than the first drive. But all
(25:21):
pretty much almost all the other metrics were radically improved.
And the reason for this was because we encountered a
very serious well, I considered a very serious storm a
Tesla classified it as moderate, and we learned something that
a lot of people you know, hint at, but no
(25:41):
one I'm aware of has ever done, like a scientific,
public public study of the Tesla operational design domain. What
are the limits of when and where the car can operate?
And the limit is moderate rain is a hard boundary
for Tesla FSD moderate cart Well, I thought it was
(26:02):
heavy rain. However, Tesla added in UH recent software update
they added a weather Alive weather overlay to their GPS,
so UH on the map they you know, have shades
of green for severity of rain, and it indicated moderate rain.
And so I was quite confident we could drive into
(26:23):
it and we could go keep going with no effect
on completion of our drive. And as soon about two
thirds of the way into the drive, as we UH
enter that storm, FSD was basically unusable. It would it
began to uh give you give us involuntary disengagements with
(26:44):
no warning, over and over and over. So after that,
now it's like it was. It was the heart of
the hard alert. So the dashboard lights up red, it
says take over immediately, and you have.
Speaker 2 (27:00):
Second second wow, and yeah, Alex, this.
Speaker 3 (27:05):
Was I think Indiana. We're going west to east and
uh so approximately two thirds into the drive, and you know,
in the first drive we had two uh involuntary dis engagements,
and in this drive we also had two on the
early early stages, but once we hit this boundary we
had two minutes apart, followed by dozens. Oh. And so
(27:27):
for the purposes of comparing apples to apples and safety,
we decided to just declare the drive over and uh
and then study the data and declare that the Tesla
operational ziine domain is moderate rain. Uh not moderate rain.
And I don't mean, you know, like a little sprinkle.
I mean if there is a storm front classified as
(27:49):
moderate that is covering the entire geographic area over what
will be dozens or hundreds of miles, you have no
chance of going through that with no tes no chance
with hardware four, which is the current what goes to
current vehicles and FSD thirteen point two point two, which
is the current well it was current when we went
(28:10):
across Tesla's now at thirteen to seven is the latest
version which we'll be using to go across.
Speaker 2 (28:18):
Within Oh you're going to go again in a few weeks.
Speaker 3 (28:21):
So the project, which we call the Tesla FSD Cannibal
Run project, the project will continue. We will keep going
across as many times as necessary until Tesla's are capable
of driverless operation.
Speaker 2 (28:35):
Let's go back to the storm. I'm going to call
it a storm because I know you said moderate rain,
but I got you know that Midwest, those clouds gather
and it can get scared. So what broke down? Is
it the cameras can't see, the centers aren't operating, the
software is not digesting. What where did it go wrong?
Or do we know? So?
Speaker 3 (28:55):
Well? We I think we have pretty good information. So
the both the Flear camera was becoming obscured by frosts
and then the Tesla cameras were obstructed front and sides,
and we had warnings from the Tesla system saying camera obstructed.
(29:15):
You know clean now, you know, remove any debris. So
we pulled over multiple times to clean it. We tried
cleaning it with the windshield wipers. However, the FSD is
so sensitive that sometimes running the wiper over the flour
facing cameras will trigger a disengagement, which happened. We didn't
(29:38):
like that. So after multiple attempts to clean it and
we had rain nex, we had everything you could put
on that car was on that car, it became clear
that there was no way that we could continue. And
you know, within a few days of our canceling the
second drive, Tesla issued a bulletin advising all owners that
(30:00):
cleaning of the camera on the inside of the windshield
would be necessary for consistent and safe operation. And so
that was something previously I think that only occurred during
periodic maintenance, but now they actually update a level like
frequency that needs to occur. Now, what this tells us
is that the Tesla cameras are are not currently heated,
(30:24):
nor do they have any self cleaning capabilities, whereas all
the companies doing multimodal a V and ROBOTAXI the deploying
right now have some physical mechanical method of cleaning the
lenses on those cameras, and that is essential if one
(30:47):
is to be a four season robotaxi or a four
season probably on vehicles. But if you need to operate
a season all season, you need to need some active
technology that can resolve to breathe.
Speaker 2 (31:00):
That's some hardware. Some more hardware. That's hardware. Okay, let's
let's pivot real quickly to Austin because Elon in the
most recent call said robotaxis Austin unsupervised. It's on is
what's your take there when you heard him say that?
Speaker 3 (31:19):
So we do not yet know what hardware will go
on the cyber taxi. That's number one, number two Austin. Uh,
you know there is snow in Austin, So I have
to believe one of two things is gonna happen. They're
going to deploy a you know, a three and a
half season vehicle hardware, hardware, hardware that will work three
(31:42):
and a half seasons. So if they have just heated cameras,
that will probably that will maximize the ability to deploy
without an external and unattractive you know, wiper or jet
or there's many ways to do external mechanical cleaning. But
if they just go with the heating, that'll get them
most of the way there, because if they're going to
(32:03):
deploy a fleet, if it's Tesla Network, Tesla app, it's
theirs alone. If the odd of that vehicle excludes media,
heavy rain and medium snow, then people will not be
able to rely on that as a personal vehicle replacement
because on those days they need to get to work,
(32:25):
they won't be able to ride in a Tesla robotaxi, test,
a cybertaxi, cybercat. So they'll have to use something else,
either human driven Uber or something we haven't seen yet.
But Tesla could you know, they could have it both
ways by allowing potentially human driven Tesla's onto the Tesla
network to fill out the odd limitation of the cybercab.
(32:49):
Or they could just partner with Uber. And this is
what way most thinking. You have your Rubotaxi on the
Uber platform and then let people choose and so that
so the user no matter what, no matter what they
want to drive in, there's always an option. It's rebot,
it could be robotaxi. You know, three hundred and sixty
(33:10):
two days a year and the other three days they
don't operate because ven odi de limitation. They get a
human driver. So the a few ways to skin this,
I find I think it's unlikely Tesla will release something
that just won't work a few days a year, but
you know, the first year is a novelty. And I
think for the purposes of advancing Tesla's narrative or perpetuating
(33:33):
Tesla's narrative, that they could get by just by deploying,
you know, a dozen cars that operate most of the time.
That would be a major piece of news. And although
I used to work at r go AI and we
were a multi modal, lighter, you know, lighter equipped developer,
I do believe that one could deploy camera only solution
(33:56):
as long as you are as long as one understands
that comes with its own limitations.
Speaker 2 (34:01):
Did we cover the other disengagements beyond the storm? What
were the other events?
Speaker 3 (34:07):
Let's see, we had only one safety related disengagement. Tesla
FIST thirteen is dramatically better than f SC twelve in
terms of safety. Safety behaviors in a highway domain, you know,
which is generally lower complexity than urban and so we
had on this one, I think no truck laning incursions,
(34:29):
which was a major problem for the FST twelve. We
had one. I try to remember what it was. I'm
not sure I can remember what it was.
Speaker 2 (34:38):
But one over thousands of miles is correct.
Speaker 3 (34:41):
Correct, And yeah, the system is without a question in
improving the thing where there has not been or there
might have been a regression was in.
Speaker 2 (34:55):
I love that way you pivoted with that there wasn't
there might have been.
Speaker 3 (34:59):
Every well, well, it's unclear because we don't again, you know,
Tesla's not when I was at Argo, when something went wrong,
there was a team of people in the next morning,
you'd have a report and it would get fixed, then
you move on. Tesla doesn't share that information with users.
So we had on we had a system what's called
(35:20):
system reset. We had to really reboot the system in
motion and you don't know why, but you know, Tesla's
like a system error occurred, and so you have some
work to do. But it is without question, thirteen is
an improvement over twelve. And uh, you know, we'll need
more data and more drives to understand, to understand more
(35:40):
than the contours of the improvement, like what's the true
improvement rate?
Speaker 2 (35:44):
All right, outside of the storm. We're setting that aside
for a moment. If I was going to ask you
on a scale of one to ten, your degree of confidence, comfort,
relaxation in that vehicle driving it across the country, what
would you get? What number would you put there?
Speaker 3 (35:59):
You know? I use that every day, Yeah, like every
day in Phoenix and drive it along distance a lot.
I think it's probably about a six and point five
or seven that you know. But monitoring it with my
level of experience, yes, I think most people have a
false sense of security. And then there is a subset
of them who have the skill to save the car
(36:23):
from a bad situation.
Speaker 2 (36:24):
Okay, so here people don't. Someone with less experience might
have a higher number.
Speaker 3 (36:30):
Yeah, yeah, or the same, But the real question is,
you know, close calls and saves. You know, yesterday I
saw a Tesla cybertruck owner who's clearly a Tesla fan,
you know, confess that he had his FSD engaged at
thirty miles an hour. He wasn't paying attention. He usually does,
(36:51):
he claims, but this time he wasn't and the cyber truck,
the FSD drove him into a telephone poll and I
think it totaled the cyber truck, and so he said
it was my fault. I wasn't monitoring it, but it's
fist's fault for doing it, but it's my fault for
not catching it. This is a level of forgiveness. Yeah,
(37:12):
that what's the word. There's some there's some dissonance here
because two things went wrong, not just one. Yes, And
I think that a lot of Tesla ownors. I think
it's all cool until it isn't. And I think the
ones who hold large TESLA positions will forgive more than
they should.
Speaker 4 (37:30):
That's a great that's a great summary right there. Don't
forget more than they should. Alex always entertaining to listen
to you recount these stories.
Speaker 3 (37:40):
Well, but I should tell you to be very important
before we part ways. The overall system engagement rate improved,
and it's a very impressive number. The first trip it
was ninety eight point I think five to three percent
total time engaged, and on the second drive, up until
the point where we reached the odd boundary, we were
(38:02):
doing ninety nine point five to one percent system engagement.
It's that's it's a very ninety nine point five one
percent of the time. So that means.
Speaker 2 (38:11):
Help me, help me then sort out the gap between
that ninety nine plus and your six point five to
seven is that highway versus city.
Speaker 3 (38:19):
Or where the six point five to seven come.
Speaker 2 (38:22):
I think you gave me a six point five to
seven out of ten.
Speaker 3 (38:25):
Oh that's well, that's my confidence in the system. Six
point five to seven.
Speaker 2 (38:32):
But if you're scoring in ninety nine, you think I
would have anticipated you're going to say nine plus, but
you no.
Speaker 3 (38:38):
No, no. I Remember, a system could be on ninety
nine percent of the time. You could be absolutely terrified
by it all time, but you could also have it
on five percent of the time and give it a
ten for confidence. So the relationship between these things, it's
not an inverse correlation, uh, nor is it a direct correlation.
It's you know, it's uh confidence in dress wild systems on.
(39:02):
I am very what the system is on. In good conditions,
in a low complexity area, my confidence is a nine
point five out of ten, okay, without question, and then
it can only go down from there.
Speaker 2 (39:18):
Yeah right, okay. Place matters, location matters density.
Speaker 1 (39:24):
Yeah.
Speaker 3 (39:24):
Well, you know the thing about location is that a
lot of Tesla fans believe that way Mo's public declaration
of a geofence is a sign of weakness and also
an achilles heel in the business model and This is
a profound misunderstanding of how autonomus vehicles are developed because
geofence is a subset of the operational design domain, and
(39:46):
so there are multiple criteria for an ODD, and an
ODD can expand or compress in real time based on events.
So if if TESLA was, if FSD was a mature product,
or let's just say it was driverless capable today and
it could be today or it close to it, then
(40:06):
what they would need to do is declare transparently to
the user that the ODD boundary is contracting. So the GPS,
the route planner would look at the map at the
storm approaching, it would calculate the ETA to the storm
boundary and then it would say it would then project
that you know, twenty five miles prior to across the
(40:28):
storm boundary, you need to be awake and ready, or
it could be ten miles whatever, whatever do you want
to call it.
Speaker 2 (40:36):
And I think most people would be perfect. I would
be perfectly okay with that.
Speaker 3 (40:40):
God Yeah. And then if you are, and so you're
going to manually drive it through the storm, and then
as you the storm is thinning, it would give you
an estimate of when drivers operation can resume that. Yeah,
and so that is rebotaxi companies do this in real time,
except that it's transparent and invisible. Sorry, it's to the
(41:00):
passenger because the passers don't care. With the robotaxi companies
their products, the ux essential for loyalty is one where
you are only picked up and dropped off when the
vehicle can do so in your desired locations. You would
not wear. I want to have a robotaxi product. That
storm's approaching, they pick you up so that they can
(41:20):
book the ride because the business, you know, the economic
let's just say the corner report looks better having an
extra pickup, but that person was dumped five minutes later
in the side of the road because it was raining,
so you know. So that's that's difference between personal ownership
and fleet operation of avs.
Speaker 2 (41:39):
Alex this week's Future of Transportation newsletter by Riley Brennan,
in my mind, the single most valuable newsletter out there
in our business.
Speaker 3 (41:48):
Agreed.
Speaker 2 (41:49):
I'd say an entire paragraph devoted to you, and I
want to read it. Alex Roy's second FSD cannonball run,
the one we're talking about today, ended prematurely due to weather.
What he is doing. What Alex is doing with these
cannonball runs is more important than most of all motorsports
put together, very little which has any attachment to road cars.
Now ask yourself, will the winning car in this year's
(42:12):
Indy five hundred have more impact on how we're driving
in ten years or Alex Roy's cross country disengagement av runs.
How beautiful is that. That's a fantastic testimony to your work,
well deserved and not only fascinating to hear what you
went through, but also the real world implications for the
(42:33):
rest of us, Like how close are we to this
thing called unsupervised full self driving?
Speaker 3 (42:41):
I think if Tesla would you know, wanted to walk
or they wanted to deconstruct the narrative of sold to
Wall Street and shareholders and give us a simpler system,
I think that one could have driver less unsupervised f
(43:02):
SD you know, soon much sooner? Mm hmm. I think
this is probably Elon's call. If we can, I think
we could have it, you know, in a few months.
If the If the o d D started small, for example,
the supercharger, parking lots and.
Speaker 2 (43:21):
Around started to interrupt for listeners.
Speaker 3 (43:24):
O d D Operational Design domain means.
Speaker 2 (43:27):
That's that's a that's a basic yeah vernacular.
Speaker 3 (43:31):
Yeah yeah, okay, So that and that odd is you know,
lighting complexity, weather, time of day for those the location.
Uh it's or some you know, not limited to those,
but that those are the basic things. So if Tesla
just said, hey, h f SD is now driverless and
unsupervised within three blocks of a supercharger in good weather, uh,
(43:57):
you're good. You can literally get you, oh you want
to it out of your car now at this restaurant,
the car will go to the charger. That'll that of course,
is only useful if there's someone to plug it in
for you at the charger. That's a different problem.
Speaker 4 (44:10):
But it wouldn't be that crazy elon, it's at the charge,
it's not charging yet, I'm at the restaurant.
Speaker 3 (44:17):
Uh huh, I could see, you know, there are there
are a couple of startups I've seen who are you know,
who want to you know, inter basically be in a
uh convenience layer to FSD when it's driverless, and uh,
it is entirely possible that one of them could be
a guy at each supercharger who just plugs him in
(44:39):
for you. And if that with some kind of O
D D boundary that customers will accept. They could go
driver less soon, but it's going to be limited and
then it'll go from there. So Tesla has always been
a maximalist narrative company, will work everywhere, all the time,
don't worry about it, and so, and then behind that
maximalist now arrative, they're doing limited things, whereas Weimo and
(45:03):
everyone else on Earth is doing a minimum viable product
narrative and then slowly expanding what that minimum viable product is.
And so expectations for Weimo's everybody else are low. Or
because we only see the geographic limitation, you don't see
(45:25):
the other lipmard. I want to see limitations areas of
the odd boundary as they expand a contract, whereas so
in Tesla's case, we only see the big promise and disappointment.
And then so you bring your lens to that narrative
based on.
Speaker 2 (45:41):
Who you are. Final question for you, Alex. Early on
you talked about the relationship between Uber and Weimo, so
they're having a kind of partnership. Some people, there seems
to be two schools of thought. Some people think that
Uber will prevail. Others think that a driver list, robo
taxi way, most system will prevail. What's your view on
(46:03):
where that goes?
Speaker 3 (46:05):
Uh? Well, Riley Brennan, I think someone said this of Riley.
They said that the if it's the Riley Brennan rule
or the maxim is that whomever owns the point of
sale is the most important position in any vertical. And
(46:25):
you know right now Uber is they are the Kleenex
and of hailing a vehicle, it's them, and so I
think that ship has sailed. And so Uber no mystery
where they hired Dara to be to run the show.
He ran Expedia, which is Expedia. They're not the they're
(46:47):
not the Kleenex of travel, but they are one of
a couple of companies that do the same thing, totally
agnostic on product. So Uber will be a winner in
autonomy because they will be the the the place where
you can get anything you want. And I think it
will have many avs on the platform. So Weimo is
(47:07):
like the way what Expedia is to travel and aviation.
Weimo is like an airline that happens to make his
own its own planes h and you will have many
such companies on Uber and uh, Weymo will make more
money if you book direct, but most people won't. They
(47:27):
will go through the Uber app.
Speaker 2 (47:29):
The convenience because today I want the Weimo vehicle tomorrow
morning and take a different.
Speaker 3 (47:34):
One the way. People are smart. They understand that the
progress in the capital this free market system is measured by,
or progress of a product is measured by how much
friction does it reduce for the customer. And so if
if I live where weimos operate, and I do, I
(47:56):
don't want a human driven Uber. I want a Weimo
and nothing else. But I happen to know that weaymos
are available twenty four to seven where I live, so
I just go straight to the Waimo apps.
Speaker 2 (48:06):
Attractive to you? About Waymo? Why do you say that
you want Waymo driver list rather than all right?
Speaker 3 (48:13):
So clean? Huh? Consistent?
Speaker 2 (48:16):
It is important, isn't it clean?
Speaker 3 (48:18):
Yeah? Consistent? I can I get to sit up front.
Vehicle behavior is universally terrific, comfortable, safe, Whether it's amash
we safe, We're perceptibly safe. I don't know. I don't
care because it is it drives a speed limit, feels great,
(48:39):
and so everything about it appeals to me.
Speaker 2 (48:41):
It's the vehicle. There's a pacifica or in Phoenix.
Speaker 3 (48:46):
It's just the Jaguar I pays, okay, so love it.
Couldn't love it more. And so if I'm in a
city where Weymo's available, but I don't know how many
or where or when, I'll just go to the Uber
app and then if there's a weymo option, I'll pick it,
but if not, I'll take something else. And so, because
I don't want to have to open two apps, so
(49:08):
as Wimo's become more ubiquitous, I'll probably just default to
the Uber app just if I'm out of town, and
if I'm home for everything, Waymos only exists and they
only have one product band currently, which is a premium vehicle,
and in time, like Uber themselves, Waymo will offer more
vehicles and more bands, and then all these platforms years
(49:31):
from now will just be you know, every AV developer,
every AV company will have multiple bands. And then the
only reason you'll be in a human driven vehicle is
because the weather's bad and or a new class of
service and product have emerged, which would be people who
(49:53):
absolutely want a human in the car even if they're
not driving. Could be a tour guide, could be an
elderly person who wants an assistant, could be any number
of things. That day will come, and so uber will
always be a winner here because they offer a great
They'll be able to offer every great product there is
to any customer with less friction than having to cross shop.
(50:15):
You know, I don't know who the competitors to Weymo will.
Speaker 2 (50:18):
Be, or weimo introduced some humanoid into the driver's seat
for conversational purposes and help you in and out of
the car.
Speaker 3 (50:28):
I mean, I think there are people who will want
a human around to do stuff and well, and there
are the new startups will emerge to service that need.
Speaker 2 (50:36):
Phenomenal Alex, I always enjoy our conversations and I enjoyed
listening to you for two reasons, one original thinking and
two a man of action. You don't find that combination
very often in the world, and that's why I appreciate
you so much.
Speaker 3 (50:52):
Well, thank you. That's why I love your show. You
too go to the source to learn facts, that's spec facts,
and then teach us something. So that's why I love
your show.
Speaker 2 (51:04):
Awesome, Alex Troy, thank you so much for joining us.
How do people get in touch?
Speaker 3 (51:09):
My website is new Industry, Venture Capital ni VC dot.
Speaker 2 (51:14):
US beautiful, all right, all the best, keep us up
to date on the next run. When will that be
and what will that be? East to west?
Speaker 3 (51:24):
It'll be east to west. And details when I have them.
Speaker 2 (51:29):
Okay, details when you have them, Alex Troy, We'll see
you soon. So the next time we're out with friends
or at a party and we want to sound really
(51:50):
up to date on Tesla's FSD systems, what exactly can
we say that we know for certain? You know, hey,
listen up, guys, I have the intel. Here. There are
three takeaways. First, confirmed, a team of people have taken
the Tesla model esque coast to coast that three thousand
miles in FSD mode and had fewer than a dozen
(52:11):
disengagements along the way. No matter how you look at it.
That's pretty damp impressive. Second, the single greatest impediment to
perfection and autonomy, well, it's heavy rains or snowstorms, the
things that block the cameras. Mother nature is stubborn. Confirmed.
And Third, this one I really like is the question
isn't really when will we have full self driving cars?
(52:35):
But where this thing called odd operational design domain is key?
Where are these vehicles deployed matters. By defining those areas,
Tesla and weay Mood could soon be making millions of
trips without drivers, provided that they are operating within that
defined area. Tesla will be starting Robotaxi rides in Austin
(52:57):
in June, and Elon Musk feels confident that they'll scale
up to millions of rides a month in a hurry.
How cool is the hat? Let's see if they can
get it done. Hey, I hope you enjoyed this episode
of the Driving with Done podcast. Do send me your messages,
your questions DM me. I'm on x on LinkedIn, love
(53:19):
to get your notes. My name is Michael Donn and
this is the Driving with Done podcast.
Speaker 1 (53:26):
Where you meet the experts creating the technologies that will
power tomorrow's cars electric autonomous software. To find this is
a Driving with Done podcast. Thank you for joining this
episode of the Driving with Done podcast. To connect with
Michael Donn, visit Doneinsights dot com or find Michael on.
Speaker 3 (53:46):
X or LinkedIn.
Speaker 1 (53:48):
This is the Driving with Done podcast
We're top executives and crazy entrepreneurs gathered to talk about
the future of electric vehicles. This is the Driving with
Done podcast.
Speaker 2 (00:12):
Hello and welcome to the Driving with Don podcasts. I'm
your host, Michael Dunn. Now you ever wonder what it
would be like to just jump in your Tesla and
take off across the country in full self driving mode.
That is, no hands, no feet, the car drives the
car all by itself. Hmm wouldn't that be exciting, scary, thrilling, terrifying. Well,
(00:38):
guess what someone else has already been there and done
that twice. His name is Alex Roy. He's the co
founder of New Industry Venture Capital, and he and his
co pilots recently made two trips coast to coast California,
New York in his brand new Model Less. Why what
were they all about? Their mission measure just how good
(01:00):
Tesla's full self driving capabilities are today in twenty twenty five. Now,
Roy is no stranger to testing the limits of vehicles.
He is the record holder for the electric car cannonball
run that he achieved in his own Tesla several years ago.
Now he's turned his attention and his expertise to autonomous driving.
(01:24):
His question, how soon will we be able to ride
in the backseat of a car, sleeping, take a nap,
and have the car deliver us coast to coast. I
first met Alex Roy in twenty eighteen at a private
industry event south of Munich. He's one of a kind, colorful,
smart and professional. I just know you'll enjoy his vivid
(01:44):
first hand account of these cannonball runs, both the triumphs
and the reversals. Reversals. Yeah, like the time the team
encountered a good old fashioned Midwestern downpour thunderstorm. It was
in Indiana. I wasn't there, but I picture rains coming
down like a cow pissing on a flat rock. How
(02:04):
exactly did the model esque respond to that water deluge?
Let's find out with Alex Roy, partner New Industry Venture
Capital on the Driving with dun podcast. Alex Roy, the
(02:32):
one and only. Thank you so much for joining me
today on the Driving with Done podcast.
Speaker 3 (02:36):
Great to see you, Michael, Love Coming and Shell love
your show. Love you Love Coming.
Speaker 2 (02:40):
On the first day back from way across the Pacific
in a country called New zeal Zealand, Yeah, what was
going on over there?
Speaker 3 (02:49):
Well, my my friend, and well I think he's your
friend too. Rob Conybeer of Shastadventures loves going out in
New Zealand. He organized a group of American investors to
go down and meet with local investors and local startups
UH to familiarize you know, Americans with the tech ecosystem
and meet some members of the government. We met the
(03:11):
Prime Minister and the head of Immigration and Education and
a lot of interesting folks UH to encourage investment and
and to witness an announcement of a new UH, new
policy which makes it easier for foreign investors to come
invest and potentially by real estate and get you know, visus.
(03:32):
So it was my first time down there.
Speaker 2 (03:35):
Any sightings of evs or a v's in New Zealand.
Speaker 3 (03:38):
Not a single AV in the country I'm aware of,
but a lot of a lot of hybrids every cab
as a prius, I wrote it a B y D
which is cool and UH was just blown away by
the culture, the people, the governance UH. And I didn't
know there was a startup ecosystem down there as healthy as.
Speaker 2 (03:56):
We knew, not at all.
Speaker 3 (04:00):
What a fantastic place. And Rob put a great, a
great event together fantastic. I can't wait to go back.
Speaker 2 (04:06):
We're here to talk about autonomous vehicles and in particular
you're cross country run. But one more point on New
Zealand before we leave. You mentioned riding in a BYD.
Impressions highs and lows.
Speaker 3 (04:20):
Very nice, really well made. The interior was groovy in
a way that only kind of renault like modern renaults
are groovy. Maybe it had a little flavor of like
cicro en, like the what is it the C four
which has the goofy the goofy exterior rubber panels. The
(04:40):
interior of the BYD. The rear passenger door release was
is a curved. It's like a curve. It's a crescent
which rotates around the mid range and tweeter. Assembly for
the rear passenger.
Speaker 2 (04:58):
Is good or not good, I don't know.
Speaker 3 (05:01):
Huh uh. It's probably easily ripped off the door if
one it gets feisty. But I thought it was clever.
I thought it was clever, It was cute, and uh
it was. It was a good car. The driver loved it.
He absolutely loved that car. So my first time into
b Idea, I was impressed.
Speaker 2 (05:18):
Now bigger than Ford, bigger than Nissan bigger than Honda
chasing GM. Wow.
Speaker 3 (05:25):
Uh you know, I think the reputation is probably well
earned from everything I've seen and heard, And of course
I'm going to have some interesting announcements I think later
this year around a Chinese vehicle. But that's for another show,
not this episode.
Speaker 2 (05:41):
Okay, you're right. This episode take us back to the beginning.
Coast to coast time and place. You're it's it's the
Tesla FSD cannonball run across the United States. Where did
you start?
Speaker 3 (05:57):
We start? We started, well, we've done a couple of time.
So the first time we started at the Red Ball
Garage in midtown Manhattan and ended at the Portofino Inn
in Redondo Beach, California. On the second drive, we did
it in reverse order, same locations. The point of it
is that there are very well, first of all, there
are very few people who are doing really authoritative reviews
(06:23):
of driver assistance, and then there are basically no reviews
of driverless vehicles like robotaxis. And so since Tesla full
self driving software is the world's most anticipated and popular
such package, we thought that someone should take Tesla and
(06:46):
their claims of imminent readiness to go driverless. Let's take them,
and let's take those claims, and let's actually put some
metrics behind that. And I had not seen anyone do
that in a meaningful way. So since the nurber ring
is has been in the metric your Nerve Burgering time
is the metric for every human driven car that is
(07:06):
remotely fun to drive. Everyone takes their cars that come
up with a metric. I thought, what is the what
is analogous to that for autonomous vehicles? And so there
we be two. If it's a personally owned vehicle capable
of driverless operation, then it has to be the cannonball
run route New York to LA Or and back. And
(07:28):
then you would take that route because it's it's something
anyone can do, and you would drive it over and
over with every software release or hardware upgrade and watch
the metrics evolve over time, and then you could begin
to learn the rate of improvement of the system performance,
whether it's software hardware changes in a domain which is
clearly recognizable and also very famous. And so a few
(07:52):
months ago Joel Johnson and C Fox and I friends
from Porsche and my consulting life, decided we were going
to go do this.
Speaker 2 (08:01):
Three of you guys, one vehicle.
Speaker 3 (08:04):
Two of Only two of us have gone at a time,
although we'll probably go three of us on the next drive.
And we would take my brand new Tessa Model S
and we would fill it with every possible piece of
hardware after market we could to measure everything. Everything, We
everything everything, and then on every subsequent drive we would
(08:24):
add or subtract something to improve our data gathering.
Speaker 2 (08:27):
Everything related to autonomous driving or everything related to everything
like EV range times what.
Speaker 3 (08:35):
Well, not really interested in the EV range because there
are people who only do that. They and if you
just do that, you will get more, you'll get better
information than a catch all drive. We are interested in
the automation, autonomy and rate of improvement towards driverless operation.
So the first drive, which was in mid December of
(08:55):
twenty twenty four, we got Fleer Teledine Flair, which is
you know, big supply of thermal cameras and imaging equipment
to sponsor us in the form of a pre production
prototype Flear camera, which will be I think it's going
on an OEM vehicle this year or next year.
Speaker 2 (09:12):
How do you spell that?
Speaker 3 (09:13):
Fl I R and I have a lot of experience
with FLEAR because I used a Flear in two thousand
and five and six on my BMW to set the
original Cannonball run record, and that made it a lot
safer to drive at one hundred and fifty miles an
hour in the dark and spot animals or police. But
(09:33):
in this case, since Tesla is a camera only system,
we thought it'd be it would improve our safety to
have a Fleer camera mounted on the front of the
car with the display mounted on top of the dash.
That would show us thermal imagery, which is something that
Tesla cannot do, and that would be a backstop. Will
(09:54):
the Flear potentially see something that the Tesla's you know,
passive traditional camera do not see, and by recording using
a GoPro over our shoulders facing forward, in the event
of an incident of any kind it could be a
fantom breaking incident or heaven forbid, a near miss or
even a collision, then we could correlate what the Fleer
(10:16):
sees to what the Tesla sees and understand whether the
Tesla might have performed better if it had a let's
just say, a different or multi modal sensor package. And
so so this Flear.
Speaker 2 (10:30):
Does it give you more distance and sees through the
dark or what's its secret?
Speaker 3 (10:35):
It can see temperature. You know, it's a thermal thermal sensor,
so it can see at night. It will see well
beyond the headlights of the car, well beyond in the daytime.
You know, it's it's pretty much the same. But what
it will do, which was fascinating on the second drive,
is it can see through you know, fog rain, you
(10:58):
can see through obscurations and so. And it also has
built in even the pre production unit has bounty boxes
and classification of objects. So when we were looking out
the window and couldn't see very far and even the
Flear display was obscured, it would still show bounding boxes
and classifications even on the basically missed it over just
(11:21):
gray background. So it's a very impressive unit. And I
you know, I'm not paid by Flear, but I wish
to me this is the magic sensor that every car
should have because they're very inexpensive.
Speaker 2 (11:33):
Where do you mount it? And how big is it?
Speaker 3 (11:35):
It's a pencil cam. It's like the size of this
little pencil in my head that I don't know. I
think it's like a couple of inches long. I had
it mounted to my front license plate bracket and it
runs a cable to the pasture compartment and connects to
a screen. I think they're they're looking at like eight
ass applications, but honestly, I think it's a great it's
a great technology. I'd like to see it used. And
(11:58):
so in in December we went on our first drive
and that was kind of just a benchmarking drive, and
on that one we were measure looking for two metrics,
because the metrics are not time for that's for something else,
(12:19):
traditional canniball records. And in the future, when driverless operation
is capable is possible with two or more companies, then
you would put them head to head, like how fast
can they go across? But for our purposes, we want
to measure the rate of improvement towards driverless operation, meaning
unsupervised operation, and to do that you need to gather
(12:40):
data on two metrics. The first one is number of disengagements.
The goal is zero. It has to be otherwise it's
not driverless, and you need in vehicle supervision. Potentially you
could have remote supervision, but at highway speeds not realistic
with the current state of the art, or at least
not in the eight not at eighty miles an hour.
Speaker 2 (13:02):
So to put some color on this unsupervised, I think
it was you to coin the term like if you're
sleeping in the backseat of the car, that qualifies.
Speaker 3 (13:10):
Otherwise you're correct, correct, because I mean it is. Every
autonomous vehicle will have a customer service function and some
remote operations function, and they're the goal of the companies
is to reduce that supervision or operations function, the remote
one to never but periodic and something that can be
done safely, which means that there has been low latency,
(13:34):
low latency and you know, a high ratio of vehicles
managed to per remote operator. So if there's if there's
a if it's one to one and someone's got a
steering wheel in an office, you're going you're going out
of business. That's not good. That doesn't scale. And so
that's the first thing is number of disengagements. And then
the second metric is percent system engagement rate. So over
(13:56):
your entire time driving, what percentage of that time is
the system engaged? Because if you have a dis engagement
and currently you're going to have a bunch, how long
in between until you can re engage the system. And
so this is really important because if you have a disengagement,
you want it to be if possible seconds seconds. And
(14:19):
in the event that these vehicles become unsupervised and but
still have a remote operations function that's necessary, well, they
want to minimize that too, because the shorter the time
the remote operators engage with a vehicle in motion, the
better and safer. So those those are the numbers.
Speaker 2 (14:37):
Can I jump into a couple of questions, Yeah, yeah,
when we talk about remote operator in Tesla's case, that
does that exist?
Speaker 3 (14:46):
So Tesla has long stated that they're just going to
be They're gonna be autonomous, and they have avoided the
topic of any remote operations. Customer service function however, uh,
there is strong indication that they're hiring people for an
operations team, and whether or not they admit it, customer
(15:07):
service is requires a remote operations function because eventually the vehicle,
every autonomous vehicle, like every technology, will have an issue MM.
Speaker 2 (15:18):
So that's a great point. So thing go one without
exception is going to have an issue.
Speaker 3 (15:23):
With exception, and especially if if the cyber taxis don't
have steering wheels and that thing accounts as a scenario
cannot resolve. Someone's got to get that thing moving again,
whether or not that's a passenger inside. So who's doing that?
Because a passenger can't do it if there's no steering
wheel path, and even if there was way, MOO doesn't
want you getting behind the wheel and moving the car.
(15:45):
So remote operations is a customer service function. There's no
product made in the history of modernity that doesn't have
a customer service function of some kind. And so, uh,
we'll see what tesla how tesla resolves at later. Right now,
the owners of the vehicles are performing this function.
Speaker 2 (16:03):
You are You didn't know it, but you are the
remote operator inside the camp.
Speaker 3 (16:07):
Yeah, tester, supervisor, whatever you want to call it.
Speaker 2 (16:10):
Yeah, all right. So you're measuring two things, number of
disengagements and the length of time between getting disengaged and
getting re engaged. Yeah, okay, So what does what is
a disengagement any example? What does one look like for
someone who hasn't been in your seat?
Speaker 3 (16:28):
What does it look at the highest level, there are
two types of disengagements, involuntary and volunteery. Involuntary would be
that the system, for whatever reason can no longer drive
and it then says, you know, take over. In a
perfect world, the system would know that would warn you
(16:49):
of a necessary takeover, you know, minutes ahead of the driver,
having the supervisor happy to do so. When you know,
when you are not engaged in the driving task, you
begin to zone out. It's very hard not to, especially
over many hours. In our case. You know, we're looking
we're talking in access of forty hours behind the wheel,
and so studies have shown that it may take up
(17:11):
to forty seconds. Is how long it may take someone
behind the seat to cognitively re engage safely and take
over if you're sitting right there in your task. Is
just that it could be milliseconds. Okay, So that's an
involuntary disengagement is no bueno. You don't want it, but
they happen. And then the other one. The voluntary disengagement
(17:37):
can happen for a variety of reasons. It could be
that the vehicle is performing a maneuver that's illegal and
we don't want it to proceed because the liability falls
on us if we're pulled over. It could be for
the vehicle approaching an unsafe scenario and it is and
it's unclear that the vehicle is going to resolve it safely.
(17:58):
An example would be that a truck is entering into
your lane, uh, and it's doing so very closely, and
the vehicle, your vehicle, the Tesla, does not appear to
be breaking in time and colision might be imminent. So
everybody has a pucker factor, it's a little different on
that one.
Speaker 2 (18:17):
And then how are you on to scale a one
to ten on that puck that truck.
Speaker 3 (18:23):
I'm pretty good. I'm pretty good because I've used so
many driver resistant systems, have been in Sony astonomouses.
Speaker 2 (18:28):
You come to trust it more than a first time
or would Yeah.
Speaker 3 (18:32):
I mean this is the the devil's The devil's math
of testing or supervising test vehicles is you know, what
is your pucker factor? Because if you if you disengage
every time you think the vehicle might not get it,
you're gonna have hundreds of disengagement and but if you
never do, you will eventually have a crash. So you
(18:53):
I like, I mean, I'm always lean towards how we
let the vehicle go as far as it can until
it's imminent and then take over. Well, it's important also
to note that not everyone is equally capable of taking
over and resolving those scenarios well or safely.
Speaker 2 (19:10):
And your average driver in America probably wouldn't want to
have anything to do with it because it's just it
can be a little bit terrifying that truck boots over
some carbs moving into your lane. How much do I
trust this vehicle.
Speaker 3 (19:22):
That is correct? It can be hair raising. I have
good news on that score when we get to the
actual results. Another type of voluntary disengagement would be navigation
related where the GPS indicates that the vehicle is going
to go here, but the vehicle, whatever reason, doesn't actually
follow what's on screen that the nav system and the
(19:42):
vehicle like motion planner don't decide that they're going to disagree,
and the vehicle makes a left turn instead of a right.
So which point you have to choose. Do you let
it play out and see if it makes the U
turn or do you just let it keep going in
the wrong direction? And I like to you know, if
it's a quick turnaround, well, I think I like to
(20:03):
see if the vehicle can resolve it. And then the
last sip of disengagement would be for charging. So on
the first drive, Tesla FSD twelve five six four was
the software version we used. That that version was not
capable of navigating into and out of Tesla supercharger stalls,
(20:26):
so we had to manually take a voluntary takeover for
each of those those stops.
Speaker 2 (20:31):
Why do you think that is if you were just
pulling into a regular parking lot, would have been able
to do so? Or is there something specific about the
supercharger station.
Speaker 3 (20:39):
I think it's specific to the supercharger stations because Tesla
has had parking functionality for some time, which is with
mixed mixed results, but they've never they did not They
never claimed supercharger capability with FSD version twelve. They did
for version thirteen, which we use in the second drive.
Speaker 2 (21:00):
You saw a difference.
Speaker 3 (21:01):
Oh yes, So on the first drive we had a
total of thirty one disengagements, of which twenty one were
for charging. Is that right? Thirty two disengagements with twenty
one for charging.
Speaker 2 (21:14):
Total you're talking total disengagements.
Speaker 3 (21:16):
Yeah, between two thousand between New York and La.
Speaker 2 (21:20):
Two thousands. Phenomenal.
Speaker 3 (21:21):
Yeah, if you take away the charging, yeah, it's quite good. Yeah,
less than a dozen disengagements of all types and and
on a second drive, the numbers were radically improved. So yeah,
the vehicle was capable of pulling into a charge is
successfully almost every time. It's instead of twenty one disengagements,
(21:43):
we had eight for charging, which is extraordinary. And the
other metrics.
Speaker 2 (21:49):
So we're out. You're rolling down Highway eight or ten
or forty whichever one, going east to west, and you
you're you're how much of this is the caught you saying,
I'm driving along the car chooses a super charge station
or do you guys choose it?
Speaker 3 (22:05):
So when you Tesla has an extraordinarily good GPS and
a charging planner, So when you are in New York
at the start line and you select your finish line,
it will prepopulate all the chargers across the whole way, country,
the whole way with as you know, charging times that
give you enough margin that you're safe. We you know,
(22:26):
I like to play with the NAV while in motion,
hoping to optimize for a shorter overall trip. It's just
it's a long trip and if you go based on
the system, it will add it's hours longer than it
needs to be. So we will add extra charging stops
sometimes with shorter charges to see if we can't game numbers.
Speaker 2 (22:49):
Shave off some time. Okay, not to go too deep
on this, but it's fascinating. It's real world practical stuff. Food.
What do you did you pack? Food? Did you stop
for a deep Uh?
Speaker 3 (23:00):
Well, we try. One tries to bring healthy food because
it's a it's a long trip and i'm you know,
getting up there, and my cardiologist is strict, so yeah,
I mean we you know, see Fox, she's a former
professional athlete and very much into fitness, so she insisted
on a very healthy, healthy meal plan. So no fried foods.
(23:24):
We're not buying you know, crap at the stops. You're
packing you know, nuts, fruits, and some vegetable snacks, a
lot of water. I want to stay away from coffee
and energy drinks because they they're not just unhealthy, they
also you can affect your cardiovascular system, which can change
your mood and your reaction times. And you don't want
(23:44):
you want to have a as much as as you
can a kind of flat cognitive you know, kind of
behavioral envelope, because you have to be monitoring the system
at all times, and especially when you have involuntary disengagement
warnings can occur at any time. When they do, you
don't want to be panicked or jerk the wheel. And
you also don't want to jerk the wheel when everything's
(24:05):
going well. It's very easy to accidentally disengage Tesla f SD.
Speaker 2 (24:11):
That's right. So I'm driving down the highway here and
they'll hey, put your hand back on the wheel. And
if I tuget it too hard, I'm out right right.
Speaker 3 (24:20):
You want to prevent that, And I'm at a point
I think of experience, it no longer happens to me. Johnson,
Joel Johnson is a very experienced driver and has quite
a bit of Tesla. Time's a He's a tall man
who's very fit, and so even let the weight of
his arm just landing on the wheels. Yeah, it's enough.
(24:42):
So he that happened a couple of times. We don't
count those. No, No, that's not systemat.
Speaker 2 (24:48):
That does that's a not counter. Don't count that one. Okay,
so we're cruising. Remind us again to say that again.
The numb total number on the most recent drive was x,
of which hel you were charging.
Speaker 3 (25:01):
All right, So on the most recent drive we had
fourteen disengagements, down from I think it's thirty two on
the first drive, and of which four were involuntary disengagements,
which is actually higher than the first drive. But all
(25:21):
pretty much almost all the other metrics were radically improved.
And the reason for this was because we encountered a
very serious well, I considered a very serious storm a
Tesla classified it as moderate, and we learned something that
a lot of people you know, hint at, but no
(25:41):
one I'm aware of has ever done, like a scientific,
public public study of the Tesla operational design domain. What
are the limits of when and where the car can operate?
And the limit is moderate rain is a hard boundary
for Tesla FSD moderate cart Well, I thought it was
(26:02):
heavy rain. However, Tesla added in UH recent software update
they added a weather Alive weather overlay to their GPS,
so UH on the map they you know, have shades
of green for severity of rain, and it indicated moderate rain.
And so I was quite confident we could drive into
(26:23):
it and we could go keep going with no effect
on completion of our drive. And as soon about two
thirds of the way into the drive, as we UH
enter that storm, FSD was basically unusable. It would it
began to uh give you give us involuntary disengagements with
(26:44):
no warning, over and over and over. So after that,
now it's like it was. It was the heart of
the hard alert. So the dashboard lights up red, it
says take over immediately, and you have.
Speaker 2 (27:00):
Second second wow, and yeah, Alex, this.
Speaker 3 (27:05):
Was I think Indiana. We're going west to east and
uh so approximately two thirds into the drive, and you know,
in the first drive we had two uh involuntary dis engagements,
and in this drive we also had two on the
early early stages, but once we hit this boundary we
had two minutes apart, followed by dozens. Oh. And so
(27:27):
for the purposes of comparing apples to apples and safety,
we decided to just declare the drive over and uh
and then study the data and declare that the Tesla
operational ziine domain is moderate rain. Uh not moderate rain.
And I don't mean, you know, like a little sprinkle.
I mean if there is a storm front classified as
(27:49):
moderate that is covering the entire geographic area over what
will be dozens or hundreds of miles, you have no
chance of going through that with no tes no chance
with hardware four, which is the current what goes to
current vehicles and FSD thirteen point two point two, which
is the current well it was current when we went
(28:10):
across Tesla's now at thirteen to seven is the latest
version which we'll be using to go across.
Speaker 2 (28:18):
Within Oh you're going to go again in a few weeks.
Speaker 3 (28:21):
So the project, which we call the Tesla FSD Cannibal
Run project, the project will continue. We will keep going
across as many times as necessary until Tesla's are capable
of driverless operation.
Speaker 2 (28:35):
Let's go back to the storm. I'm going to call
it a storm because I know you said moderate rain,
but I got you know that Midwest, those clouds gather
and it can get scared. So what broke down? Is
it the cameras can't see, the centers aren't operating, the
software is not digesting. What where did it go wrong?
Or do we know? So?
Speaker 3 (28:55):
Well? We I think we have pretty good information. So
the both the Flear camera was becoming obscured by frosts
and then the Tesla cameras were obstructed front and sides,
and we had warnings from the Tesla system saying camera obstructed.
(29:15):
You know clean now, you know, remove any debris. So
we pulled over multiple times to clean it. We tried
cleaning it with the windshield wipers. However, the FSD is
so sensitive that sometimes running the wiper over the flour
facing cameras will trigger a disengagement, which happened. We didn't
(29:38):
like that. So after multiple attempts to clean it and
we had rain nex, we had everything you could put
on that car was on that car, it became clear
that there was no way that we could continue. And
you know, within a few days of our canceling the
second drive, Tesla issued a bulletin advising all owners that
(30:00):
cleaning of the camera on the inside of the windshield
would be necessary for consistent and safe operation. And so
that was something previously I think that only occurred during
periodic maintenance, but now they actually update a level like
frequency that needs to occur. Now, what this tells us
is that the Tesla cameras are are not currently heated,
(30:24):
nor do they have any self cleaning capabilities, whereas all
the companies doing multimodal a V and ROBOTAXI the deploying
right now have some physical mechanical method of cleaning the
lenses on those cameras, and that is essential if one
(30:47):
is to be a four season robotaxi or a four
season probably on vehicles. But if you need to operate
a season all season, you need to need some active
technology that can resolve to breathe.
Speaker 2 (31:00):
That's some hardware. Some more hardware. That's hardware. Okay, let's
let's pivot real quickly to Austin because Elon in the
most recent call said robotaxis Austin unsupervised. It's on is
what's your take there when you heard him say that?
Speaker 3 (31:19):
So we do not yet know what hardware will go
on the cyber taxi. That's number one, number two Austin. Uh,
you know there is snow in Austin, So I have
to believe one of two things is gonna happen. They're
going to deploy a you know, a three and a
half season vehicle hardware, hardware, hardware that will work three
(31:42):
and a half seasons. So if they have just heated cameras,
that will probably that will maximize the ability to deploy
without an external and unattractive you know, wiper or jet
or there's many ways to do external mechanical cleaning. But
if they just go with the heating, that'll get them
most of the way there, because if they're going to
(32:03):
deploy a fleet, if it's Tesla Network, Tesla app, it's
theirs alone. If the odd of that vehicle excludes media,
heavy rain and medium snow, then people will not be
able to rely on that as a personal vehicle replacement
because on those days they need to get to work,
(32:25):
they won't be able to ride in a Tesla robotaxi, test,
a cybertaxi, cybercat. So they'll have to use something else,
either human driven Uber or something we haven't seen yet.
But Tesla could you know, they could have it both
ways by allowing potentially human driven Tesla's onto the Tesla
network to fill out the odd limitation of the cybercab.
(32:49):
Or they could just partner with Uber. And this is
what way most thinking. You have your Rubotaxi on the
Uber platform and then let people choose and so that
so the user no matter what, no matter what they
want to drive in, there's always an option. It's rebot,
it could be robotaxi. You know, three hundred and sixty
(33:10):
two days a year and the other three days they
don't operate because ven odi de limitation. They get a
human driver. So the a few ways to skin this,
I find I think it's unlikely Tesla will release something
that just won't work a few days a year, but
you know, the first year is a novelty. And I
think for the purposes of advancing Tesla's narrative or perpetuating
(33:33):
Tesla's narrative, that they could get by just by deploying,
you know, a dozen cars that operate most of the time.
That would be a major piece of news. And although
I used to work at r go AI and we
were a multi modal, lighter, you know, lighter equipped developer,
I do believe that one could deploy camera only solution
(33:56):
as long as you are as long as one understands
that comes with its own limitations.
Speaker 2 (34:01):
Did we cover the other disengagements beyond the storm? What
were the other events?
Speaker 3 (34:07):
Let's see, we had only one safety related disengagement. Tesla
FIST thirteen is dramatically better than f SC twelve in
terms of safety. Safety behaviors in a highway domain, you know,
which is generally lower complexity than urban and so we
had on this one, I think no truck laning incursions,
(34:29):
which was a major problem for the FST twelve. We
had one. I try to remember what it was. I'm
not sure I can remember what it was.
Speaker 2 (34:38):
But one over thousands of miles is correct.
Speaker 3 (34:41):
Correct, And yeah, the system is without a question in
improving the thing where there has not been or there
might have been a regression was in.
Speaker 2 (34:55):
I love that way you pivoted with that there wasn't
there might have been.
Speaker 3 (34:59):
Every well, well, it's unclear because we don't again, you know,
Tesla's not when I was at Argo, when something went wrong,
there was a team of people in the next morning,
you'd have a report and it would get fixed, then
you move on. Tesla doesn't share that information with users.
So we had on we had a system what's called
(35:20):
system reset. We had to really reboot the system in
motion and you don't know why, but you know, Tesla's
like a system error occurred, and so you have some
work to do. But it is without question, thirteen is
an improvement over twelve. And uh, you know, we'll need
more data and more drives to understand, to understand more
(35:40):
than the contours of the improvement, like what's the true
improvement rate?
Speaker 2 (35:44):
All right, outside of the storm. We're setting that aside
for a moment. If I was going to ask you
on a scale of one to ten, your degree of confidence, comfort,
relaxation in that vehicle driving it across the country, what
would you get? What number would you put there?
Speaker 3 (35:59):
You know? I use that every day, Yeah, like every
day in Phoenix and drive it along distance a lot.
I think it's probably about a six and point five
or seven that you know. But monitoring it with my
level of experience, yes, I think most people have a
false sense of security. And then there is a subset
of them who have the skill to save the car
(36:23):
from a bad situation.
Speaker 2 (36:24):
Okay, so here people don't. Someone with less experience might
have a higher number.
Speaker 3 (36:30):
Yeah, yeah, or the same, But the real question is,
you know, close calls and saves. You know, yesterday I
saw a Tesla cybertruck owner who's clearly a Tesla fan,
you know, confess that he had his FSD engaged at
thirty miles an hour. He wasn't paying attention. He usually does,
(36:51):
he claims, but this time he wasn't and the cyber truck,
the FSD drove him into a telephone poll and I
think it totaled the cyber truck, and so he said
it was my fault. I wasn't monitoring it, but it's
fist's fault for doing it, but it's my fault for
not catching it. This is a level of forgiveness. Yeah,
(37:12):
that what's the word. There's some there's some dissonance here
because two things went wrong, not just one. Yes, And
I think that a lot of Tesla ownors. I think
it's all cool until it isn't. And I think the
ones who hold large TESLA positions will forgive more than
they should.
Speaker 4 (37:30):
That's a great that's a great summary right there. Don't
forget more than they should. Alex always entertaining to listen
to you recount these stories.
Speaker 3 (37:40):
Well, but I should tell you to be very important
before we part ways. The overall system engagement rate improved,
and it's a very impressive number. The first trip it
was ninety eight point I think five to three percent
total time engaged, and on the second drive, up until
the point where we reached the odd boundary, we were
(38:02):
doing ninety nine point five to one percent system engagement.
It's that's it's a very ninety nine point five one
percent of the time. So that means.
Speaker 2 (38:11):
Help me, help me then sort out the gap between
that ninety nine plus and your six point five to
seven is that highway versus city.
Speaker 3 (38:19):
Or where the six point five to seven come.
Speaker 2 (38:22):
I think you gave me a six point five to
seven out of ten.
Speaker 3 (38:25):
Oh that's well, that's my confidence in the system. Six
point five to seven.
Speaker 2 (38:32):
But if you're scoring in ninety nine, you think I
would have anticipated you're going to say nine plus, but
you no.
Speaker 3 (38:38):
No, no. I Remember, a system could be on ninety
nine percent of the time. You could be absolutely terrified
by it all time, but you could also have it
on five percent of the time and give it a
ten for confidence. So the relationship between these things, it's
not an inverse correlation, uh, nor is it a direct correlation.
It's you know, it's uh confidence in dress wild systems on.
(39:02):
I am very what the system is on. In good conditions,
in a low complexity area, my confidence is a nine
point five out of ten, okay, without question, and then
it can only go down from there.
Speaker 2 (39:18):
Yeah right, okay. Place matters, location matters density.
Speaker 1 (39:24):
Yeah.
Speaker 3 (39:24):
Well, you know the thing about location is that a
lot of Tesla fans believe that way Mo's public declaration
of a geofence is a sign of weakness and also
an achilles heel in the business model and This is
a profound misunderstanding of how autonomus vehicles are developed because
geofence is a subset of the operational design domain, and
(39:46):
so there are multiple criteria for an ODD, and an
ODD can expand or compress in real time based on events.
So if if TESLA was, if FSD was a mature product,
or let's just say it was driverless capable today and
it could be today or it close to it, then
(40:06):
what they would need to do is declare transparently to
the user that the ODD boundary is contracting. So the GPS,
the route planner would look at the map at the
storm approaching, it would calculate the ETA to the storm
boundary and then it would say it would then project
that you know, twenty five miles prior to across the
(40:28):
storm boundary, you need to be awake and ready, or
it could be ten miles whatever, whatever do you want
to call it.
Speaker 2 (40:36):
And I think most people would be perfect. I would
be perfectly okay with that.
Speaker 3 (40:40):
God Yeah. And then if you are, and so you're
going to manually drive it through the storm, and then
as you the storm is thinning, it would give you
an estimate of when drivers operation can resume that. Yeah,
and so that is rebotaxi companies do this in real time,
except that it's transparent and invisible. Sorry, it's to the
(41:00):
passenger because the passers don't care. With the robotaxi companies
their products, the ux essential for loyalty is one where
you are only picked up and dropped off when the
vehicle can do so in your desired locations. You would
not wear. I want to have a robotaxi product. That
storm's approaching, they pick you up so that they can
(41:20):
book the ride because the business, you know, the economic
let's just say the corner report looks better having an
extra pickup, but that person was dumped five minutes later
in the side of the road because it was raining,
so you know. So that's that's difference between personal ownership
and fleet operation of avs.
Speaker 2 (41:39):
Alex this week's Future of Transportation newsletter by Riley Brennan,
in my mind, the single most valuable newsletter out there
in our business.
Speaker 3 (41:48):
Agreed.
Speaker 2 (41:49):
I'd say an entire paragraph devoted to you, and I
want to read it. Alex Roy's second FSD cannonball run,
the one we're talking about today, ended prematurely due to weather.
What he is doing. What Alex is doing with these
cannonball runs is more important than most of all motorsports
put together, very little which has any attachment to road cars.
Now ask yourself, will the winning car in this year's
(42:12):
Indy five hundred have more impact on how we're driving
in ten years or Alex Roy's cross country disengagement av runs.
How beautiful is that. That's a fantastic testimony to your work,
well deserved and not only fascinating to hear what you
went through, but also the real world implications for the
(42:33):
rest of us, Like how close are we to this
thing called unsupervised full self driving?
Speaker 3 (42:41):
I think if Tesla would you know, wanted to walk
or they wanted to deconstruct the narrative of sold to
Wall Street and shareholders and give us a simpler system,
I think that one could have driver less unsupervised f
(43:02):
SD you know, soon much sooner? Mm hmm. I think
this is probably Elon's call. If we can, I think
we could have it, you know, in a few months.
If the If the o d D started small, for example,
the supercharger, parking lots and.
Speaker 2 (43:21):
Around started to interrupt for listeners.
Speaker 3 (43:24):
O d D Operational Design domain means.
Speaker 2 (43:27):
That's that's a that's a basic yeah vernacular.
Speaker 3 (43:31):
Yeah yeah, okay, So that and that odd is you know,
lighting complexity, weather, time of day for those the location.
Uh it's or some you know, not limited to those,
but that those are the basic things. So if Tesla
just said, hey, h f SD is now driverless and
unsupervised within three blocks of a supercharger in good weather, uh,
(43:57):
you're good. You can literally get you, oh you want
to it out of your car now at this restaurant,
the car will go to the charger. That'll that of course,
is only useful if there's someone to plug it in
for you at the charger. That's a different problem.
Speaker 4 (44:10):
But it wouldn't be that crazy elon, it's at the charge,
it's not charging yet, I'm at the restaurant.
Speaker 3 (44:17):
Uh huh, I could see, you know, there are there
are a couple of startups I've seen who are you know,
who want to you know, inter basically be in a
uh convenience layer to FSD when it's driverless, and uh,
it is entirely possible that one of them could be
a guy at each supercharger who just plugs him in
(44:39):
for you. And if that with some kind of O
D D boundary that customers will accept. They could go
driver less soon, but it's going to be limited and
then it'll go from there. So Tesla has always been
a maximalist narrative company, will work everywhere, all the time,
don't worry about it, and so, and then behind that
maximalist now arrative, they're doing limited things, whereas Weimo and
(45:03):
everyone else on Earth is doing a minimum viable product
narrative and then slowly expanding what that minimum viable product is.
And so expectations for Weimo's everybody else are low. Or
because we only see the geographic limitation, you don't see
(45:25):
the other lipmard. I want to see limitations areas of
the odd boundary as they expand a contract, whereas so
in Tesla's case, we only see the big promise and disappointment.
And then so you bring your lens to that narrative
based on.
Speaker 2 (45:41):
Who you are. Final question for you, Alex. Early on
you talked about the relationship between Uber and Weimo, so
they're having a kind of partnership. Some people, there seems
to be two schools of thought. Some people think that
Uber will prevail. Others think that a driver list, robo
taxi way, most system will prevail. What's your view on
(46:03):
where that goes?
Speaker 3 (46:05):
Uh? Well, Riley Brennan, I think someone said this of Riley.
They said that the if it's the Riley Brennan rule
or the maxim is that whomever owns the point of
sale is the most important position in any vertical. And
(46:25):
you know right now Uber is they are the Kleenex
and of hailing a vehicle, it's them, and so I
think that ship has sailed. And so Uber no mystery
where they hired Dara to be to run the show.
He ran Expedia, which is Expedia. They're not the they're
(46:47):
not the Kleenex of travel, but they are one of
a couple of companies that do the same thing, totally
agnostic on product. So Uber will be a winner in
autonomy because they will be the the the place where
you can get anything you want. And I think it
will have many avs on the platform. So Weimo is
(47:07):
like the way what Expedia is to travel and aviation.
Weimo is like an airline that happens to make his
own its own planes h and you will have many
such companies on Uber and uh, Weymo will make more
money if you book direct, but most people won't. They
(47:27):
will go through the Uber app.
Speaker 2 (47:29):
The convenience because today I want the Weimo vehicle tomorrow
morning and take a different.
Speaker 3 (47:34):
One the way. People are smart. They understand that the
progress in the capital this free market system is measured by,
or progress of a product is measured by how much
friction does it reduce for the customer. And so if
if I live where weimos operate, and I do, I
(47:56):
don't want a human driven Uber. I want a Weimo
and nothing else. But I happen to know that weaymos
are available twenty four to seven where I live, so
I just go straight to the Waimo apps.
Speaker 2 (48:06):
Attractive to you? About Waymo? Why do you say that
you want Waymo driver list rather than all right?
Speaker 3 (48:13):
So clean? Huh? Consistent?
Speaker 2 (48:16):
It is important, isn't it clean?
Speaker 3 (48:18):
Yeah? Consistent? I can I get to sit up front.
Vehicle behavior is universally terrific, comfortable, safe, Whether it's amash
we safe, We're perceptibly safe. I don't know. I don't
care because it is it drives a speed limit, feels great,
(48:39):
and so everything about it appeals to me.
Speaker 2 (48:41):
It's the vehicle. There's a pacifica or in Phoenix.
Speaker 3 (48:46):
It's just the Jaguar I pays, okay, so love it.
Couldn't love it more. And so if I'm in a
city where Weymo's available, but I don't know how many
or where or when, I'll just go to the Uber
app and then if there's a weymo option, I'll pick it,
but if not, I'll take something else. And so, because
I don't want to have to open two apps, so
(49:08):
as Wimo's become more ubiquitous, I'll probably just default to
the Uber app just if I'm out of town, and
if I'm home for everything, Waymos only exists and they
only have one product band currently, which is a premium vehicle,
and in time, like Uber themselves, Waymo will offer more
vehicles and more bands, and then all these platforms years
(49:31):
from now will just be you know, every AV developer,
every AV company will have multiple bands. And then the
only reason you'll be in a human driven vehicle is
because the weather's bad and or a new class of
service and product have emerged, which would be people who
(49:53):
absolutely want a human in the car even if they're
not driving. Could be a tour guide, could be an
elderly person who wants an assistant, could be any number
of things. That day will come, and so uber will
always be a winner here because they offer a great
They'll be able to offer every great product there is
to any customer with less friction than having to cross shop.
(50:15):
You know, I don't know who the competitors to Weymo will.
Speaker 2 (50:18):
Be, or weimo introduced some humanoid into the driver's seat
for conversational purposes and help you in and out of
the car.
Speaker 3 (50:28):
I mean, I think there are people who will want
a human around to do stuff and well, and there
are the new startups will emerge to service that need.
Speaker 2 (50:36):
Phenomenal Alex, I always enjoy our conversations and I enjoyed
listening to you for two reasons, one original thinking and
two a man of action. You don't find that combination
very often in the world, and that's why I appreciate
you so much.
Speaker 3 (50:52):
Well, thank you. That's why I love your show. You
too go to the source to learn facts, that's spec facts,
and then teach us something. So that's why I love
your show.
Speaker 2 (51:04):
Awesome, Alex Troy, thank you so much for joining us.
How do people get in touch?
Speaker 3 (51:09):
My website is new Industry, Venture Capital ni VC dot.
Speaker 2 (51:14):
US beautiful, all right, all the best, keep us up
to date on the next run. When will that be
and what will that be? East to west?
Speaker 3 (51:24):
It'll be east to west. And details when I have them.
Speaker 2 (51:29):
Okay, details when you have them, Alex Troy, We'll see
you soon. So the next time we're out with friends
or at a party and we want to sound really
(51:50):
up to date on Tesla's FSD systems, what exactly can
we say that we know for certain? You know, hey,
listen up, guys, I have the intel. Here. There are
three takeaways. First, confirmed, a team of people have taken
the Tesla model esque coast to coast that three thousand
miles in FSD mode and had fewer than a dozen
(52:11):
disengagements along the way. No matter how you look at it.
That's pretty damp impressive. Second, the single greatest impediment to
perfection and autonomy, well, it's heavy rains or snowstorms, the
things that block the cameras. Mother nature is stubborn. Confirmed.
And Third, this one I really like is the question
isn't really when will we have full self driving cars?
(52:35):
But where this thing called odd operational design domain is key?
Where are these vehicles deployed matters. By defining those areas,
Tesla and weay Mood could soon be making millions of
trips without drivers, provided that they are operating within that
defined area. Tesla will be starting Robotaxi rides in Austin
(52:57):
in June, and Elon Musk feels confident that they'll scale
up to millions of rides a month in a hurry.
How cool is the hat? Let's see if they can
get it done. Hey, I hope you enjoyed this episode
of the Driving with Done podcast. Do send me your messages,
your questions DM me. I'm on x on LinkedIn, love
(53:19):
to get your notes. My name is Michael Donn and
this is the Driving with Done podcast.
Speaker 1 (53:26):
Where you meet the experts creating the technologies that will
power tomorrow's cars electric autonomous software. To find this is
a Driving with Done podcast. Thank you for joining this
episode of the Driving with Done podcast. To connect with
Michael Donn, visit Doneinsights dot com or find Michael on.
Speaker 3 (53:46):
X or LinkedIn.
Speaker 1 (53:48):
This is the Driving with Done podcast
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com