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

[00:11] [Music]

[00:19] Hello everyone and welcome to Business

[00:21] Insiders inside self-driving the

[00:24] AIdriven evolution of autonomous

[00:27] vehicles presented by Mobile Eye. I'm

[00:29] Steve Russell, chief news editor here at

[00:32] BI. And today we're diving into one of

[00:34] the most transformative and debated

[00:37] frontiers in technology. How AI is

[00:40] turning autonomous mobility from a long

[00:42] promised dream into a fast approaching

[00:45] reality. We'll explore how automakers,

[00:48] tech innovators, and policy makers are

[00:50] working together to make autonomy safe,

[00:53] scalable, and trusted, and what that

[00:56] means for businesses, cities, and all of

[00:58] us who share the road together. First,

[01:01] we're starting today with a conversation

[01:03] presented by our sponsor, Mobile Eye,

[01:05] that goes inside the company's

[01:07] collaboration with Lyft as they work to

[01:09] bring driverless technology to scale.

[01:17] [Music]

[01:22] Thank you, Steve. I'm Dr. Deborah

[01:25] Bervishes and I'm happy to be here. Robo

[01:28] taxis already operate in a few cities,

[01:31] but taking autonomous vehicles

[01:33] mainstream comes down to three things:

[01:36] auditable safety, consumer trust, and

[01:39] economics that work. I'm joined here by

[01:42] JJ Youngworth, executive vice president

[01:45] of autonomous vehicles at Mobile Eye,

[01:48] and Stephen Hayes, VP of autonomous

[01:51] fleets and driver operations at Lyft.

[01:53] Thank you both for being here. Let's

[01:56] talk about trust and safety. It seems

[01:59] like one of the main goals of autonomous

[02:01] vehicle companies is to convince both

[02:03] the writers and the regulators that

[02:06] these kinds of vehicles are safe. So,

[02:09] Stephen, at Lyft, your role is to

[02:12] interact directly with a rider. What

[02:15] would they need to see or experience to

[02:17] feel comfortable using an autonomous

[02:19] vehicle?

[02:21] >> Great question. Uh, at Lyft, our purpose

[02:23] is to serve and connect, and that's

[02:26] something that we do uh about 800

[02:28] million times a year, helping riders get

[02:30] to where they need to go. Uh, and over

[02:32] time, AVs are going to make up a bigger

[02:34] and bigger percentage of those trips on

[02:37] our platform. And uh if you are tuned

[02:40] into this broadcast, chances are you are

[02:42] a bit of a tech enthusiast and an early

[02:45] adopter. But the reality is for most of

[02:47] the people who open up the Lyft app on a

[02:49] daily basis, they're just looking to get

[02:50] to where they need to go. Uh and that's

[02:52] where uh it is our privilege and

[02:55] responsibility to introduce millions of

[02:57] new riders to exciting autonomous uh

[03:01] technology. And in order to do that

[03:03] effectively, we need to find the right

[03:05] autonomous trip for the right rider at

[03:07] the right time. And then once they come

[03:09] in, uh we need to educate them about the

[03:12] experience that they're going to have.

[03:13] Uh and make sure it's a delightful one.

[03:15] And all the while, what is going to be

[03:18] really important for us is making sure

[03:20] that we have happy uh repeat customers

[03:22] who are getting where they need to go

[03:24] even more quickly and efficiently than

[03:26] they are today.

[03:27] >> Awesome. JJ, from your perspective at

[03:30] Mobile Eye, which three pieces of proof

[03:33] would you hand to a regulator to prove

[03:35] that autonomous vehicle technology is

[03:38] ready and safe?

[03:41] >> Yes. So, uh, of course, number one is is

[03:43] safety as you just mentioned. Um, and

[03:45] there are different metrics, different

[03:46] KPIs, uh, on on how, uh, safety is

[03:50] measured. Um, one is, you know, to look

[03:52] at, uh, the crash rates and there of

[03:54] course the goal is to be safer than

[03:56] human drivers. You know, we believe that

[03:58] eventually the technology will support

[04:00] to be 10x safer, maybe 100x safer. Um,

[04:04] and uh, you know, the technology

[04:05] basically never sleeps. Uh, it has eyes

[04:08] all around the vehicles. It can react in

[04:10] milliseconds. It doesn't have, you know,

[04:12] a second reaction time like like human

[04:14] drivers. It can see better at night uh,

[04:16] with all the sensors technologies and

[04:18] uh, redundancies. And uh, then of

[04:21] course, you know, looking at let's say

[04:23] cities, customers. I mean it's also very

[04:25] important that these vehicles you know

[04:27] are fitting into regular traffic. Uh you

[04:30] don't need you know special lanes,

[04:31] special infrastructure. Uh but you know

[04:34] regular like a regular you know human

[04:36] driver fitting in there also not being

[04:37] too slow having a certain assertiveness

[04:40] and uh then of course trust it's very

[04:42] important uh for both for riders as well

[04:45] as for you know cities and operators and

[04:48] uh companies like Lyft uh who are

[04:50] offering the services.

[04:52] I like your answer because it reminds me

[04:55] that it's not only safety that's

[04:57] important, but the vehicle also needs to

[04:59] operate in an assertive enough way that

[05:01] it inspires confidence with the writer

[05:03] that it's going to get them from point A

[05:05] to B. Uh maybe Stephen, can you comment

[05:08] on this?

[05:10] >> Yeah. Uh I think it's such an important

[05:13] and underappreciated aspect of taking

[05:17] autonomous vehicles from uh the

[05:20] prototype phase to scaled commercial

[05:22] deployments is the ride experience and

[05:24] the ride feel of the autonomous vehicle

[05:27] itself because the vehicle could from a

[05:29] technical perspective be uh incredibly

[05:32] safe. But as JJ mentioned, if it is so

[05:36] cautious that you end up waiting three

[05:38] or four different light cycles to take

[05:40] an unprotected lefthand turn, you're

[05:43] going to end up with a lot of riders who

[05:45] are like, "Well, that was that was kind

[05:46] of cool, but this is not the way that I

[05:48] I'm going to get choose to get around on

[05:50] a day-to-day basis." And so the ride

[05:53] feel and being able to kind of fine-tune

[05:56] uh the the ride experience to make sure

[05:58] that it gets you where you need to go in

[06:00] the right amount of time is going to be

[06:02] really important because today uh human

[06:04] driver trips tend to be a little bit

[06:06] shorter uh from a a trip duration

[06:08] standpoint uh than AVs. Uh so it'll be

[06:11] really interesting to see uh AV

[06:14] companies like Mobilei continuing to

[06:16] kind of like move the dial on what the

[06:18] uh ride experience and feel uh of the AV

[06:22] after you know continuing to master all

[06:24] of the fundamentals of the autonomous

[06:25] driving because the the style is is

[06:27] actually very important.

[06:29] >> Sure. Yeah. It's fascinating how it

[06:31] works. So let's move on to another topic

[06:34] about economic e the economics of scale.

[06:37] We know that we can already hail a

[06:39] driverless taxi in cities like San

[06:41] Francisco, Austin, and Phoenix, but for

[06:44] most of the country, autonomous vehicles

[06:47] still feels like maybe five years away.

[06:50] What are the make or break economic

[06:52] realities of turning a test program into

[06:55] a viable business? And why has it been

[06:58] so hard to make self-driving mainstream

[07:00] everywhere?

[07:04] >> I can jump in and take a take a stab at

[07:06] that. and then I'll hand it over to JJ

[07:08] who is definitely in the best position

[07:10] to speak to the the engineering of what

[07:12] makes it hard to build a self-driving

[07:14] vehicle. Um you know the the reality you

[07:17] said Deborah is there are there are

[07:18] cities around the country where you see

[07:20] autonomous vehicles. Some of them are

[07:22] commercially deployed, some of them are

[07:24] in testing, but that we should all

[07:26] remember represents the very tip of the

[07:29] iceberg. And underneath that deployed

[07:32] asset, there is an entire value chain uh

[07:36] of different partners and ecosystem of

[07:39] players that needs to be marching in

[07:41] lock step in order to support the

[07:43] commercialization of that asset. And I

[07:46] think this is really important. So just

[07:47] want to unpack it for for a moment. uh

[07:50] uh that value chain starts with

[07:52] companies like Mobilei which are

[07:53] building the self-driving technology. It

[07:56] spans to OEMs, the auto manufacturers

[07:59] who are building and producing the

[08:00] vehicles. And today uh we're generally

[08:03] taking uh retrofitted vehicles. Uh so

[08:06] they're not they're not built for

[08:07] autonomous specifically and that means

[08:09] you need to kind of like tear them apart

[08:11] and then put them back together with the

[08:12] tech stack on it and that has

[08:13] significant implications from a cost and

[08:15] scale standpoint. And then once you have

[08:17] the vehicle and you have the AV stack on

[08:19] it, then you need a fleet manager and an

[08:22] operator and a financing partner who's

[08:24] going to hold that vehicle. Uh, and then

[08:27] you need a mobility marketplace where

[08:29] you can deploy that asset and

[08:30] commercialize it. And you need a

[08:32] front-end customer experience, a way

[08:34] that riders can interact with your

[08:36] technology. And uh in order to go from

[08:40] hundreds to thousands of vehicles,

[08:42] again, you really need all the different

[08:44] component parts of that value chain

[08:46] coming together uh in order to uh

[08:49] achieve sustainable economics. And uh I

[08:52] think that's where the the industry has

[08:54] a a growing appreciation for the

[08:56] complexity of doing that. Uh and that's

[08:58] where our partnership with Mobile Eye,

[09:00] the fact that Lyft owns and operates a

[09:02] subsidiary called Flex Drive. We 15,000

[09:05] vehicles that we directly manage today

[09:07] and obviously a thriving marketplace.

[09:08] These are all really important

[09:10] ingredients

[09:12] >> that is

[09:12] >> yeah maybe to add to that just just

[09:14] quickly um um is you know from a

[09:17] technical side I think you know in order

[09:19] to scale it's very important to have you

[09:22] know a product uh that uh basically has

[09:25] you know high efficiency and also is

[09:27] built from a cost perspective and u

[09:30] actually from an overall technical

[09:32] approach uh in a way that you can

[09:34] actually go quickly from city to city

[09:35] because this is something you know where

[09:37] we look back the last five years uh the

[09:40] deployment rate you know has been very

[09:41] slow. We are still not at the beginning

[09:43] of the actual scaling phase and you know

[09:45] for us as mobile you know we always say

[09:47] like safety is first and then

[09:49] scalability second and efficiency third.

[09:52] >> Yeah I love that. I mean from what

[09:54] you're saying it's pretty obvious that

[09:56] no single company is going to scale

[09:58] autonomous vehicles alone but

[10:00] collaboration in this in industry is

[10:02] notoriously tough. So can you talk about

[10:06] like a gridlock or or or a bottleneck in

[10:09] operational uh work between the OEMs and

[10:13] the mobility platforms and the provider

[10:15] that you feel is slowing down progress.

[10:19] >> So um based on my experience now and

[10:22] I've been in this space now also since

[10:24] uh you know about 15 years um this is

[10:27] actually not the case. I mean yes there

[10:28] is competition between you know

[10:30] automakers uh and there's competition

[10:32] between the technology providers and so

[10:34] on. So looking at these, you know, four

[10:36] or five like value chain layers uh that

[10:39] Stephen just explained basically um

[10:43] there's a lot of collaboration and and

[10:45] you know it's also there are different

[10:47] business models you know you look at you

[10:48] know maybe Whimo an all-in kind of more

[10:50] vertical type of approach and and look

[10:53] at our you know approach and

[10:54] partnerships uh for example with Lyft um

[10:57] and and with you know Folkswagen as our

[10:59] main strategic partner for uh uh you

[11:03] know the first VW ID as you know

[11:04] platform vehicle, beautiful vehicle also

[11:07] nicely integrated technology from from

[11:09] our side and uh also with others. So we

[11:12] try to be very open in this regards to

[11:14] actually work with different platform

[11:16] providers on the vehicle side but also

[11:18] on the go to market side and we think

[11:20] this is the better approach this open

[11:22] approach

[11:24] >> and I I I'll just chime in and agree

[11:26] with uh JJ's sentiment that I think

[11:29] whereas in the very early stages you saw

[11:32] some players say hey we need to

[11:33] vertically integrate and own the whole

[11:35] thing ourselves and both from a

[11:38] complexity standpoint as well as the

[11:40] growing recog recognition that the

[11:42] market opportunity here is enormous and

[11:45] we are in the bottom of the first

[11:47] inning. Uh I think more and more players

[11:50] are realizing that we can go further by

[11:53] partnering together uh and finding

[11:55] partners that have very complimentary

[11:57] skill sets.

[11:59] >> Great. Talk to me about the future.

[12:02] What's the next milestone that's going

[12:04] to really matter in the next two years?

[12:09] I'll answer this from a from a customer

[12:11] perspective and I'd love to hear JJ's uh

[12:13] perspective from a technical one. Um the

[12:16] metric that we at Lyft are going to be

[12:19] obsessing over for the next uh two years

[12:21] and beyond with uh our autonomous

[12:24] partners is the percentage of riders who

[12:28] opt in after taking uh an autonomous

[12:30] trip into the next autonomous trip.

[12:33] Because again, going back to what I said

[12:34] in the beginning, for a lot of people,

[12:36] AVs just aren't on their radar or it's a

[12:39] bit of a novelty. It's like, yeah, I'm

[12:41] coming to San Francisco. I want to try

[12:42] this new experience. It's kind of like

[12:44] going on a a new theme park ride at

[12:46] Disneyland. And then there are people

[12:49] who uh are are getting habituated to

[12:51] taking AVs, and we want more and more

[12:53] people in that category. And so you only

[12:55] have one chance to make a first

[12:57] impression. And we want to make sure

[12:59] that we are delighting writers in

[13:01] setting appropriate expectations and

[13:03] that they're giving that back to us by

[13:04] saying, "Yeah, I will take Navy with you

[13:06] again."

[13:08] >> And um I I personally, you know, I'm I'm

[13:11] very interested in and you know, looking

[13:13] forward to, you know, kind of this dual

[13:15] path, you know, of bringing this

[13:16] technology to market on the one hand

[13:18] with fleets. I mean this is kind of

[13:20] natural because the technology is you

[13:22] know pretty expensive at the moment. But

[13:25] then the second path was consumer

[13:26] vehicles. uh basically you know letting

[13:29] people um own or lease such vehicles

[13:32] maybe they even put them into fleets

[13:33] when they don't use them themselves but

[13:35] basically you know having consumer AVs

[13:37] and then fleet AVs um and and you know

[13:40] looking at you know which type of

[13:41] services what type of vehicle you know

[13:43] interiors exteriors I mean there's going

[13:45] to be a lot of innovation uh in in the

[13:48] in the vehicle design you know maybe

[13:49] there will be even collaboration between

[13:51] automakers and you know design studios

[13:54] or furniture companies or you you know,

[13:57] uh, interior designers to come up with

[13:59] completely, uh, new new designs,

[14:01] partnerships. You know, you can have an

[14:03] office on wheels, you can have a lounge

[14:04] on wheels, you can have, you know, movie

[14:07] theater on wheels, anything you want.

[14:08] And maybe, you know, depending on your

[14:10] needs and and once uh you order, you

[14:13] know, this or that type of uh vehicle or

[14:15] or rider service.

[14:17] >> I love the office on wheels. Okay, so

[14:20] this is a short question for both of

[14:22] you. What is one myth about autonomous

[14:26] vehicle safety that you want retired?

[14:32] >> I'll go. Uh I think the myth about

[14:34] autonomous vehicle safety is that safety

[14:37] is enough because safety is critical.

[14:40] It's necessary and it's not efficient.

[14:42] As we talked about earlier, if riders

[14:44] feel like the ride is jerky or it's too

[14:46] cautious, they're not going to be future

[14:48] customers of it. And so of course we

[14:51] need safety and we need a delightful

[14:54] experience around it.

[14:56] I personally actually think that uh one

[14:59] of those myth is that you know let's say

[15:01] as regular pedestrians or u you know

[15:04] maybe uh um another vehicle you know

[15:07] driver uh that you might act uh

[15:10] differently you know in front of an AV

[15:12] that people think oh it's an AV I can

[15:14] just you know jump in front of it you

[15:16] know even at you know two feet or three

[15:19] feet I mean basically you know people

[15:21] still need to consider you know the

[15:22] physical limits of you know breaking

[15:24] distance and uh uh and and uh you know

[15:28] reaction and so on. So I think uh you

[15:31] know from that perspective and I know

[15:32] there's you know there are discussions

[15:34] about you know should AVs have this you

[15:36] know light blue light you know when they

[15:37] are active so that people can see

[15:39] pedestrians and so on. Oh this is an AV

[15:41] or an AV function is on. I have to say

[15:43] I'm kind of against this and and uh

[15:45] believe that it's better that people

[15:47] have respect uh you know for any type of

[15:50] vehicles uh and any type of sizes uh

[15:53] because at the end of the day um these

[15:55] uh uh vehicles can only let's say react

[15:58] and break a certain uh let's say with a

[16:00] certain momentum and and brake power and

[16:02] and and so on you know basically just

[16:04] physical um u limitations and I think

[16:07] it's important uh that people you know

[16:10] treat AVs the same way as you know

[16:12] human-driven vehicles just, you know,

[16:14] safer and better.

[16:16] >> Yeah, it sounds like there's a lot to do

[16:18] to educate the market about this new

[16:20] technology. So, thank you JJ and Stephen

[16:23] for the great discussion. I really

[16:25] enjoyed it. It's fascinating to see how

[16:27] technology, trust, and collaboration are

[16:30] shaping the next chapter of mobility.

[16:33] Steve, back to you.

[16:36] >> Thank you, Deborah, JJ, and Stephen.

[16:38] That was a fascinating look at what

[16:40] comes next for autonomous fleets and

[16:42] innovation. Now, we're shifting gears to

[16:45] zoom out and look at the bigger picture.

[16:47] How cities, automakers, and regulators

[16:50] are shaping the infrastructure and the

[16:52] mindset needed to make autonomy work for

[16:55] everyday people.

[16:59] [Music]

[17:07] I'm thrilled to be joined by two leaders

[17:09] in this space. James Philin is the vice

[17:12] president of autonomy and AI at Rivian.

[17:15] And before joining the automaker, James

[17:17] spent years at the forefront of

[17:19] autonomous vehicle development, leading

[17:21] software and perception teams at both

[17:23] Whimo and Zuks, where he helped advance

[17:26] the systems that allow self-driving cars

[17:28] to see and understand the world around

[17:31] them. Now, he's building technology that

[17:33] makes advanced driver assistance and

[17:35] autonomy integral to Rivian vehicles.

[17:38] We're also joined by Charlie Tyson, who

[17:40] plays a key role in Michigan's

[17:42] autonomous vehicle pilot programs,

[17:44] advancing the state's efforts to turn

[17:46] transportation innovation into policy

[17:49] and infrastructure. He works at the

[17:51] intersection of government, industry,

[17:53] and research to make Michigan a national

[17:56] test bed for next generation mobility.

[17:58] James and Charlie, thank you both so

[18:00] much for being here.

[18:03] >> So, let's get right into it. is he

[18:06] >> let's talk about where we are the state

[18:08] of play today where we really are right

[18:10] now. How would you each describe the

[18:12] state of autonomous vehicle technology

[18:15] right now and specifically what's real

[18:18] what's still experimental and what's

[18:21] misunderstood. James, let's go to you

[18:23] first.

[18:25] >> Yeah, I mean I think you're starting to

[18:27] see the phase where um you know full

[18:29] autonomy has gone from the sort of

[18:30] science project into an actual product.

[18:32] Um, and you can go up to, you know, San

[18:34] Francisco, I can drive 40 miles north of

[18:36] me right now. And, you know, it was just

[18:39] going to be flooded with Whimos. Um, and

[18:41] now Zuks is as well. So, I'm I sort of

[18:44] um felt a few years ago that actually

[18:45] the the fundamental problems had been

[18:48] solved. That wasn't that doesn't mean

[18:49] that every every problem has been

[18:51] solved, but it was moving into a more

[18:52] engineering um and deployment and

[18:54] scalability phase.

[18:56] Um, and then I think, you know, with

[18:59] that hat on, you got to think how does

[19:01] this change how people um use

[19:04] transportation in the future. And I'm

[19:06] still a big believer in um personally

[19:08] owned vehicles. I think that for every

[19:11] mile done in a robo taxi, probably in

[19:13] the future, you know, more than 10 times

[19:15] will be done in personally owned

[19:17] vehicles. I think the economics of a

[19:19] robo taxi versus a personally owned

[19:20] vehicle still mean that those two modes

[19:23] will be around for a very long time. And

[19:25] so that's why I made the the leap to

[19:26] Rivian. Essentially, can we bring that

[19:29] sort of L4 technology back to the

[19:31] consumer space and really provide value

[19:32] for our customers.

[19:34] >> And Charlie, what do you think?

[19:37] >> Yeah, I think James hit the nail on the

[19:39] head there, but um from a state's

[19:41] perspective, I think that we are we are

[19:44] going from like the testing phase to

[19:47] some level of commercial operations, but

[19:49] I think we're seeing it in certain

[19:51] states. Um, for example, um, James

[19:53] mentioned California. We're seeing, um,

[19:56] I actually was just in Arizona. I took

[19:58] Whimos in Phoenix. Um, commercial

[20:01] operations really impressive. Um, but I

[20:03] think there are still some challenges.

[20:05] Um, the technology seems to be there for

[20:08] the most part. But um consumer consumer

[20:10] adoption, how does it how do these um

[20:13] techn how do these vehicles whether

[20:15] they're consumer AVs or um consumer

[20:18] vehicles with um some level of of uh

[20:22] autonomy or their um autonomous vehicle

[20:25] fleets? How do we integrate them into

[20:27] our existing transportation network? I

[20:29] think that's still a challenge that

[20:31] we're working on um here in Michigan and

[20:32] and really throughout the throughout the

[20:34] nation.

[20:35] >> You know, it's really interesting. I

[20:36] mean, I guess the one question I have

[20:38] for both of you is how do you get people

[20:40] to really feel comfortable about getting

[20:42] in an autonomous vehicle? I think about

[20:45] my parents who say that they'll never

[20:47] get in one, that they don't just don't

[20:49] trust the technology. And it seems like

[20:50] there's maybe a generational divide.

[20:52] Boomers feel one way, millennials,

[20:54] genzers. I'm curious how you all think

[20:57] about this trust issue, particularly

[20:59] from a generational perspective. And

[21:01] James, do you want to start first?

[21:04] Yeah, I mean my sense is it's pretty

[21:05] non-existent. I mean my my parents are

[21:07] also in that generation and um you know

[21:11] they're interested in what I'm doing but

[21:12] they don't really you know trust all

[21:14] these systems but you know the last time

[21:16] they visited we took a Whimo around um

[21:19] it's very you know very quickly becomes

[21:21] normal actually and um I feel like

[21:23] there's that initial kind of wow moment

[21:26] hesitation but then it's completely

[21:28] normalized people you know just go about

[21:30] their day. So, I think it's um I don't

[21:32] think the trust issue is is going to be

[21:34] um a persistent one. I also think that

[21:37] as people get more used to those higher

[21:40] levels of autonomy, they'll expect more

[21:41] autonomy in their vehicles. So, I think

[21:43] that trust issue actually trickles down

[21:45] and what you'll see is a huge um sort of

[21:47] competitive swing towards um you

[21:50] personally owned vehicles that offer

[21:51] those higher levels of autonomy. Um, so

[21:53] I think that's kind of a bit of the race

[21:55] you see right right now and one I think

[21:57] Rivian is, you know, perfectly poised to

[21:59] execute on.

[22:01] >> Do you agree, Charlie?

[22:02] >> Yeah. And I think Yeah, I agree with

[22:04] that totally. And I I think that, um, we

[22:07] just have to give individuals the the

[22:09] chance to experience the technology. Um

[22:12] that's why I think that although we want

[22:14] to see and we're pushing for commercial

[22:16] operations, um these pilot projects and

[22:19] these these testing activity that allows

[22:21] consumers to experience the technology

[22:23] in different use cases is still really

[22:25] critical. Um we've seen a number of

[22:28] pilot projects uh in Michigan um in some

[22:32] of our larger cities like Detroit of

[22:34] course, Grand Rapids, Ann Arbor. And the

[22:38] feedback from the surveys that we we

[22:40] sent out um was really interesting. Um

[22:43] uh most most of the the riders I think

[22:45] 90% of the riders in the survey

[22:48] respondents um you came back saying that

[22:50] they would absolutely love to take

[22:52] another um trip in an autonomous vehicle

[22:54] and they would recommend it to their

[22:55] their peers. Um they what was

[22:58] interesting though was um we asked a

[23:00] question in the survey um around

[23:03] removing the safety driver and I think

[23:05] that is a big piece here. That's where

[23:08] we started to see some of the um the

[23:10] comfort levels um change a little bit.

[23:12] Uh the survey respondents um I think it

[23:14] was about 5050, you know, responding

[23:18] that they would be willing to get get in

[23:19] an autonomous vehicle without a safety

[23:21] driver. So I think just um getting

[23:23] people more um accustomed with the

[23:26] technology um sharing the benefits um

[23:29] and also I think uh it's it's critical

[23:32] to um be able to you know educate the

[23:35] public um and just give them that

[23:37] experience to to get inside the vehicle

[23:40] um provide feedback uh and and let them

[23:43] know that these aren't being forced on

[23:45] them. They're they're being deployed in

[23:47] safe ways and and kind of a a crawl,

[23:49] walk, run approach. I think that's

[23:51] critical.

[23:52] >> Charlie, you mentioned these these

[23:53] pilots. Um, I guess my question is how

[23:56] how do you measure readiness? Is it, you

[23:59] know, the the number of miles driven? Is

[24:01] it disengagement rates? Is it just

[24:03] consumer trust or something else

[24:05] entirely? Like h how do we know that

[24:07] these things truly are ready to be out

[24:09] on the road?

[24:12] >> Yeah, I think um it's different for it's

[24:14] kind of case by case, different for each

[24:16] project. Um, for example, we had a

[24:18] project in Northern Michigan looking at

[24:21] um supporting a a a full-size autonomous

[24:24] transit bus deployed at Sleeping Bear

[24:27] Dunes National Park. So, looking at a a

[24:30] very unique use case in northern

[24:32] Michigan. Um, a majority of the of the

[24:35] riders were tourists visiting the area.

[24:37] they either uh they had challenges with

[24:40] with parking, but um I think uh it the

[24:44] way that we kind of measured read

[24:46] readiness for that project was how many

[24:47] times um the the autonomous vehicle um

[24:51] had to be essentially how many times the

[24:54] um safety driver had to take control of

[24:57] the vehicle. Um were there challenges

[24:59] with connectivity in that region due to

[25:02] to Wi-Fi or um you know infrastructure

[25:05] uh capabilities? Uh so I think that's

[25:07] the big thing is looking at

[25:08] disengagement um and just looking at

[25:11] based on the use case based on the

[25:12] region what are the key factors to

[25:14] enable um adoption and and that kind of

[25:16] is how we determine what are the key uh

[25:19] metrics around uh success.

[25:23] >> I'm curious

[25:23] >> yeah maybe I can just add to that please

[25:25] um yeah and just just talk about the

[25:27] Rivian process. So we we actually do an

[25:30] extensive kind of release readiness um

[25:33] process every month for our software and

[25:35] that involves you know many aspects to

[25:37] it the sort of metrics across the stack

[25:39] but also a key part of that is actually

[25:40] simulation. So we take millions of miles

[25:42] of real customer data and we can replay

[25:45] them through our stack and kind of

[25:47] measure the performance the safety the

[25:49] smoothness um and all those aspects and

[25:51] I think all of those pieces go into that

[25:53] readiness uh report. So I think that's a

[25:55] very important part is um having that

[25:58] scale of data and also the ability to to

[26:00] replay it and to to you know gain

[26:02] insights from it.

[26:03] >> Charlie, you mentioned in your survey

[26:05] data that you cited that there's a

[26:07] distinction between how people feel uh

[26:09] when they're in a self-driving car that

[26:12] does not have a driver versus one that

[26:14] does have a driver. And how do you sort

[26:16] of bridge the gap between people's

[26:18] perceptions on on those two experiences?

[26:23] Yeah, it's a great question. Um, I think

[26:26] it's understandable that someone may

[26:28] have um less willingness to get an

[26:32] autonomous vehicle without a safety

[26:33] driver. But I think that goes to the

[26:36] importance of getting autonomous fleets

[26:38] out there. For example, some of these

[26:40] states that have been doing it without a

[26:41] safety driver. Again, was just in a

[26:44] Whimo in Arizona. Um, and got out of the

[26:48] vehicle feeling totally safe. Um and I

[26:50] and so that kind of goes to um the

[26:52] importance of um states and and um

[26:56] industry working together, government

[26:58] and industry working together to safely

[26:59] deploy these vehicles and provide them

[27:01] and um you know provide the opportunity

[27:04] for the public to get in get in the

[27:06] vehicle. Um ideally we move from safety

[27:10] drivers to to non-safety drivers and

[27:12] fully autonomous vehicles. Um and that's

[27:14] what we're working on doing here in

[27:15] Michigan. But there are some challenges,

[27:16] right? And I think that's um why

[27:18] Michigan feels that we are um posi

[27:21] positioned well to be a kind of a great

[27:24] test bed to to move from from the

[27:27] testing phase to commercial operations.

[27:29] For example, how do we um ensure that

[27:32] autonomous vehicles can operate in in

[27:34] harsh weather conditions? In order to be

[27:36] able to fully um to see widescale

[27:38] adoption of AVs, we need them to be able

[27:40] to operate in in harsh weather

[27:42] conditions, rain, snow, etc. And so, um,

[27:45] there's certain states and and Michigan

[27:47] definitely feels we are one of them

[27:48] that, um, you not only due to our

[27:51] automotive heritage, but also, um, just

[27:54] our our demographics, our our weather

[27:56] conditions, um, we are positioned well

[27:58] to be able to to test these different,

[28:00] um, uses and to hopefully support, um,

[28:03] that path towards commercialization

[28:04] while we do so with, um, public

[28:07] engagement and and providing, you know,

[28:09] individuals the experience to, uh, to

[28:11] get in the vehicles. I always kind of go

[28:13] back to, you know, 10, 20 years ago, how

[28:16] many parents um were on social media?

[28:19] Not very many. I was always I was always

[28:20] getting flack from my parents for being

[28:22] on Facebook or or um you know, whatever

[28:25] in Instagram, for example. But, you

[28:27] know, now my grandma's on Facebook and

[28:29] she's she has a smartphone and and

[28:31] they're on it more than I am. So I think

[28:33] it just takes time for adoption and for

[28:35] for uh the public to get um accustomed

[28:38] to technology and feel comfortable in it

[28:40] and and I think industry is doing a

[28:42] great job and it's important for

[28:43] government to to align and to partner

[28:46] with industry to to make this path as

[28:48] smooth as possible.

[28:49] >> You know it's a great point you raised

[28:51] and just this week we obviously had the

[28:53] big Amazon AWS outage that knocked out

[28:56] wide parts of the internet and how

[28:58] people go about their everyday lives. To

[29:00] me, this is one of the concerns perhaps

[29:02] for for autonomous vehicles is what

[29:05] happens if there's an outage and then

[29:06] these things don't operate the way that

[29:08] they're supposed to operate. Um, what do

[29:11] you do? How do you adapt? Um, so how do

[29:13] you how do you both think about that?

[29:14] James, do you want to take that one

[29:15] first?

[29:18] >> Yeah. So, I think I think it's key to to

[29:20] in your safety case sort of think

[29:22] through these um these outages that

[29:24] could happen in the cloud. So the the

[29:26] the software that's on our Rivian

[29:28] vehicles um it it doesn't require a

[29:31] cloud connection. So the idea is that we

[29:32] can be um sort of in a safe state even

[29:36] if that external connection goes down.

[29:38] So there's all the processing happens on

[29:39] vehicle. I think um it's pretty

[29:42] important when you build a resilient

[29:44] system that you're not taking those

[29:45] dependencies unnecessarily or if you are

[29:47] that you have um sort of good backup

[29:49] processes. Um

[29:53] >> and Charlie, what do you think? And I

[29:55] would I would say I don't have a highly

[29:56] technical answer there. Um obviously

[29:58] that's it's a it's a challenge. We also

[30:00] worry about cyber security um the you

[30:03] know grid resiliency. But um one of the

[30:06] things that we want we try to do here in

[30:07] Michigan and um within the office of

[30:10] future mobility electrification is is we

[30:13] reach out we we like to engage industry

[30:15] and to call out some of these challenges

[30:16] and say how can you help us solve some

[30:19] of these challenges that we're talking

[30:20] about here? for example, um you know,

[30:22] the Amazon um outage, you know, we on an

[30:26] ongoing basis, we like to um call out or

[30:30] identify some of these ch these major

[30:31] challenges and concerns and

[30:32] considerations and and to work with ind

[30:34] industry to solve them. And that's why

[30:36] our our grant programs and our um pilot

[30:39] projects have been so successful.

[30:41] Identifying challenges and finding uh

[30:44] solution providers to address them and

[30:46] working together in public private

[30:47] partnerships is something that we are

[30:48] we've seen a lot of success in and

[30:50] continue to do. You know, Charlie, you

[30:53] mentioned you use the social media

[30:54] comparison 20 years ago, how people were

[30:56] just starting to figure out how to use

[30:58] social media. The older generation

[31:00] wasn't on it and now all of a sudden

[31:02] they're all on it. Um what do you think

[31:04] is the biggest variable now to achieving

[31:06] a fully autonomous world?

[31:12] Yeah, I think um

[31:16] I would probably say one economics. How

[31:18] do we make it economically feasible for

[31:20] the fleet operators or the auto um

[31:22] automakers um to fully integrate

[31:24] autonomous systems into their vehicles?

[31:26] Um I think it's going to be critical to

[31:29] continue supporting and seeing um

[31:32] increased level of autonomy and consumer

[31:35] u vehicles and and pro production

[31:37] vehicles. I think that will really help

[31:40] um you know uh drivers and individuals

[31:44] um feel comfortable with the technology

[31:46] getting it into their day-to-day

[31:48] vehicle. I think that's critical and

[31:50] also um being able to support um you

[31:53] know commercial fleets throughout um our

[31:56] nation and doing so uh and continuing to

[31:58] do so in a safe way. But um I I think

[32:02] addressing some of the challenges with

[32:04] infrastructure and addressing some of

[32:06] the challenges with um um deploying

[32:08] harsh weather conditions is something

[32:10] that is really important. Um but again,

[32:13] I'd probably, you know, go back to how

[32:15] do we get this um how do we get

[32:17] increased level of of autonomy, excuse

[32:20] me, into consumer vehicles and into more

[32:23] production vehicles. I think that's

[32:24] going to be um a critical way of of

[32:26] increasing adoption.

[32:30] James, what do you think?

[32:33] >> Um, yeah, I think I think, you know,

[32:35] Charlie's right that um a certain amount

[32:37] of this is just just time that people,

[32:39] you know, it takes time to try and then

[32:41] gain acceptance. I think it's actually

[32:43] changing, you know, pretty quickly. So,

[32:45] we we released our, you know, hands-free

[32:47] feature earlier this year and since

[32:50] then, you know, every month we've

[32:51] essentially seen the number of miles

[32:52] driven hands-free increase. So I think

[32:54] it's around 20% of all Rivian models now

[32:56] are done in that hands-free mode. So I

[32:59] think as these features get better, they

[33:00] become more capable, people become more

[33:02] comfortable with them, they get used to

[33:04] the, you know, the UI, the UX aspects.

[33:06] Um, and then also, you know, the system

[33:09] is learning through the data that we're

[33:10] able to gather um during those those

[33:13] drive events and all of that kind of

[33:15] ladders into this um sort of upward

[33:17] spiral of improvement. So, I think it's

[33:19] important that um OEMs are able to

[33:23] really learn from their fleets. I think

[33:25] that's something very new that they

[33:27] haven't had to do in the past. Um and we

[33:29] think it's a it's a key competitive

[33:31] advantage for us.

[33:33] >> It's so interesting as as adoption rates

[33:35] increase, I have to think too that

[33:38] driving fatality fatalities will

[33:39] ultimately go down. I mean, we have

[33:41] 40,000 driving fatalities a year right

[33:43] now with cars that people drive, right?

[33:46] Traditional cars. Um but so but at the

[33:49] same time though once if god forbid a

[33:53] whimo happens to kill someone who's

[33:55] crossing the street you're gonna get a

[33:56] there's gonna be a ton of attention on

[33:58] that and people are going to say

[34:00] computers kill people and how do you how

[34:02] do you go about this or or what do we do

[34:04] to combat this? So how do you all think

[34:06] about that and and the the safety

[34:08] element of all of this?

[34:12] >> Yeah, maybe I can start. So um I think

[34:15] we we have to recognize that the the

[34:17] baseline here is the average human

[34:18] driver and you know there's far too many

[34:20] fatalities in the US. I think the last

[34:23] stat I heard is something like a 747

[34:25] full of people die every day in the US

[34:27] you know on the roads in the US. So I

[34:29] think you know for me that's the number

[34:32] we have to drive down. We're not saying

[34:34] that these um systems are going to be

[34:37] perfect. I think that's an unrealistic

[34:39] expectation. In fact, I think if we have

[34:40] that expectation, we'll delay launching

[34:42] something that could be saving lives um

[34:45] potentially now. Um and so I think

[34:48] that's the number we should focus on and

[34:49] drive down. Um we spend a lot of work,

[34:52] you know, Rivian on our active safety

[34:54] systems. Um so a lot of that is actually

[34:56] fed by the same world model investments

[34:59] um on the ML side uh that are powering

[35:02] sort of the L2 plus features. And so

[35:04] really our goal is for you know Rivian

[35:06] vehicles to be the safest vehicles to be

[35:07] in and around. um through those active

[35:10] safety features. So you can imagine

[35:11] almost like a safety bubble where we're

[35:13] preventing the vehicle from getting into

[35:15] these collisions and I think systems

[35:17] like that deployed widely on consumer

[35:19] vehicles will really start to move the

[35:20] needle um on these fatality rates.

[35:25] Yeah, I think it's a really important um

[35:27] question and um I think one of the

[35:30] important things here is to um make sure

[35:33] that we are being with the government

[35:35] being in government um working closely

[35:38] with industry to ensure that safety

[35:40] standards are are top-notch and are um

[35:43] are we're aligning on um you know making

[35:47] sure that bad actors aren't able to

[35:50] access autonomous vehicles or you know,

[35:53] in cyber security um capabilities are

[35:55] are in place. Um and then making sure

[35:58] that we educate the community um that we

[36:00] that the vehicles are in. Um I think

[36:03] most most communities want to um know

[36:07] that the roads have be roads have gotten

[36:10] less and less safe. Um people, you know,

[36:14] technology has the ability to actually

[36:16] um make our make our road safer and the

[36:19] average person is not going to drive as

[36:22] safe as autonomous vehicle. Um they are

[36:26] um you know just to be able to trust the

[36:28] technology and and that's going to take

[36:29] some time and um just constant

[36:31] engagement with with communities that

[36:33] the the vehicles are in. I think that's

[36:34] really important. When you talk about

[36:36] trusting the technology, it gets me to

[36:38] this broader question in the industry

[36:40] about what's the better approach to

[36:42] autonomy. Is it, you know, it's the

[36:44] cameras versus the LAR question. Is it

[36:46] using machine learning or a rules-based

[36:49] system? James, what do you think?

[36:53] >> Yeah, I mean, I feel like those

[36:54] arguments have sort of largely been been

[36:56] settled. Um, so I think you really want

[36:59] to use and leverage machine learning for

[37:01] as much of the driving task as possible.

[37:03] The reason you you should do that is

[37:05] because um specifying driving in rules

[37:07] is actually it's very complicated. You

[37:09] end up with huge you know spaghetti code

[37:12] of heristics. Um it is actually not well

[37:15] specified in many cases. So you know

[37:17] think of the example of a lot of

[37:18] vehicles entering a a stop intersection.

[37:21] You know there are rules on how that's

[37:22] supposed to be handled but that's not

[37:24] how humans actually navigate those

[37:26] things. And so to sort of encode all of

[37:28] that in in a rulesbased system is is

[37:30] almost impossible. So we believe in um

[37:33] doing as much in the machine learning

[37:34] model as possible really learning from

[37:36] um customer driving data. There's an

[37:38] effort we have ongoing at the moment

[37:40] called the Rivian large driving model

[37:42] and this is supposed to be an offboard

[37:44] um huge model that can um essentially

[37:46] learn all the nuances of human driving

[37:49] from all the data we receive and then

[37:51] but then I think you have to you have to

[37:53] sort of couch that in a system that

[37:56] provides those those guardrails, right?

[37:58] So we we can use the ML um as much as

[38:01] possible especially to to have this sort

[38:03] of humanistic um and sort of nuanced

[38:05] understanding of how to drive but we can

[38:08] still have guardrails on that system

[38:09] that say okay I don't want to collide

[38:11] with anything I don't want to run a red

[38:12] light um I need to be cautious in this

[38:14] situation and so I think it's it's that

[38:16] combination using ML as much as possible

[38:19] because that is the most scalable and

[38:20] sort of powerful approach but then

[38:22] having still a rules-based um uh set of

[38:26] uh features at the end that you can use

[38:28] to kind of guarantee certain aspects

[38:30] about the driving behavior. And then I

[38:32] think on the when you come to sort of

[38:34] cameras versus versus lighter versus

[38:36] other modalities, I think for us we

[38:38] would say um you know more modalities is

[38:41] better and you know Charlie was talking

[38:42] about um uh you know adverse weather you

[38:45] see in Michigan and I think it you know

[38:48] it's exactly those cases where those

[38:50] additional modalities really really can

[38:52] help. Um so you know for us it's it's

[38:55] really you know can you get the right

[38:57] sensing sort of independence and um

[39:00] different views of the scene at an

[39:02] economically you know affordable price

[39:04] point for consumers and that's what

[39:06] we're really focused on.

[39:08] Yeah, and I I would say we are here in

[39:10] Michigan trying to to support industry

[39:13] and and getting to to where we don't

[39:16] industry and and fleets um autonomous

[39:19] vehicles aren't relying on um

[39:22] infrastructure,

[39:23] but at the same time, how do we improve

[39:25] infrastructure to make um to in increase

[39:29] redundant redundancy, improve safety? Um

[39:32] and so that's kind of what the approach

[39:34] we're taking. How do we support machine

[39:36] learning and autonomous vehicles that

[39:38] aren't completely reliable on VTOX or

[39:42] you know vehicle to infrastructure

[39:43] communication but being able to have

[39:45] extra redundancy um with roadside

[39:48] roadside units and and infrastructure

[39:50] technology that is able to provide kind

[39:52] of a backbone and um additional layers

[39:54] of safety. Uh we for example we have a

[39:57] um a corridor an autonomous vehicle

[39:59] corridor being built built out uh

[40:02] between Detroit and Ann Arbor about a 30

[40:04] m 39 mile segment of of interstate that

[40:07] will have um vehicle to vehicle

[40:10] communication technology um technology

[40:13] ve um vehicle to infrastructure

[40:16] capabilities um that will support the

[40:18] integration of autonomous fleets into

[40:20] our our normal traffic. Um, so I think

[40:22] that's the approach that will will be

[40:25] most su successful down the line and the

[40:27] approach that we're taking here in

[40:28] Michigan.

[40:29] >> You know, you've both touched on the

[40:31] weather issue and Charlie specifically,

[40:33] obviously you're sitting in Michigan

[40:34] where you guys have some pretty severe

[40:36] weather there. Um, what I mean, how do

[40:40] you is this the biggest variable to

[40:42] achieving a fully autonomous world? Is

[40:44] this how do you combat severe weather

[40:46] conditions?

[40:50] I would love to hear um James' thoughts

[40:51] from a technical perspective, but I I

[40:53] would say um you know, Michigan, yes, we

[40:56] have harsh winters, but our summers are

[40:58] absolutely beautiful. Please come visit.

[40:59] Um but I would say that um you know how

[41:02] do we support uh new technology that

[41:06] might be able to help um ensure that the

[41:07] sensors are clean um if there's if in in

[41:10] a rainstorm for example or how do we

[41:12] make sure how do we support industry in

[41:14] new technology that provides better

[41:16] cameras that cameras that can work in an

[41:19] adver adverse condition. So that's kind

[41:21] of what we're doing here is is providing

[41:23] the the platform throughout the state to

[41:26] be able to test new technology and

[41:28] identify, you know, startups and

[41:30] technology providers that are working on

[41:32] some cutting edge technology that will

[41:34] help AVs work in all types of

[41:37] conditions.

[41:39] >> Yeah, I think um you know Charlie

[41:40] touched on some of the the things you

[41:42] know I think you'd start with the

[41:43] sensors, right? So yeah, this sensors

[41:45] see clean and unluded. Can they can they

[41:48] see the scene? you know, in some some

[41:49] foggy conditions, some, you know, very

[41:51] heavy snow, um you actually can't really

[41:53] rely on cameras. You have to then um you

[41:55] know, start using radars. Um

[41:58] you know, even non-weather related

[42:00] scenarios like, you know, the sun, a low

[42:02] sun, you know, shining directly into the

[42:03] cameras that can be challenging um from

[42:05] a vision only. Again, that's where, you

[42:08] know, radar and LAR can really help. So,

[42:09] I think you sort of start start with

[42:11] that. You need to have that um that kind

[42:13] of patchwork of sensing capability and

[42:16] sort of different different sensors as

[42:17] well.

[42:18] Um I think then there's then the data

[42:20] piece is very key. So um how do people

[42:23] actually drive in snowstorms? It's not

[42:25] the same way um that you drive in an

[42:27] unluded, you know, sunny day, right? So

[42:29] do you have the machine learning and do

[42:31] you have the data flywheel that's that's

[42:33] telling you how to handle those

[42:34] situations? I think that's a that's a

[42:36] big advantage that um a kind of fully

[42:39] integrated um sort of databased OEM like

[42:43] Rivian has over for example like a rower

[42:46] taxi where you have to actually send

[42:47] those fleets to go and gather this

[42:49] specific data in these different

[42:50] conditions. Um and then finally um yeah

[42:54] sort of how how do the rules you know I

[42:56] talked about those guardrails. Do those

[42:57] guardrails need to change? you know, for

[42:59] example, um you know, when you when

[43:01] you've got snow on the ground, people

[43:03] often don't follow the lanes. They can't

[43:04] see them. So, you you sort of have these

[43:06] virtual lanes that pop up. Does that

[43:08] need to be taken into account in your in

[43:10] your guardrails? Um so, it's sort of

[43:12] it's multifaceted. I wouldn't say it's

[43:14] the primary challenge. I still think um

[43:17] you know, density um and complexity in

[43:21] urban areas is is typically where the

[43:24] those final um big challenges are. um

[43:27] you know James to just reason about you

[43:29] know many many objects in a scene maybe

[43:31] there's a lot of nuance negotiation and

[43:33] things happening those can be tricky to

[43:34] handle um in a good way

[43:36] >> on the density issue James you know I I

[43:38] took my first Whimo earlier this year

[43:40] out in San Francisco and like so many

[43:42] other people was just so so fascinated

[43:44] by it um and so and it seems like

[43:47] wherever you turn in San Francisco

[43:49] there's another Whimo on every street

[43:50] corner but in terms of New York City

[43:53] it's I I just walk around here and I

[43:55] think to myself how are we going to have

[43:56] these types of cars in New York City and

[43:58] they're Whimo is already testing um

[44:00] testing their fleet in in the city here.

[44:02] But I mean, how do you how do you adapt

[44:05] to these different densities in

[44:07] different cities and how all the

[44:08] different layouts and how everything is

[44:10] is so different in different places?

[44:13] >> Um yeah, so I think you know you hope

[44:15] that your that some of your base systems

[44:17] obviously generalize to those places.

[44:19] Now, of course, there's going to be, you

[44:21] know, traffic specific um kind of rules

[44:23] of the road almost that exist in that

[44:26] exist in New York but don't exist in San

[44:27] Francisco and things like that. And I

[44:28] think that's where the where that data

[44:30] flywheel really is important. And I

[44:33] mean, you talked about New York City,

[44:34] but you know, if you go outside of the

[44:35] US and you talk about, you know, a

[44:37] country like India where um you know,

[44:39] some of the driving's you know, even

[44:40] more, you know, intense I would say um

[44:43] and very different again. So you have

[44:44] sort of had to think like how does a

[44:46] system um scale and I think that it

[44:49] really sort of tips you in favor of the

[44:51] MLbased approaches right because there

[44:53] you can gather the data you can see how

[44:55] people drive you can uh you know learn

[44:57] start to learn how um to sort of mimic

[45:00] it versus you know rules based systems

[45:02] that can really they can be brittle so

[45:04] you take them to a new place and

[45:06] suddenly those rules um don't work

[45:07] anymore.

[45:09] James, you're a former way Whimo

[45:12] employee before you joined uh Rivian.

[45:14] So, I have to ask about the elephant

[45:16] room. I got to ask about Elon Musk and

[45:18] Tesla's approach uh to autonomous

[45:21] driving right now.

[45:23] How does Tesla compare to Whimo compared

[45:26] to Rivian and others? And who's who's

[45:29] getting it right and perhaps who's not

[45:31] not doing it as well?

[45:35] >> Um yes, I'm obviously not privy to, you

[45:37] know, what's going on inside Tesla. Um I

[45:39] think what I would say from the outside

[45:41] is that um I think they've

[45:45] you know on the good side they've really

[45:46] sort of pushed um the OEMs forwards in

[45:48] the sense that they took a very um sort

[45:51] of MLbased approach early on and um I

[45:55] think that is the right way to build

[45:56] these systems. Now on the on the counter

[45:58] side, I think they have um a sort of

[46:00] very uh rigid point of view I guess on

[46:03] um different sense modalities which I

[46:07] don't think is you know fully

[46:08] explainable just from an engineering

[46:09] point of view. So um I would say it's

[46:12] sort of a a mixed bag. Um I think we're

[46:15] really focused on um you know can we can

[46:18] we bring that L4 technology back to

[46:21] consumers in the best way possible. And

[46:23] I think um you know sensors are sensors

[46:25] can get you there faster and they can

[46:26] get you there in a more robust way. And

[46:28] I think the the price point of a lot of

[46:30] these sensors is is no longer that you

[46:33] know liars are $10,000 um because of the

[46:36] huge um scale that you've seen in China.

[46:38] Those are those are coming down to you

[46:40] know a few hundred which is very much in

[46:42] the in the envelope of um you know

[46:45] consumer vehicles. So, I think, you

[46:47] know, that's that's kind of the approach

[46:48] we're taking and the one that I think is

[46:49] is going to get us there um in the most

[46:52] sort of direct way.

[46:54] >> Charlie, what do you think?

[46:58] >> Yeah, you'll have to um recap that

[47:00] question one more time for me, Steve. My

[47:02] apologies.

[47:02] >> Just comparing the different approaches

[47:04] that uh automakers are taking from Whimo

[47:06] to Tesla. Um, Whimo I has this

[47:11] reputation in the industry for taking a

[47:13] slower perhaps more more cautious

[47:15] approach to how they go about things

[47:17] whereas Tesla is more of a move fast and

[47:20] break things kind of uh approach here

[47:22] and obviously they have the robo taxi

[47:23] fleet that's uh starting to be rolled

[47:25] out. So just curious your thoughts on

[47:27] the different approaches and the pros

[47:29] and cons of both.

[47:31] >> Yeah, you know I think

[47:34] we don't really have like a

[47:37] reference per se on on what a company's

[47:40] approach would be. Being with the state

[47:42] of Michigan's economic development

[47:43] department, we we want to support all

[47:46] companies that want to grow here in

[47:47] Michigan, hire here in Michigan. I think

[47:49] the big thing is um deploying in a safe

[47:52] way. That's that's the most important

[47:54] thing. whether you're taking a a fast

[47:56] approach with um using machine learning

[47:59] and and or you're taking a slower

[48:01] approach with a certain use case and and

[48:03] leveraging infrastructure. Um we don't

[48:06] really have a preference per se as I as

[48:08] I mentioned. It's just um we are trying

[48:10] to to make Michigan a great place for

[48:12] companies to thrive and to to uh to

[48:14] deploy their technology in a safe way

[48:16] within communities. And so um you know

[48:19] we are we think that we have a really

[48:21] strong obviously automotive industry

[48:23] here but how do we make sure that we we

[48:25] remain competitive and how do we kind of

[48:28] bridge the gap between legacy automotive

[48:30] and and the startup world in in the on

[48:33] the west coast and and bring the minds

[48:35] together to ensure that we're you know

[48:37] we see this widespread widespread

[48:39] adoption over time.

[48:41] James, what what do you think is the

[48:43] biggest bottleneck for the robo taxi

[48:45] industry right now?

[48:51] >> Um I I mean I'm no longer in, you know,

[48:53] the robo taxi industry. So um I I would

[48:56] say that um it just takes time to scale

[48:59] these things. They're, you know, fleets

[49:02] are physical things. You have to charge

[49:04] them somewhere. You have to clean them

[49:06] somewhere. you have to you know think

[49:07] about power and um the operational

[49:10] aspects and so that just takes time. Um

[49:13] so I think and and then of course

[49:16] there's you know there could also be you

[49:17] know further engineering development

[49:18] that has to happen to you know maybe

[49:20] cover specific cases that you see in new

[49:22] places. So I don't see a fundamental

[49:24] issue but I think you know this isn't

[49:27] like um I don't know Charlie mentioned

[49:29] social media right it's not like just

[49:30] downloading an app someone has to go and

[49:32] build these vehicles put all the sensors

[49:34] on they have to be kept somewhere they

[49:36] have to be you know kept clean

[49:37] operationalized and that takes time it's

[49:39] like a the physical investment aspects

[49:42] um are you know just take take time

[49:47] >> Charlie what do you think

[49:49] >> yeah you know I think um

[49:52] identif Identifying a clear use case is

[49:54] important for for fleet operations. Um

[49:57] deploying them in in a certain scenario.

[50:00] Um at least right now I think that's an

[50:02] important way to continue momentum and

[50:04] continue adoption. Um you know deploying

[50:07] them in a highly urban complex

[50:10] environment may not be the best best um

[50:13] option right now. So can we identify

[50:15] fixed routes or can we identify certain

[50:18] use cases that um fleets will thrive in

[50:20] and and um it's more feasible you know

[50:23] at this time and for example looking at

[50:26] how can we partner with large employers

[50:28] to provide um you know shuttle services

[50:30] between their facilities or can we

[50:32] deploy autonomous vehicles at airports

[50:34] from the parking to the terminals. um

[50:37] looking at deploying autonomous vehicles

[50:39] on at universities to get um students um

[50:43] engaged in to experience that

[50:45] technology. I think that's the that's a

[50:47] really um critical way at this juncture

[50:50] to continue, you know, supporting fleet

[50:52] operations uh before we see them in in

[50:55] highly complex multimodal environments.

[50:59] >> I'd like to shift gears a little bit. we

[51:01] could do a little bit of a a lightning

[51:02] round here on some of your predictions

[51:04] for what you guys think is going to come

[51:05] true in this industry in the coming

[51:07] years. So, uh I have a my older son is

[51:10] seven years old. In 10 years, he'll be

[51:12] eligible to get his driver's license

[51:13] here in New York. Will he need it and

[51:16] will he even want one? James, what do

[51:19] you think?

[51:22] >> Um so, I I grew up in London and even

[51:24] without autonomous vehicles, um you

[51:26] know, London has a fantastic, you know,

[51:28] public transport system. So when I was

[51:30] young and sticking in the city um you

[51:33] know I didn't need to drive and actually

[51:34] I took my driver's license quite late. I

[51:36] would say once you have a family and you

[51:38] start um needing to get from A to B

[51:42] that's when the real value of you know

[51:44] personally owned vehicle comes. So I I

[51:46] suspect your son will actually still um

[51:50] get a driver's license even if his early

[51:52] years are you know in in a rubber taxi.

[51:55] But I think the world in which the

[51:57] vehicle he'll be driving in I think will

[52:00] be much safer. It will have um you know

[52:03] many more autonomy modes. Um you as a

[52:06] parent, this is actually you know

[52:07] feature I'm excited about you know may

[52:09] be able to engage a teenager mode right

[52:11] which puts the vehicle in a in kind of

[52:13] like a extra safe state. Um so that uh

[52:17] you know your son can't can't speed or

[52:20] you know do donuts in the parking lot or

[52:22] whatever else he wants to do. Um, and

[52:24] so, uh, yeah, I think that's probably

[52:26] most likely. I I don't see a huge shift

[52:29] away, um, from the automobile, at least

[52:32] in the US, just because of, um, you

[52:35] would also need a a commensurate shift

[52:37] in, you know, where houses are built and

[52:39] how people are living and everything

[52:40] else. Um, but I do think that, uh,

[52:43] consumer autonomy will will become, uh,

[52:47] essentially a, you know, a must have on

[52:49] every vehicle.

[52:52] >> I would say must have on every vehicle.

[52:54] >> Your son will.

[52:55] >> Yeah, I would say your son will probably

[52:56] um need to get a driver's license. And I

[52:58] think that's okay. I think we we're

[53:00] going to continue to see more um AVs on

[53:02] the roads, but it's going to be a little

[53:04] bit longer down the line until you you

[53:06] don't need a um a driver's license. I

[53:09] actually p personally enjoy driving from

[53:11] time to time. Um there's also times

[53:13] where I I I would love to get an

[53:15] autonomous vehicle and not drive to get

[53:16] work done or to to get some rest, for

[53:19] example. Um, but uh I think that's a

[53:21] little ways out and it really depends on

[53:23] where you live your lifestyle as as

[53:26] James alluded to. If you're if you live

[53:28] in a rural community, most likely you're

[53:30] going to, you know, at least near-term

[53:33] over the next 5 10 years still want need

[53:36] a driver's license, want a driver's

[53:37] license. If you live in a highly urban

[53:39] area, maybe that's not the case.

[53:42] What percentage of cars on the road will

[53:45] be autonomous versus not in let's say 10

[53:48] years from now?

[53:51] >> 20%.

[53:53] >> What are we at right now?

[53:55] >> And what are we at right now?

[53:59] >> That's a good question. James, do you

[54:00] know?

[54:01] >> I would say sort of define autonomous.

[54:04] So, do you mean like a full robo taxi

[54:05] level autonomy or do you mean vehicles

[54:08] um that for example could provide an L3

[54:10] capability that gives you your time back

[54:12] on the road and makes makes driving

[54:13] safer? So, I think I think that's where

[54:15] there's actually a big spectrum of

[54:17] autonomy here and

[54:19] >> I I think we're I think people focus on

[54:22] the endpoint. Um but I think there's

[54:24] actually many other customer societal

[54:27] benefits you get on the way there. Um,

[54:30] so I think, you know, to be honest, I

[54:32] think every vehicle, new vehicle sold in

[54:34] 10 years time to be competitive will

[54:36] have to have, you know, close to

[54:38] best-in-class autonomy. I think it's

[54:40] becoming more and more of a consumer

[54:42] preference. We we see actually much

[54:45] higher conversion rates um when people

[54:47] try our autonomy features in the in the

[54:49] stores. I think it's something like a 3x

[54:52] conversion rate. And so I think um I

[54:54] think that that shift is really

[54:55] happening. And I think as you see, you

[54:56] know, robo taxis roll out, people will

[54:58] just expect and demand more and more. So

[55:00] I think I think this this tide is coming

[55:02] and um OEMs need to be ready.

[55:06] >> What what matters more better data or

[55:09] the algorithm?

[55:15] >> I think if you had to

[55:18] I think if you had to pick one, you

[55:20] would pick the data.

[55:22] Um but to make the best use of the data

[55:26] you need, you know, excellent, you know,

[55:29] machine learning engineers and

[55:30] approaches to to understand that data

[55:33] and to learn um you know, how to drive

[55:36] from it essentially.

[55:39] But the data is the most important. If

[55:40] you don't have the data, I think you you

[55:43] there's no getting around it really. So,

[55:44] you know, if you have if you don't have

[55:46] the Michigan snowstorm, I don't think

[55:48] there's any feasible way you could build

[55:51] an autonomy system that then can handle

[55:52] those Michigan snowtorrms. You have to

[55:54] go there and see the data.

[55:57] >> And actually, just sticking on that,

[55:58] James, um, Rivian is doing a lot in the

[56:00] Gen AI space. And so, do you want to

[56:02] talk just a little bit about um, how are

[56:04] you using Gen AI to, um, improve the

[56:06] autonomy that you guys are offering in

[56:07] your vehicles?

[56:10] >> Yeah. Yeah. So I think here's maybe two

[56:11] sens in which you know we we use genai.

[56:14] So I think um one is in this like large

[56:17] driving model that I alluded to earlier.

[56:18] So that is actually a very large

[56:21] transformer-based model. It looks a lot

[56:22] like um a large language model in the

[56:24] sense that you have you know data coming

[56:26] in in the LLM space. It's it's text in

[56:29] our in our side. It's really sensor data

[56:31] raw sensor data. And then you have these

[56:33] large transformers that that kind of

[56:34] chew on that data. And then at the end

[56:38] um we we generate tokens. And now these

[56:40] tokens are not words or or um you know

[56:42] letters in the LLM case. They're

[56:44] actually little snippets of trajectories

[56:46] and we kind of we piece them together

[56:48] and that gives you the the best um the

[56:51] model's best interpretation of the

[56:52] future driving path. Um so I think you

[56:56] know that that model is very heavily

[56:57] inspired by a lot of the work that's

[56:59] happening on LMS and of course we we

[57:01] kind of uh slipstream on the work that's

[57:03] happening there. I think that's one

[57:05] sense. The other sense is um we're also

[57:08] you know uh seeing a big sort of um

[57:10] productivity boost from using some of

[57:12] these genai tools. Um uh I would say not

[57:16] replacing people but really making them

[57:18] more effective. So um you know improving

[57:21] the velocity that you can write code

[57:22] that you can test code that you can do

[57:25] things like code review and um uh and

[57:28] you know interface with APIs and things

[57:30] like that. So I I think um yeah it it is

[57:33] an accelerant and and definitely

[57:35] um on both sides. I think very essential

[57:37] to the work we're doing on autonomy.

[57:40] >> Charlie, what's the biggest myth that

[57:42] you'd like to debunk about self-driving?

[57:48] >> That the techn is not ready. I think the

[57:50] technology industry has been um

[57:52] incredible. At least

[57:55] I think in the United States, you know,

[57:56] the technology being developed by

[57:58] industry, by academia is um has gotten

[58:01] us to a point where it's ready. Let's

[58:03] let's get these techn let's get these

[58:04] vehicles out there and get let's allow

[58:06] people to experience them. Um but I

[58:08] that's the big thing. I think too many

[58:10] people think that the technology is not

[58:12] ready, it's not safe, but um I think

[58:14] that's just the opposite.

[58:16] >> James, what do you think?

[58:19] >> Yeah, I you know, plus plus one to

[58:21] Charlie. Um I think um

[58:24] uh yeah I I think we we we just need to

[58:26] get more people experiencing um what is

[58:28] out there and I think we

[58:30] you know we need to push um US OEMs to

[58:34] to be more tech forward um you know you

[58:37] see how the level of autonomy features

[58:39] that are present in Chinese um OEMs and

[58:42] I think uh you know I kind of feel like

[58:45] OEM's got a little bit complacent um in

[58:48] in that regard and I think we have to

[58:49] push everyone forwards Um, so that's

[58:51] what we're, you know, we're trying to do

[58:53] here at Rivian.

[58:55] >> And finally for you, James, what's one

[58:57] word to describe the state of

[58:58] self-driving technology today?

[59:02] >> I think on the cusp, I know it's not one

[59:04] word, but it's, you know, yeah, one

[59:06] idea. So, I think it's really on the

[59:08] cusp where you um you're seeing

[59:10] significant deployments um in certain

[59:12] cities and I think um in in dense

[59:15] metros, I think that will happen quite

[59:17] quickly. And then I think you'll see um

[59:20] a trickle down into consumer vehicles

[59:22] for most other trips

[59:24] >> on the cusp. I like that. Charlie, what

[59:26] do you think? One word to describe the

[59:28] state of self-driving technology today

[59:31] >> here. I think it's here and we're going

[59:33] to see it more and more every day.

[59:36] >> Well, thank you James and Charlie for

[59:39] these insights. Uh and to everyone who

[59:41] joined us today for inside self-driving.

[59:43] I also want to thank our partner MobileI

[59:45] for their support of today's program.

[59:47] I'm Steve Russell from Business Insider.

[59:50] Thank you for being with us. We'll see

[59:51] you next time.

[1:00:03] [Music]

[1:00:34] [Music]