---
title: 'Inside Self-Driving: The AI-Driven Evolution of Autonomous Vehicles'
source: 'https://youtube.com/watch?v=CbtbN4dyfwE'
video_id: 'CbtbN4dyfwE'
date: 2026-06-30
duration_sec: 3638
---

# Inside Self-Driving: The AI-Driven Evolution of Autonomous Vehicles

> Source: [Inside Self-Driving: The AI-Driven Evolution of Autonomous Vehicles](https://youtube.com/watch?v=CbtbN4dyfwE)

## Summary



## Transcript

[Music]
Hello everyone and welcome to Business
Insiders inside self-driving the
AIdriven evolution of autonomous
vehicles presented by Mobile Eye. I'm
Steve Russell, chief news editor here at
BI. And today we're diving into one of
the most transformative and debated
frontiers in technology. How AI is
turning autonomous mobility from a long
promised dream into a fast approaching
reality. We'll explore how automakers,
tech innovators, and policy makers are
working together to make autonomy safe,
scalable, and trusted, and what that
means for businesses, cities, and all of
us who share the road together. First,
we're starting today with a conversation
presented by our sponsor, Mobile Eye,
that goes inside the company's
collaboration with Lyft as they work to
bring driverless technology to scale.
[Music]
Thank you, Steve. I'm Dr. Deborah
Bervishes and I'm happy to be here. Robo
taxis already operate in a few cities,
but taking autonomous vehicles
mainstream comes down to three things:
auditable safety, consumer trust, and
economics that work. I'm joined here by
JJ Youngworth, executive vice president
of autonomous vehicles at Mobile Eye,
and Stephen Hayes, VP of autonomous
fleets and driver operations at Lyft.
Thank you both for being here. Let's
talk about trust and safety. It seems
like one of the main goals of autonomous
vehicle companies is to convince both
the writers and the regulators that
these kinds of vehicles are safe. So,
Stephen, at Lyft, your role is to
interact directly with a rider. What
would they need to see or experience to
feel comfortable using an autonomous
vehicle?
>> Great question. Uh, at Lyft, our purpose
is to serve and connect, and that's
something that we do uh about 800
million times a year, helping riders get
to where they need to go. Uh, and over
time, AVs are going to make up a bigger
and bigger percentage of those trips on
our platform. And uh if you are tuned
into this broadcast, chances are you are
a bit of a tech enthusiast and an early
adopter. But the reality is for most of
the people who open up the Lyft app on a
daily basis, they're just looking to get
to where they need to go. Uh and that's
where uh it is our privilege and
responsibility to introduce millions of
new riders to exciting autonomous uh
technology. And in order to do that
effectively, we need to find the right
autonomous trip for the right rider at
the right time. And then once they come
in, uh we need to educate them about the
experience that they're going to have.
Uh and make sure it's a delightful one.
And all the while, what is going to be
really important for us is making sure
that we have happy uh repeat customers
who are getting where they need to go
even more quickly and efficiently than
they are today.
>> Awesome. JJ, from your perspective at
Mobile Eye, which three pieces of proof
would you hand to a regulator to prove
that autonomous vehicle technology is
ready and safe?
>> Yes. So, uh, of course, number one is is
safety as you just mentioned. Um, and
there are different metrics, different
KPIs, uh, on on how, uh, safety is
measured. Um, one is, you know, to look
at, uh, the crash rates and there of
course the goal is to be safer than
human drivers. You know, we believe that
eventually the technology will support
to be 10x safer, maybe 100x safer. Um,
and uh, you know, the technology
basically never sleeps. Uh, it has eyes
all around the vehicles. It can react in
milliseconds. It doesn't have, you know,
a second reaction time like like human
drivers. It can see better at night uh,
with all the sensors technologies and
uh, redundancies. And uh, then of
course, you know, looking at let's say
cities, customers. I mean it's also very
important that these vehicles you know
are fitting into regular traffic. Uh you
don't need you know special lanes,
special infrastructure. Uh but you know
regular like a regular you know human
driver fitting in there also not being
too slow having a certain assertiveness
and uh then of course trust it's very
important uh for both for riders as well
as for you know cities and operators and
uh companies like Lyft uh who are
offering the services.
I like your answer because it reminds me
that it's not only safety that's
important, but the vehicle also needs to
operate in an assertive enough way that
it inspires confidence with the writer
that it's going to get them from point A
to B. Uh maybe Stephen, can you comment
on this?
>> Yeah. Uh I think it's such an important
and underappreciated aspect of taking
autonomous vehicles from uh the
prototype phase to scaled commercial
deployments is the ride experience and
the ride feel of the autonomous vehicle
itself because the vehicle could from a
technical perspective be uh incredibly
safe. But as JJ mentioned, if it is so
cautious that you end up waiting three
or four different light cycles to take
an unprotected lefthand turn, you're
going to end up with a lot of riders who
are like, "Well, that was that was kind
of cool, but this is not the way that I
I'm going to get choose to get around on
a day-to-day basis." And so the ride
feel and being able to kind of fine-tune
uh the the ride experience to make sure
that it gets you where you need to go in
the right amount of time is going to be
really important because today uh human
driver trips tend to be a little bit
shorter uh from a a trip duration
standpoint uh than AVs. Uh so it'll be
really interesting to see uh AV
companies like Mobilei continuing to
kind of like move the dial on what the
uh ride experience and feel uh of the AV
after you know continuing to master all
of the fundamentals of the autonomous
driving because the the style is is
actually very important.
>> Sure. Yeah. It's fascinating how it
works. So let's move on to another topic
about economic e the economics of scale.
We know that we can already hail a
driverless taxi in cities like San
Francisco, Austin, and Phoenix, but for
most of the country, autonomous vehicles
still feels like maybe five years away.
What are the make or break economic
realities of turning a test program into
a viable business? And why has it been
so hard to make self-driving mainstream
everywhere?
>> I can jump in and take a take a stab at
that. and then I'll hand it over to JJ
who is definitely in the best position
to speak to the the engineering of what
makes it hard to build a self-driving
vehicle. Um you know the the reality you
said Deborah is there are there are
cities around the country where you see
autonomous vehicles. Some of them are
commercially deployed, some of them are
in testing, but that we should all
remember represents the very tip of the
iceberg. And underneath that deployed
asset, there is an entire value chain uh
of different partners and ecosystem of
players that needs to be marching in
lock step in order to support the
commercialization of that asset. And I
think this is really important. So just
want to unpack it for for a moment. uh
uh that value chain starts with
companies like Mobilei which are
building the self-driving technology. It
spans to OEMs, the auto manufacturers
who are building and producing the
vehicles. And today uh we're generally
taking uh retrofitted vehicles. Uh so
they're not they're not built for
autonomous specifically and that means
you need to kind of like tear them apart
and then put them back together with the
tech stack on it and that has
significant implications from a cost and
scale standpoint. And then once you have
the vehicle and you have the AV stack on
it, then you need a fleet manager and an
operator and a financing partner who's
going to hold that vehicle. Uh, and then
you need a mobility marketplace where
you can deploy that asset and
commercialize it. And you need a
front-end customer experience, a way
that riders can interact with your
technology. And uh in order to go from
hundreds to thousands of vehicles,
again, you really need all the different
component parts of that value chain
coming together uh in order to uh
achieve sustainable economics. And uh I
think that's where the the industry has
a a growing appreciation for the
complexity of doing that. Uh and that's
where our partnership with Mobile Eye,
the fact that Lyft owns and operates a
subsidiary called Flex Drive. We 15,000
vehicles that we directly manage today
and obviously a thriving marketplace.
These are all really important
ingredients
>> that is
>> yeah maybe to add to that just just
quickly um um is you know from a
technical side I think you know in order
to scale it's very important to have you
know a product uh that uh basically has
you know high efficiency and also is
built from a cost perspective and u
actually from an overall technical
approach uh in a way that you can
actually go quickly from city to city
because this is something you know where
we look back the last five years uh the
deployment rate you know has been very
slow. We are still not at the beginning
of the actual scaling phase and you know
for us as mobile you know we always say
like safety is first and then
scalability second and efficiency third.
>> Yeah I love that. I mean from what
you're saying it's pretty obvious that
no single company is going to scale
autonomous vehicles alone but
collaboration in this in industry is
notoriously tough. So can you talk about
like a gridlock or or or a bottleneck in
operational uh work between the OEMs and
the mobility platforms and the provider
that you feel is slowing down progress.
>> So um based on my experience now and
I've been in this space now also since
uh you know about 15 years um this is
actually not the case. I mean yes there
is competition between you know
automakers uh and there's competition
between the technology providers and so
on. So looking at these, you know, four
or five like value chain layers uh that
Stephen just explained basically um
there's a lot of collaboration and and
you know it's also there are different
business models you know you look at you
know maybe Whimo an all-in kind of more
vertical type of approach and and look
at our you know approach and
partnerships uh for example with Lyft um
and and with you know Folkswagen as our
main strategic partner for uh uh you
know the first VW ID as you know
platform vehicle, beautiful vehicle also
nicely integrated technology from from
our side and uh also with others. So we
try to be very open in this regards to
actually work with different platform
providers on the vehicle side but also
on the go to market side and we think
this is the better approach this open
approach
>> and I I I'll just chime in and agree
with uh JJ's sentiment that I think
whereas in the very early stages you saw
some players say hey we need to
vertically integrate and own the whole
thing ourselves and both from a
complexity standpoint as well as the
growing recog recognition that the
market opportunity here is enormous and
we are in the bottom of the first
inning. Uh I think more and more players
are realizing that we can go further by
partnering together uh and finding
partners that have very complimentary
skill sets.
>> Great. Talk to me about the future.
What's the next milestone that's going
to really matter in the next two years?
I'll answer this from a from a customer
perspective and I'd love to hear JJ's uh
perspective from a technical one. Um the
metric that we at Lyft are going to be
obsessing over for the next uh two years
and beyond with uh our autonomous
partners is the percentage of riders who
opt in after taking uh an autonomous
trip into the next autonomous trip.
Because again, going back to what I said
in the beginning, for a lot of people,
AVs just aren't on their radar or it's a
bit of a novelty. It's like, yeah, I'm
coming to San Francisco. I want to try
this new experience. It's kind of like
going on a a new theme park ride at
Disneyland. And then there are people
who uh are are getting habituated to
taking AVs, and we want more and more
people in that category. And so you only
have one chance to make a first
impression. And we want to make sure
that we are delighting writers in
setting appropriate expectations and
that they're giving that back to us by
saying, "Yeah, I will take Navy with you
again."
>> And um I I personally, you know, I'm I'm
very interested in and you know, looking
forward to, you know, kind of this dual
path, you know, of bringing this
technology to market on the one hand
with fleets. I mean this is kind of
natural because the technology is you
know pretty expensive at the moment. But
then the second path was consumer
vehicles. uh basically you know letting
people um own or lease such vehicles
maybe they even put them into fleets
when they don't use them themselves but
basically you know having consumer AVs
and then fleet AVs um and and you know
looking at you know which type of
services what type of vehicle you know
interiors exteriors I mean there's going
to be a lot of innovation uh in in the
in the vehicle design you know maybe
there will be even collaboration between
automakers and you know design studios
or furniture companies or you you know,
uh, interior designers to come up with
completely, uh, new new designs,
partnerships. You know, you can have an
office on wheels, you can have a lounge
on wheels, you can have, you know, movie
theater on wheels, anything you want.
And maybe, you know, depending on your
needs and and once uh you order, you
know, this or that type of uh vehicle or
or rider service.
>> I love the office on wheels. Okay, so
this is a short question for both of
you. What is one myth about autonomous
vehicle safety that you want retired?
>> I'll go. Uh I think the myth about
autonomous vehicle safety is that safety
is enough because safety is critical.
It's necessary and it's not efficient.
As we talked about earlier, if riders
feel like the ride is jerky or it's too
cautious, they're not going to be future
customers of it. And so of course we
need safety and we need a delightful
experience around it.
I personally actually think that uh one
of those myth is that you know let's say
as regular pedestrians or u you know
maybe uh um another vehicle you know
driver uh that you might act uh
differently you know in front of an AV
that people think oh it's an AV I can
just you know jump in front of it you
know even at you know two feet or three
feet I mean basically you know people
still need to consider you know the
physical limits of you know breaking
distance and uh uh and and uh you know
reaction and so on. So I think uh you
know from that perspective and I know
there's you know there are discussions
about you know should AVs have this you
know light blue light you know when they
are active so that people can see
pedestrians and so on. Oh this is an AV
or an AV function is on. I have to say
I'm kind of against this and and uh
believe that it's better that people
have respect uh you know for any type of
vehicles uh and any type of sizes uh
because at the end of the day um these
uh uh vehicles can only let's say react
and break a certain uh let's say with a
certain momentum and and brake power and
and and so on you know basically just
physical um u limitations and I think
it's important uh that people you know
treat AVs the same way as you know
human-driven vehicles just, you know,
safer and better.
>> Yeah, it sounds like there's a lot to do
to educate the market about this new
technology. So, thank you JJ and Stephen
for the great discussion. I really
enjoyed it. It's fascinating to see how
technology, trust, and collaboration are
shaping the next chapter of mobility.
Steve, back to you.
>> Thank you, Deborah, JJ, and Stephen.
That was a fascinating look at what
comes next for autonomous fleets and
innovation. Now, we're shifting gears to
zoom out and look at the bigger picture.
How cities, automakers, and regulators
are shaping the infrastructure and the
mindset needed to make autonomy work for
everyday people.
[Music]
I'm thrilled to be joined by two leaders
in this space. James Philin is the vice
president of autonomy and AI at Rivian.
And before joining the automaker, James
spent years at the forefront of
autonomous vehicle development, leading
software and perception teams at both
Whimo and Zuks, where he helped advance
the systems that allow self-driving cars
to see and understand the world around
them. Now, he's building technology that
makes advanced driver assistance and
autonomy integral to Rivian vehicles.
We're also joined by Charlie Tyson, who
plays a key role in Michigan's
autonomous vehicle pilot programs,
advancing the state's efforts to turn
transportation innovation into policy
and infrastructure. He works at the
intersection of government, industry,
and research to make Michigan a national
test bed for next generation mobility.
James and Charlie, thank you both so
much for being here.
>> So, let's get right into it. is he
>> let's talk about where we are the state
of play today where we really are right
now. How would you each describe the
state of autonomous vehicle technology
right now and specifically what's real
what's still experimental and what's
misunderstood. James, let's go to you
first.
>> Yeah, I mean I think you're starting to
see the phase where um you know full
autonomy has gone from the sort of
science project into an actual product.
Um, and you can go up to, you know, San
Francisco, I can drive 40 miles north of
me right now. And, you know, it was just
going to be flooded with Whimos. Um, and
now Zuks is as well. So, I'm I sort of
um felt a few years ago that actually
the the fundamental problems had been
solved. That wasn't that doesn't mean
that every every problem has been
solved, but it was moving into a more
engineering um and deployment and
scalability phase.
Um, and then I think, you know, with
that hat on, you got to think how does
this change how people um use
transportation in the future. And I'm
still a big believer in um personally
owned vehicles. I think that for every
mile done in a robo taxi, probably in
the future, you know, more than 10 times
will be done in personally owned
vehicles. I think the economics of a
robo taxi versus a personally owned
vehicle still mean that those two modes
will be around for a very long time. And
so that's why I made the the leap to
Rivian. Essentially, can we bring that
sort of L4 technology back to the
consumer space and really provide value
for our customers.
>> And Charlie, what do you think?
>> Yeah, I think James hit the nail on the
head there, but um from a state's
perspective, I think that we are we are
going from like the testing phase to
some level of commercial operations, but
I think we're seeing it in certain
states. Um, for example, um, James
mentioned California. We're seeing, um,
I actually was just in Arizona. I took
Whimos in Phoenix. Um, commercial
operations really impressive. Um, but I
think there are still some challenges.
Um, the technology seems to be there for
the most part. But um consumer consumer
adoption, how does it how do these um
techn how do these vehicles whether
they're consumer AVs or um consumer
vehicles with um some level of of uh
autonomy or their um autonomous vehicle
fleets? How do we integrate them into
our existing transportation network? I
think that's still a challenge that
we're working on um here in Michigan and
and really throughout the throughout the
nation.
>> You know, it's really interesting. I
mean, I guess the one question I have
for both of you is how do you get people
to really feel comfortable about getting
in an autonomous vehicle? I think about
my parents who say that they'll never
get in one, that they don't just don't
trust the technology. And it seems like
there's maybe a generational divide.
Boomers feel one way, millennials,
genzers. I'm curious how you all think
about this trust issue, particularly
from a generational perspective. And
James, do you want to start first?
Yeah, I mean my sense is it's pretty
non-existent. I mean my my parents are
also in that generation and um you know
they're interested in what I'm doing but
they don't really you know trust all
these systems but you know the last time
they visited we took a Whimo around um
it's very you know very quickly becomes
normal actually and um I feel like
there's that initial kind of wow moment
hesitation but then it's completely
normalized people you know just go about
their day. So, I think it's um I don't
think the trust issue is is going to be
um a persistent one. I also think that
as people get more used to those higher
levels of autonomy, they'll expect more
autonomy in their vehicles. So, I think
that trust issue actually trickles down
and what you'll see is a huge um sort of
competitive swing towards um you
personally owned vehicles that offer
those higher levels of autonomy. Um, so
I think that's kind of a bit of the race
you see right right now and one I think
Rivian is, you know, perfectly poised to
execute on.
>> Do you agree, Charlie?
>> Yeah. And I think Yeah, I agree with
that totally. And I I think that, um, we
just have to give individuals the the
chance to experience the technology. Um
that's why I think that although we want
to see and we're pushing for commercial
operations, um these pilot projects and
these these testing activity that allows
consumers to experience the technology
in different use cases is still really
critical. Um we've seen a number of
pilot projects uh in Michigan um in some
of our larger cities like Detroit of
course, Grand Rapids, Ann Arbor. And the
feedback from the surveys that we we
sent out um was really interesting. Um
uh most most of the the riders I think
90% of the riders in the survey
respondents um you came back saying that
they would absolutely love to take
another um trip in an autonomous vehicle
and they would recommend it to their
their peers. Um they what was
interesting though was um we asked a
question in the survey um around
removing the safety driver and I think
that is a big piece here. That's where
we started to see some of the um the
comfort levels um change a little bit.
Uh the survey respondents um I think it
was about 5050, you know, responding
that they would be willing to get get in
an autonomous vehicle without a safety
driver. So I think just um getting
people more um accustomed with the
technology um sharing the benefits um
and also I think uh it's it's critical
to um be able to you know educate the
public um and just give them that
experience to to get inside the vehicle
um provide feedback uh and and let them
know that these aren't being forced on
them. They're they're being deployed in
safe ways and and kind of a a crawl,
walk, run approach. I think that's
critical.
>> Charlie, you mentioned these these
pilots. Um, I guess my question is how
how do you measure readiness? Is it, you
know, the the number of miles driven? Is
it disengagement rates? Is it just
consumer trust or something else
entirely? Like h how do we know that
these things truly are ready to be out
on the road?
>> Yeah, I think um it's different for it's
kind of case by case, different for each
project. Um, for example, we had a
project in Northern Michigan looking at
um supporting a a a full-size autonomous
transit bus deployed at Sleeping Bear
Dunes National Park. So, looking at a a
very unique use case in northern
Michigan. Um, a majority of the of the
riders were tourists visiting the area.
they either uh they had challenges with
with parking, but um I think uh it the
way that we kind of measured read
readiness for that project was how many
times um the the autonomous vehicle um
had to be essentially how many times the
um safety driver had to take control of
the vehicle. Um were there challenges
with connectivity in that region due to
to Wi-Fi or um you know infrastructure
uh capabilities? Uh so I think that's
the big thing is looking at
disengagement um and just looking at
based on the use case based on the
region what are the key factors to
enable um adoption and and that kind of
is how we determine what are the key uh
metrics around uh success.
>> I'm curious
>> yeah maybe I can just add to that please
um yeah and just just talk about the
Rivian process. So we we actually do an
extensive kind of release readiness um
process every month for our software and
that involves you know many aspects to
it the sort of metrics across the stack
but also a key part of that is actually
simulation. So we take millions of miles
of real customer data and we can replay
them through our stack and kind of
measure the performance the safety the
smoothness um and all those aspects and
I think all of those pieces go into that
readiness uh report. So I think that's a
very important part is um having that
scale of data and also the ability to to
replay it and to to you know gain
insights from it.
>> Charlie, you mentioned in your survey
data that you cited that there's a
distinction between how people feel uh
when they're in a self-driving car that
does not have a driver versus one that
does have a driver. And how do you sort
of bridge the gap between people's
perceptions on on those two experiences?
Yeah, it's a great question. Um, I think
it's understandable that someone may
have um less willingness to get an
autonomous vehicle without a safety
driver. But I think that goes to the
importance of getting autonomous fleets
out there. For example, some of these
states that have been doing it without a
safety driver. Again, was just in a
Whimo in Arizona. Um, and got out of the
vehicle feeling totally safe. Um and I
and so that kind of goes to um the
importance of um states and and um
industry working together, government
and industry working together to safely
deploy these vehicles and provide them
and um you know provide the opportunity
for the public to get in get in the
vehicle. Um ideally we move from safety
drivers to to non-safety drivers and
fully autonomous vehicles. Um and that's
what we're working on doing here in
Michigan. But there are some challenges,
right? And I think that's um why
Michigan feels that we are um posi
positioned well to be a kind of a great
test bed to to move from from the
testing phase to commercial operations.
For example, how do we um ensure that
autonomous vehicles can operate in in
harsh weather conditions? In order to be
able to fully um to see widescale
adoption of AVs, we need them to be able
to operate in in harsh weather
conditions, rain, snow, etc. And so, um,
there's certain states and and Michigan
definitely feels we are one of them
that, um, you not only due to our
automotive heritage, but also, um, just
our our demographics, our our weather
conditions, um, we are positioned well
to be able to to test these different,
um, uses and to hopefully support, um,
that path towards commercialization
while we do so with, um, public
engagement and and providing, you know,
individuals the experience to, uh, to
get in the vehicles. I always kind of go
back to, you know, 10, 20 years ago, how
many parents um were on social media?
Not very many. I was always I was always
getting flack from my parents for being
on Facebook or or um you know, whatever
in Instagram, for example. But, you
know, now my grandma's on Facebook and
she's she has a smartphone and and
they're on it more than I am. So I think
it just takes time for adoption and for
for uh the public to get um accustomed
to technology and feel comfortable in it
and and I think industry is doing a
great job and it's important for
government to to align and to partner
with industry to to make this path as
smooth as possible.
>> You know it's a great point you raised
and just this week we obviously had the
big Amazon AWS outage that knocked out
wide parts of the internet and how
people go about their everyday lives. To
me, this is one of the concerns perhaps
for for autonomous vehicles is what
happens if there's an outage and then
these things don't operate the way that
they're supposed to operate. Um, what do
you do? How do you adapt? Um, so how do
you how do you both think about that?
James, do you want to take that one
first?
>> Yeah. So, I think I think it's key to to
in your safety case sort of think
through these um these outages that
could happen in the cloud. So the the
the software that's on our Rivian
vehicles um it it doesn't require a
cloud connection. So the idea is that we
can be um sort of in a safe state even
if that external connection goes down.
So there's all the processing happens on
vehicle. I think um it's pretty
important when you build a resilient
system that you're not taking those
dependencies unnecessarily or if you are
that you have um sort of good backup
processes. Um
>> and Charlie, what do you think? And I
would I would say I don't have a highly
technical answer there. Um obviously
that's it's a it's a challenge. We also
worry about cyber security um the you
know grid resiliency. But um one of the
things that we want we try to do here in
Michigan and um within the office of
future mobility electrification is is we
reach out we we like to engage industry
and to call out some of these challenges
and say how can you help us solve some
of these challenges that we're talking
about here? for example, um you know,
the Amazon um outage, you know, we on an
ongoing basis, we like to um call out or
identify some of these ch these major
challenges and concerns and
considerations and and to work with ind
industry to solve them. And that's why
our our grant programs and our um pilot
projects have been so successful.
Identifying challenges and finding uh
solution providers to address them and
working together in public private
partnerships is something that we are
we've seen a lot of success in and
continue to do. You know, Charlie, you
mentioned you use the social media
comparison 20 years ago, how people were
just starting to figure out how to use
social media. The older generation
wasn't on it and now all of a sudden
they're all on it. Um what do you think
is the biggest variable now to achieving
a fully autonomous world?
Yeah, I think um
I would probably say one economics. How
do we make it economically feasible for
the fleet operators or the auto um
automakers um to fully integrate
autonomous systems into their vehicles?
Um I think it's going to be critical to
continue supporting and seeing um
increased level of autonomy and consumer
u vehicles and and pro production
vehicles. I think that will really help
um you know uh drivers and individuals
um feel comfortable with the technology
getting it into their day-to-day
vehicle. I think that's critical and
also um being able to support um you
know commercial fleets throughout um our
nation and doing so uh and continuing to
do so in a safe way. But um I I think
addressing some of the challenges with
infrastructure and addressing some of
the challenges with um um deploying
harsh weather conditions is something
that is really important. Um but again,
I'd probably, you know, go back to how
do we get this um how do we get
increased level of of autonomy, excuse
me, into consumer vehicles and into more
production vehicles. I think that's
going to be um a critical way of of
increasing adoption.
James, what do you think?
>> Um, yeah, I think I think, you know,
Charlie's right that um a certain amount
of this is just just time that people,
you know, it takes time to try and then
gain acceptance. I think it's actually
changing, you know, pretty quickly. So,
we we released our, you know, hands-free
feature earlier this year and since
then, you know, every month we've
essentially seen the number of miles
driven hands-free increase. So I think
it's around 20% of all Rivian models now
are done in that hands-free mode. So I
think as these features get better, they
become more capable, people become more
comfortable with them, they get used to
the, you know, the UI, the UX aspects.
Um, and then also, you know, the system
is learning through the data that we're
able to gather um during those those
drive events and all of that kind of
ladders into this um sort of upward
spiral of improvement. So, I think it's
important that um OEMs are able to
really learn from their fleets. I think
that's something very new that they
haven't had to do in the past. Um and we
think it's a it's a key competitive
advantage for us.
>> It's so interesting as as adoption rates
increase, I have to think too that
driving fatality fatalities will
ultimately go down. I mean, we have
40,000 driving fatalities a year right
now with cars that people drive, right?
Traditional cars. Um but so but at the
same time though once if god forbid a
whimo happens to kill someone who's
crossing the street you're gonna get a
there's gonna be a ton of attention on
that and people are going to say
computers kill people and how do you how
do you go about this or or what do we do
to combat this? So how do you all think
about that and and the the safety
element of all of this?
>> Yeah, maybe I can start. So um I think
we we have to recognize that the the
baseline here is the average human
driver and you know there's far too many
fatalities in the US. I think the last
stat I heard is something like a 747
full of people die every day in the US
you know on the roads in the US. So I
think you know for me that's the number
we have to drive down. We're not saying
that these um systems are going to be
perfect. I think that's an unrealistic
expectation. In fact, I think if we have
that expectation, we'll delay launching
something that could be saving lives um
potentially now. Um and so I think
that's the number we should focus on and
drive down. Um we spend a lot of work,
you know, Rivian on our active safety
systems. Um so a lot of that is actually
fed by the same world model investments
um on the ML side uh that are powering
sort of the L2 plus features. And so
really our goal is for you know Rivian
vehicles to be the safest vehicles to be
in and around. um through those active
safety features. So you can imagine
almost like a safety bubble where we're
preventing the vehicle from getting into
these collisions and I think systems
like that deployed widely on consumer
vehicles will really start to move the
needle um on these fatality rates.
Yeah, I think it's a really important um
question and um I think one of the
important things here is to um make sure
that we are being with the government
being in government um working closely
with industry to ensure that safety
standards are are top-notch and are um
are we're aligning on um you know making
sure that bad actors aren't able to
access autonomous vehicles or you know,
in cyber security um capabilities are
are in place. Um and then making sure
that we educate the community um that we
that the vehicles are in. Um I think
most most communities want to um know
that the roads have be roads have gotten
less and less safe. Um people, you know,
technology has the ability to actually
um make our make our road safer and the
average person is not going to drive as
safe as autonomous vehicle. Um they are
um you know just to be able to trust the
technology and and that's going to take
some time and um just constant
engagement with with communities that
the the vehicles are in. I think that's
really important. When you talk about
trusting the technology, it gets me to
this broader question in the industry
about what's the better approach to
autonomy. Is it, you know, it's the
cameras versus the LAR question. Is it
using machine learning or a rules-based
system? James, what do you think?
>> Yeah, I mean, I feel like those
arguments have sort of largely been been
settled. Um, so I think you really want
to use and leverage machine learning for
as much of the driving task as possible.
The reason you you should do that is
because um specifying driving in rules
is actually it's very complicated. You
end up with huge you know spaghetti code
of heristics. Um it is actually not well
specified in many cases. So you know
think of the example of a lot of
vehicles entering a a stop intersection.
You know there are rules on how that's
supposed to be handled but that's not
how humans actually navigate those
things. And so to sort of encode all of
that in in a rulesbased system is is
almost impossible. So we believe in um
doing as much in the machine learning
model as possible really learning from
um customer driving data. There's an
effort we have ongoing at the moment
called the Rivian large driving model
and this is supposed to be an offboard
um huge model that can um essentially
learn all the nuances of human driving
from all the data we receive and then
but then I think you have to you have to
sort of couch that in a system that
provides those those guardrails, right?
So we we can use the ML um as much as
possible especially to to have this sort
of humanistic um and sort of nuanced
understanding of how to drive but we can
still have guardrails on that system
that say okay I don't want to collide
with anything I don't want to run a red
light um I need to be cautious in this
situation and so I think it's it's that
combination using ML as much as possible
because that is the most scalable and
sort of powerful approach but then
having still a rules-based um uh set of
uh features at the end that you can use
to kind of guarantee certain aspects
about the driving behavior. And then I
think on the when you come to sort of
cameras versus versus lighter versus
other modalities, I think for us we
would say um you know more modalities is
better and you know Charlie was talking
about um uh you know adverse weather you
see in Michigan and I think it you know
it's exactly those cases where those
additional modalities really really can
help. Um so you know for us it's it's
really you know can you get the right
sensing sort of independence and um
different views of the scene at an
economically you know affordable price
point for consumers and that's what
we're really focused on.
Yeah, and I I would say we are here in
Michigan trying to to support industry
and and getting to to where we don't
industry and and fleets um autonomous
vehicles aren't relying on um
infrastructure,
but at the same time, how do we improve
infrastructure to make um to in increase
redundant redundancy, improve safety? Um
and so that's kind of what the approach
we're taking. How do we support machine
learning and autonomous vehicles that
aren't completely reliable on VTOX or
you know vehicle to infrastructure
communication but being able to have
extra redundancy um with roadside
roadside units and and infrastructure
technology that is able to provide kind
of a backbone and um additional layers
of safety. Uh we for example we have a
um a corridor an autonomous vehicle
corridor being built built out uh
between Detroit and Ann Arbor about a 30
m 39 mile segment of of interstate that
will have um vehicle to vehicle
communication technology um technology
ve um vehicle to infrastructure
capabilities um that will support the
integration of autonomous fleets into
our our normal traffic. Um, so I think
that's the approach that will will be
most su successful down the line and the
approach that we're taking here in
Michigan.
>> You know, you've both touched on the
weather issue and Charlie specifically,
obviously you're sitting in Michigan
where you guys have some pretty severe
weather there. Um, what I mean, how do
you is this the biggest variable to
achieving a fully autonomous world? Is
this how do you combat severe weather
conditions?
I would love to hear um James' thoughts
from a technical perspective, but I I
would say um you know, Michigan, yes, we
have harsh winters, but our summers are
absolutely beautiful. Please come visit.
Um but I would say that um you know how
do we support uh new technology that
might be able to help um ensure that the
sensors are clean um if there's if in in
a rainstorm for example or how do we
make sure how do we support industry in
new technology that provides better
cameras that cameras that can work in an
adver adverse condition. So that's kind
of what we're doing here is is providing
the the platform throughout the state to
be able to test new technology and
identify, you know, startups and
technology providers that are working on
some cutting edge technology that will
help AVs work in all types of
conditions.
>> Yeah, I think um you know Charlie
touched on some of the the things you
know I think you'd start with the
sensors, right? So yeah, this sensors
see clean and unluded. Can they can they
see the scene? you know, in some some
foggy conditions, some, you know, very
heavy snow, um you actually can't really
rely on cameras. You have to then um you
know, start using radars. Um
you know, even non-weather related
scenarios like, you know, the sun, a low
sun, you know, shining directly into the
cameras that can be challenging um from
a vision only. Again, that's where, you
know, radar and LAR can really help. So,
I think you sort of start start with
that. You need to have that um that kind
of patchwork of sensing capability and
sort of different different sensors as
well.
Um I think then there's then the data
piece is very key. So um how do people
actually drive in snowstorms? It's not
the same way um that you drive in an
unluded, you know, sunny day, right? So
do you have the machine learning and do
you have the data flywheel that's that's
telling you how to handle those
situations? I think that's a that's a
big advantage that um a kind of fully
integrated um sort of databased OEM like
Rivian has over for example like a rower
taxi where you have to actually send
those fleets to go and gather this
specific data in these different
conditions. Um and then finally um yeah
sort of how how do the rules you know I
talked about those guardrails. Do those
guardrails need to change? you know, for
example, um you know, when you when
you've got snow on the ground, people
often don't follow the lanes. They can't
see them. So, you you sort of have these
virtual lanes that pop up. Does that
need to be taken into account in your in
your guardrails? Um so, it's sort of
it's multifaceted. I wouldn't say it's
the primary challenge. I still think um
you know, density um and complexity in
urban areas is is typically where the
those final um big challenges are. um
you know James to just reason about you
know many many objects in a scene maybe
there's a lot of nuance negotiation and
things happening those can be tricky to
handle um in a good way
>> on the density issue James you know I I
took my first Whimo earlier this year
out in San Francisco and like so many
other people was just so so fascinated
by it um and so and it seems like
wherever you turn in San Francisco
there's another Whimo on every street
corner but in terms of New York City
it's I I just walk around here and I
think to myself how are we going to have
these types of cars in New York City and
they're Whimo is already testing um
testing their fleet in in the city here.
But I mean, how do you how do you adapt
to these different densities in
different cities and how all the
different layouts and how everything is
is so different in different places?
>> Um yeah, so I think you know you hope
that your that some of your base systems
obviously generalize to those places.
Now, of course, there's going to be, you
know, traffic specific um kind of rules
of the road almost that exist in that
exist in New York but don't exist in San
Francisco and things like that. And I
think that's where the where that data
flywheel really is important. And I
mean, you talked about New York City,
but you know, if you go outside of the
US and you talk about, you know, a
country like India where um you know,
some of the driving's you know, even
more, you know, intense I would say um
and very different again. So you have
sort of had to think like how does a
system um scale and I think that it
really sort of tips you in favor of the
MLbased approaches right because there
you can gather the data you can see how
people drive you can uh you know learn
start to learn how um to sort of mimic
it versus you know rules based systems
that can really they can be brittle so
you take them to a new place and
suddenly those rules um don't work
anymore.
James, you're a former way Whimo
employee before you joined uh Rivian.
So, I have to ask about the elephant
room. I got to ask about Elon Musk and
Tesla's approach uh to autonomous
driving right now.
How does Tesla compare to Whimo compared
to Rivian and others? And who's who's
getting it right and perhaps who's not
not doing it as well?
>> Um yes, I'm obviously not privy to, you
know, what's going on inside Tesla. Um I
think what I would say from the outside
is that um I think they've
you know on the good side they've really
sort of pushed um the OEMs forwards in
the sense that they took a very um sort
of MLbased approach early on and um I
think that is the right way to build
these systems. Now on the on the counter
side, I think they have um a sort of
very uh rigid point of view I guess on
um different sense modalities which I
don't think is you know fully
explainable just from an engineering
point of view. So um I would say it's
sort of a a mixed bag. Um I think we're
really focused on um you know can we can
we bring that L4 technology back to
consumers in the best way possible. And
I think um you know sensors are sensors
can get you there faster and they can
get you there in a more robust way. And
I think the the price point of a lot of
these sensors is is no longer that you
know liars are $10,000 um because of the
huge um scale that you've seen in China.
Those are those are coming down to you
know a few hundred which is very much in
the in the envelope of um you know
consumer vehicles. So, I think, you
know, that's that's kind of the approach
we're taking and the one that I think is
is going to get us there um in the most
sort of direct way.
>> Charlie, what do you think?
>> Yeah, you'll have to um recap that
question one more time for me, Steve. My
apologies.
>> Just comparing the different approaches
that uh automakers are taking from Whimo
to Tesla. Um, Whimo I has this
reputation in the industry for taking a
slower perhaps more more cautious
approach to how they go about things
whereas Tesla is more of a move fast and
break things kind of uh approach here
and obviously they have the robo taxi
fleet that's uh starting to be rolled
out. So just curious your thoughts on
the different approaches and the pros
and cons of both.
>> Yeah, you know I think
we don't really have like a
reference per se on on what a company's
approach would be. Being with the state
of Michigan's economic development
department, we we want to support all
companies that want to grow here in
Michigan, hire here in Michigan. I think
the big thing is um deploying in a safe
way. That's that's the most important
thing. whether you're taking a a fast
approach with um using machine learning
and and or you're taking a slower
approach with a certain use case and and
leveraging infrastructure. Um we don't
really have a preference per se as I as
I mentioned. It's just um we are trying
to to make Michigan a great place for
companies to thrive and to to uh to
deploy their technology in a safe way
within communities. And so um you know
we are we think that we have a really
strong obviously automotive industry
here but how do we make sure that we we
remain competitive and how do we kind of
bridge the gap between legacy automotive
and and the startup world in in the on
the west coast and and bring the minds
together to ensure that we're you know
we see this widespread widespread
adoption over time.
James, what what do you think is the
biggest bottleneck for the robo taxi
industry right now?
>> Um I I mean I'm no longer in, you know,
the robo taxi industry. So um I I would
say that um it just takes time to scale
these things. They're, you know, fleets
are physical things. You have to charge
them somewhere. You have to clean them
somewhere. you have to you know think
about power and um the operational
aspects and so that just takes time. Um
so I think and and then of course
there's you know there could also be you
know further engineering development
that has to happen to you know maybe
cover specific cases that you see in new
places. So I don't see a fundamental
issue but I think you know this isn't
like um I don't know Charlie mentioned
social media right it's not like just
downloading an app someone has to go and
build these vehicles put all the sensors
on they have to be kept somewhere they
have to be you know kept clean
operationalized and that takes time it's
like a the physical investment aspects
um are you know just take take time
>> Charlie what do you think
>> yeah you know I think um
identif Identifying a clear use case is
important for for fleet operations. Um
deploying them in in a certain scenario.
Um at least right now I think that's an
important way to continue momentum and
continue adoption. Um you know deploying
them in a highly urban complex
environment may not be the best best um
option right now. So can we identify
fixed routes or can we identify certain
use cases that um fleets will thrive in
and and um it's more feasible you know
at this time and for example looking at
how can we partner with large employers
to provide um you know shuttle services
between their facilities or can we
deploy autonomous vehicles at airports
from the parking to the terminals. um
looking at deploying autonomous vehicles
on at universities to get um students um
engaged in to experience that
technology. I think that's the that's a
really um critical way at this juncture
to continue, you know, supporting fleet
operations uh before we see them in in
highly complex multimodal environments.
>> I'd like to shift gears a little bit. we
could do a little bit of a a lightning
round here on some of your predictions
for what you guys think is going to come
true in this industry in the coming
years. So, uh I have a my older son is
seven years old. In 10 years, he'll be
eligible to get his driver's license
here in New York. Will he need it and
will he even want one? James, what do
you think?
>> Um so, I I grew up in London and even
without autonomous vehicles, um you
know, London has a fantastic, you know,
public transport system. So when I was
young and sticking in the city um you
know I didn't need to drive and actually
I took my driver's license quite late. I
would say once you have a family and you
start um needing to get from A to B
that's when the real value of you know
personally owned vehicle comes. So I I
suspect your son will actually still um
get a driver's license even if his early
years are you know in in a rubber taxi.
But I think the world in which the
vehicle he'll be driving in I think will
be much safer. It will have um you know
many more autonomy modes. Um you as a
parent, this is actually you know
feature I'm excited about you know may
be able to engage a teenager mode right
which puts the vehicle in a in kind of
like a extra safe state. Um so that uh
you know your son can't can't speed or
you know do donuts in the parking lot or
whatever else he wants to do. Um, and
so, uh, yeah, I think that's probably
most likely. I I don't see a huge shift
away, um, from the automobile, at least
in the US, just because of, um, you
would also need a a commensurate shift
in, you know, where houses are built and
how people are living and everything
else. Um, but I do think that, uh,
consumer autonomy will will become, uh,
essentially a, you know, a must have on
every vehicle.
>> I would say must have on every vehicle.
>> Your son will.
>> Yeah, I would say your son will probably
um need to get a driver's license. And I
think that's okay. I think we we're
going to continue to see more um AVs on
the roads, but it's going to be a little
bit longer down the line until you you
don't need a um a driver's license. I
actually p personally enjoy driving from
time to time. Um there's also times
where I I I would love to get an
autonomous vehicle and not drive to get
work done or to to get some rest, for
example. Um, but uh I think that's a
little ways out and it really depends on
where you live your lifestyle as as
James alluded to. If you're if you live
in a rural community, most likely you're
going to, you know, at least near-term
over the next 5 10 years still want need
a driver's license, want a driver's
license. If you live in a highly urban
area, maybe that's not the case.
What percentage of cars on the road will
be autonomous versus not in let's say 10
years from now?
>> 20%.
>> What are we at right now?
>> And what are we at right now?
>> That's a good question. James, do you
know?
>> I would say sort of define autonomous.
So, do you mean like a full robo taxi
level autonomy or do you mean vehicles
um that for example could provide an L3
capability that gives you your time back
on the road and makes makes driving
safer? So, I think I think that's where
there's actually a big spectrum of
autonomy here and
>> I I think we're I think people focus on
the endpoint. Um but I think there's
actually many other customer societal
benefits you get on the way there. Um,
so I think, you know, to be honest, I
think every vehicle, new vehicle sold in
10 years time to be competitive will
have to have, you know, close to
best-in-class autonomy. I think it's
becoming more and more of a consumer
preference. We we see actually much
higher conversion rates um when people
try our autonomy features in the in the
stores. I think it's something like a 3x
conversion rate. And so I think um I
think that that shift is really
happening. And I think as you see, you
know, robo taxis roll out, people will
just expect and demand more and more. So
I think I think this this tide is coming
and um OEMs need to be ready.
>> What what matters more better data or
the algorithm?
>> I think if you had to
I think if you had to pick one, you
would pick the data.
Um but to make the best use of the data
you need, you know, excellent, you know,
machine learning engineers and
approaches to to understand that data
and to learn um you know, how to drive
from it essentially.
But the data is the most important. If
you don't have the data, I think you you
there's no getting around it really. So,
you know, if you have if you don't have
the Michigan snowstorm, I don't think
there's any feasible way you could build
an autonomy system that then can handle
those Michigan snowtorrms. You have to
go there and see the data.
>> And actually, just sticking on that,
James, um, Rivian is doing a lot in the
Gen AI space. And so, do you want to
talk just a little bit about um, how are
you using Gen AI to, um, improve the
autonomy that you guys are offering in
your vehicles?
>> Yeah. Yeah. So I think here's maybe two
sens in which you know we we use genai.
So I think um one is in this like large
driving model that I alluded to earlier.
So that is actually a very large
transformer-based model. It looks a lot
like um a large language model in the
sense that you have you know data coming
in in the LLM space. It's it's text in
our in our side. It's really sensor data
raw sensor data. And then you have these
large transformers that that kind of
chew on that data. And then at the end
um we we generate tokens. And now these
tokens are not words or or um you know
letters in the LLM case. They're
actually little snippets of trajectories
and we kind of we piece them together
and that gives you the the best um the
model's best interpretation of the
future driving path. Um so I think you
know that that model is very heavily
inspired by a lot of the work that's
happening on LMS and of course we we
kind of uh slipstream on the work that's
happening there. I think that's one
sense. The other sense is um we're also
you know uh seeing a big sort of um
productivity boost from using some of
these genai tools. Um uh I would say not
replacing people but really making them
more effective. So um you know improving
the velocity that you can write code
that you can test code that you can do
things like code review and um uh and
you know interface with APIs and things
like that. So I I think um yeah it it is
an accelerant and and definitely
um on both sides. I think very essential
to the work we're doing on autonomy.
>> Charlie, what's the biggest myth that
you'd like to debunk about self-driving?
>> That the techn is not ready. I think the
technology industry has been um
incredible. At least
I think in the United States, you know,
the technology being developed by
industry, by academia is um has gotten
us to a point where it's ready. Let's
let's get these techn let's get these
vehicles out there and get let's allow
people to experience them. Um but I
that's the big thing. I think too many
people think that the technology is not
ready, it's not safe, but um I think
that's just the opposite.
>> James, what do you think?
>> Yeah, I you know, plus plus one to
Charlie. Um I think um
uh yeah I I think we we we just need to
get more people experiencing um what is
out there and I think we
you know we need to push um US OEMs to
to be more tech forward um you know you
see how the level of autonomy features
that are present in Chinese um OEMs and
I think uh you know I kind of feel like
OEM's got a little bit complacent um in
in that regard and I think we have to
push everyone forwards Um, so that's
what we're, you know, we're trying to do
here at Rivian.
>> And finally for you, James, what's one
word to describe the state of
self-driving technology today?
>> I think on the cusp, I know it's not one
word, but it's, you know, yeah, one
idea. So, I think it's really on the
cusp where you um you're seeing
significant deployments um in certain
cities and I think um in in dense
metros, I think that will happen quite
quickly. And then I think you'll see um
a trickle down into consumer vehicles
for most other trips
>> on the cusp. I like that. Charlie, what
do you think? One word to describe the
state of self-driving technology today
>> here. I think it's here and we're going
to see it more and more every day.
>> Well, thank you James and Charlie for
these insights. Uh and to everyone who
joined us today for inside self-driving.
I also want to thank our partner MobileI
for their support of today's program.
I'm Steve Russell from Business Insider.
Thank you for being with us. We'll see
you next time.
[Music]
[Music]