Transcript: Magna’s Vision for Autonomy Begins at Level 2

Executive Summary

In this episode of The Road to Autonomy podcast, Grayson Brulte sat down with David Doria, Director of Engineering – Automated Driving, Magna to discuss how the supplier is navigating the rapid transition to automated, software-defined vehicles. 

David shares his expert perspective on the critical debates shaping the industry, from the necessity of HD maps in complex urban environments to the evolving role of sensors like Lidar and 4D imaging radar. He also explains the profound shift from traditional, rule-based drive policies to end-to-end AI systems and how this will fundamentally change vehicle compute, software complexity, and safety validation.

Key The Road to Autonomy Episode Questions Answered

Key The Road to Autonomy Topics & Timestamps

[00:23] The Rapidly Changing Automotive Market

David Doria discusses the major shifts in the automotive industry, including advancements in hardware, an increasing volume of sensors, the exponential growth of compute power, and rising customer expectations, particularly in Asian markets.

[02:00] Centralized Architecture and Sensor Fusion 

The conversation covers the benefits of moving to a centralized vehicle architecture, which simplifies the system and allows for easier integration of sensor fusion and future hardware updates without a complete system rewrite.

[04:00] The Map vs. Mapless Debate in Autonomous Driving

David shares his perspective, arguing that while simple Level 2 features don’t need maps, they become highly valuable for more complex scenarios like highway navigation and urban driving by providing essential prior information.

[09:00] The Role and Evolution of Lidar in ADAS

Magna’s approach to Lidar is discussed. While it’s not used in their current production ADAS systems, it serves as the “gold standard” for ground-truthing during development and is seen as critical for redundancy in future Level 3 systems.

[11:00] Magna’s Work with 4D Imaging Radar 

David highlights Magna’s imaging radar product. It offers a much higher point density than traditional radar, enabling better object classification and prediction, similar to LiDAR but with the weather-robust properties of radar.

[15:00] The Shift to AI and Machine Learning in Drive Policy 

A key inflection point is identified: the move from using AI only for perception to applying it to the “drive policy” (the vehicle’s decision-making). This shifts the industry from rule-based algorithms to machine learning models that learn driving behavior.

[17:00] The Future of “End-to-End” Autonomous Systems

The ultimate goal of “end-to-end” systems is explored, where AI processes sensor data directly to produce vehicle commands. This aims to replace all human-written algorithms but creates new challenges in safety and validation.

[22:00] Magna’s Strategy on Partnerships vs. In-House Development

David explains that Magna maintains a flexible strategy, continuously evaluating whether it’s best to build technology in-house, form a partnership, or make an acquisition, with no fixed philosophical approach.

[29:00] How AI Will Reduce the Lines of Code in a Vehicle 

Contrary to the current trend, David predicts that the adoption of AI and neural networks will eventually reverse the growth of code complexity, as a single AI model can replace thousands of lines of human-written code.

[30:00] The Importance of Testing, Simulation, and Digital Twins

With the rise of AI-based systems, the focus of validation shifts. David emphasizes the critical role of simulation, synthetic data, and digital twins to test for countless edge cases that are impossible to cover with physical driving alone.

[37:00] How Autonomy Will Scale Globally (L2 vs. L4) 

The discussion concludes by looking at how different levels of autonomy will scale. Currently, Level 2 consumer ADAS and Level 4 robotaxis operate in separate markets with different hardware, software, and regulations, and a convergence is still far in the future.

Full Episode Transcript

The Road to Autonomy Introduction: Welcome to the Road to Autonomy, the podcast shaping how Wall Street and industry executives think about the autonomy economy for today’s executives and investors. Simply following the headlines is not enough. That’s why our clients go beyond the podcast and the headlines. To get exclusive market intelligence and industry briefings, see what they see. Visit road to autonomy.com. That’s road to autonomy.com.

Grayson Brulte: David, the automotive industry is changing. Magnus sitting at the epicenter of this, it’s becoming automated and it’s becoming software defined. You, you’re an engineer. How are you thinking about this rapidly changing automotive market? 

David Doria: Hey, Grayson. Uh, thanks for having me. It’s, it’s, uh, great to be here. The, yeah, you, you nailed it. I mean, rapidly changing is definitely the name of the game here. Uh, when you look at, you know, the hardware is changing more and more sensors, different kinds of sensors, uh, volume of sensors on each vehicle, cost of sensors going down. Then you’ve got the compute platforms, you know, the, the, uh, curves as you go, you know, very rapidly through time. The, the compute is growing almost exponentially. Uh, and the cost is also coming down. The expectation from, uh, customers is going up. You know, you look across markets, especially look into the Asian markets, and, uh, you have some pretty amazing things happening at very low price points, um, as far as urban navigation on autopilot and those types of features coming out. So, uh, yeah, rapidly changing and, you know, we’re, uh, we’re doing our best to stay ahead of the game there.

Grayson Brulte: How do you stay ahead of the game? ’cause if you, if you look at the Chinese market in particular, they have. Well received, well regarded vehicles at at low price points that are intuitive, software defined. Is that the early hints of a trend that might work its way to the east? 

David Doria: Yeah, I mean, I, I think one of the big shifts, uh, towards a centralized architecture is, is really helpful there. So, I mean, once you can kind of simplify the vehicle system. Uh, to have a kind of single large compute, then you can think about a lot of things like sensor fusion and you can think about replacing sensors over, you know, vehicle iterations. It’s not a whole rewrite of the whole system. Uh, so, you know, getting that in place kind of in the now timeframe so that in the next few years you can kind of start to move those sophistication forward. You know, that’s what we’re trying to do.

Grayson Brulte: From a systems perspective, and I’m gonna correlate this to let’s say a Boeing and Airbus, a, a large air jet jet airline where they have the, the fly by wire system. Then they have the IFE, the inflight entertainment system. The wires never cross. I I, is it fair to think about that from, from a vehicle standpoint where you have the autonomous driving system, the ADAS system? And then you have the, let’s call it the, the, in the v vehicle infotainment, where those wires never cross or the bits and the bites they never touch.

David Doria: Yeah, I think, you know, different customers have different viewpoints on that. Uh, I think we’re kind of exploring both ways. Some really wanna merge things and some really want to keep them separate. So, uh, you know, do we do ’em on different boards? Do we do ’em on the same board on different chips that are isolated? Or do we treat the whole system together? I mean, I think all those are still options that there’s not, uh, single way forward. Uh, you know, that’s kind of agreed on the, in the whole market.

Grayson Brulte: Why do you think that is? Is it, is it just the, the engineering cultures of your customers? Or, or, or is it just, we’re still such in the early days, we’re still trying to figure it out.

David Doria: Yeah. I think a lot of it may be kind of legacy mindsets that are, that are just tough to move. I mean, there’s a lot of momentum in some of this stuff. , You know, I, I’ve been, uh, kind of automotive adjacent for a lot of my career and I’ve only become, you know, in automotive for the last few years. So, I mean, some of that mindset is, uh. You know, a little bit clashes with some very modern, you know, general engineering principles and software engineering principles, and just ways to think about, uh, you know, networking across various components and things like that, that, uh, some, some customers have adopted very quickly and others have a lot more momentum. So, uh, you know, MAGNA tries to supply, you know, all of those players and, you know, we, we don’t necessarily intend to drive them in a direction, but rather kind of support them in what they’re trying to do.

Grayson Brulte: You mentioned the adjacent you, you, I would say you had one foot in automotive technically because you were at a world renowned. Mapping company and now you’re at a world renowned automotive supplier, in my opinion, does a lot more than just does that. You’re building the future at Magna. Which raises the question about maps there. There’s a big debate in traditional automotive. There’s a big debate in Silicon Valley Map mapped mapless. How are you thinking about that? ’cause you have years and years experience developing maps.

David Doria: it’s been a fun ride. Uh, so the, the, the HD maps, you know, maybe mid 2010s started to be kind of a real phenomenon. Uh, the thought at the time was you need an HD mapping company. Uh, they’re gonna maintain the world’s maps, and then you’re gonna build adas features on top of those maps. Uh, I still fairly well subscribe to that. I mean, I have a lot of bias because of that background, I guess. But, uh, you know, as, as the features become more sophisticated, you kind of start to see the value in that. Uh, you know, as you, if you look at a simple L two system with your adaptive cruise control and your lane centering. Those you don’t need maps for, right? Those are just very responsive to the environment. You’re looking kind of directly ahead of you. You’re looking kind of directly down at the lines ahead of you, and that’s all you need to, to control those features. These days, if you start to look at, you know, uh, highway assist types of features and, and, uh, highway navigate on autopilot, where now you need to understand the relationship of some junctions from the on and off ramps. Uh, you need to understand, you know, what you have to work with as far as adjacent lanes and shoulders and things like this. The map starts to be very valuable. Uh, then you start to look in urban cases and now you’ve got intersections and multiple traffic lights controlling multiple lanes and multiple turns through the intersection. So, uh, having some of that information as a prior is very valuable. Uh, you know, I think there are for sure companies that would argue, hey. You, if you can see it, then you’re a human, then you can, then you can drive it the same way with a camera and, and, and a, a chip, uh, you know, probably true and maybe in 10, 20 years that that can be more true. But, uh, I think now having that, at least as a prior, if not as a kind of a redundant system for cases where you can see everything, , is super, super valuable.

Grayson Brulte: So do you view maps, let’s say just, we’ll, we’ll see. This an L two system on a highway. I, I’m gonna use simple things ’cause I’m Florida, so Miami to Orlando, just on that, that, that stretch of 95. So let’s say that’s mapless. And then when you get into more complex, you want to go from downtown Brickell to Disney World. ’cause you have to go off the highway, believe it or not, to go to Disney World. It can’t drive right in. Does that become mapped? Is that kind of how we should think about it? Or, ’cause you’re adding a little bit of complexity on, on each end.

David Doria: You you’re talking about, do you drive highway, mapless and then switch to a different system when you get off the highway? I think it’s a little bit dangerous to have to switch systems hard like that because then you have to develop two systems. You know, I mean, I guess it’s, it’s possible. Uh, the trick with maps is freshness, right? So, so that was the whole exercise at a mapping company. So we spent, you know, a few years prototyping, uh, building HD maps from data kind of automatically. And, and that was. Kind of successful, and I won’t say easy, but you know, kind of like straightforward. Uh, but then you start to look at the logistics of how do you keep these fresh over time and immediately get into concepts of crowdsourcing, uh, because it’s just simply too slow and too expensive to drive industrial capture vehicles at the rate that you need to make sense. , And then you have kind of all kinds of conflicting reports from, from different vehicles, and you got, kind of the. Frequency of vehicles that you can expect on different kinds of roads. I mean, it becomes a, a kind of a nightmare of logistics to, to keep that fresh. And so then the question is, you know how, if you, if it’s not, if you don’t know how fresh it is, or if you can’t keep it fresh, then how much can you trust it or, you know, can you trust it at all? And the amount that you trust it, can you even get any value out of that? I think still the answer is probably yes if you’re treated in the right way. Uh, and I, you know, I frankly don’t think that we can do enough in real time without any priors yet that I would say we should not have maps. , But I mean, I think that’s something that some companies are looking at and, you know, we keep an eye on for, if there becomes a way to do it without a map, I think that would be, uh, a, a great, you know, system complexity reduction. But for now, I think the maps are useful.

Grayson Brulte: From a cost perspective, utilization perspective, you can also throw safety in there as well. Do you see primarily maps being used in controlled ODDs where robotaxis are? ’cause those vehicles are constantly going and updating the maps and they’re running in a constrained environment. Instead of that Miami Orlando, which is, it’s a common route, but you’re not gonna have the, the freshness that you are of running in a constrained ODD with robotaxis.

David Doria: Well, I mean, if you’re just using the map to to check, are you in the ODD? I mean that, that’s a kind of a simpler use case. You still be very careful there. You know that GPS precision and localization are working in harmony and all those kind of things. , So I mean, constraining the ODD is one thing, but I mean, actually using the map during the, the dry policy, so as you’re doing trajectory planning and things like this, that’s I think where it becomes more valuable.

Grayson Brulte: And so well, since we’re on the controversial topics, I’m gonna throw another one at it, LIDAR, ’cause you have Lidar experience. What are your thoughts on Lidar? 

David Doria: It’s been amazing to watch it evolve over the years. I mean, so, uh, you know, in, in grad school, a long time ago now, we were using Leica scanners where you have to take them out of literally a trunk and put them on a giant, , tripod and they take half an hour and you can literally watch the laser strips, you know, going up and down the scene. Uh, so from that to where we are today has been unbelievable to, to watch reduction. My take on LIDAR is, you know, we’re, we still use lidar for ground truthing, even though we don’t develop adas systems with lidar. So there’s, there’s some implicit, you know, we believe in lidar enough that we take that as the right answer. When we try to, you know, find what’s happening in the scene with radar and cameras and other things, we say, well, what did the lidar see? As we’re developing that, then that becomes kind of the, the, the gold standard. , You know, of course multimodal sensors are, are critical as you, as you start operating in more complex environments. You know, when you have inclement weather. , When you have nighttime, when you have, you know, all of these types, types of situations, uh, lidar may not always be the best, uh, you know, in, in a, in a nominal conditions. You know, I think it’s, it’s kind of still the gold standard though. So as the cost continues to come down, you know, we’ll watch if customers are willing to, to pay for it, and we can show them the value that you get in, in your ADAS stack out of having it there. That’d be great. You know, as you start to need more redundancies, they start looking at level three types of situations, you know, uh, that’s one way to get an an extra redundancy. And then as you start to get into, you know, magna’s not specifically looking right now at level four types of robotaxis, but those companies all seem to be using lidar. So there’s, there’s obviously something to that, uh, or they wouldn’t be doing that. Uh, so it’s definitely a hot topic.

Grayson Brulte: it’s a hot topic. Another hot topic is the the four, the 4D Radar. John Deere through Blue River recently acquired Spartan Radar and they had really great software algorithm running on top of of four D radars. Are you currently looking into 4D Radars and what you can do with them, especially from a software perspective? 

David Doria: So we have an imaging radar product that’s, uh, that’s kind of amazing. I mean, you can, you can get such a higher point density, , when you’re looking to, you know, if you use a traditional radar, you’re looking mainly at kind of obstacle identification and obstacle avoidance types of things. You know, it’s very hard to. To classify without additional sensors like cameras and other things, you know? And as you look into prediction, I mean, you get just kind of a rough blob. It’s much harder to tell which way is it oriented and, uh, those types of things. So with something like an imaging radar, you can do that much, much more effectively. I mean, it’s, it’s not, certainly not a lidar, but it’s kind of going in the direction of lidar and you get a lot of the same properties that you get with traditional radar, uh, as far as, you know, robustness to, uh, different weather conditions and those kinds of things.

Grayson Brulte: So you mentioned. At Magna from an a a s perspective is, is not using lidar on your, your a a s suites, and then when you go into higher levels of automation, you said that you are using lidar. How do you find from a development standpoint as you’re developing and testing these products, the, the sweet spot from a sensor perspective? 

David Doria: Yeah. Uh, you know, the ODD is everything, right? So if, if you are gonna be operating at high speeds, you need to be able to see far into the distance. You have time to react. , If you’re operating in urban environments, you need lots of different,, you know, sensor placement on the vehicle to try to avoid occlusion as much as you can. If you’re operating only in daytime, then you can think very differently. If you’re operating only in, uh, you know, non snow, non rain conditions, you can think very differently. So the, the problem statement is really essential. I mean, it’s, it’s very hard to say, okay, let’s assume nothing and then pick a sensor suite. Right? Just doesn’t make any sense. You, you really do have to, uh, design the. The vehicle system to the intended use case, uh, is simply too expensive to, to make it work in all conditions at the moment.

Grayson Brulte: Do you just also take into account the design of the vehicle when you’re looking at that potential ideal sensor suite? How it looks, how it functions, where, where you could put a sensor. ’cause from a consumer standpoint, I’ll call the consumer doesn’t wanna buy the Lego car. Okay, we gotta pump this sound. No, they want a smooth, I’ll use, I’ll use my father-in-law’s line. I want smooth lines kid. I want smooth lines. Consumers look for that when they purchase vehicles.

David Doria: For sure, for sure. And, and you know, something like a radar you can put behind the bumper and is literally invisible. Uh, people are getting pretty used to seeing little camera lenses poking out from places, and I don’t think that’s, that’s a big deal these days. Uh, you know, the giant spinning lidar on the top of the, the roof rack, that’s a, a very different look for sure. Uh, again, the market, the, the, the regional markets have very different opinions on that. So my understanding from my last trip to China was that, you know, customers there get excited and say, Hey, it’s got lidar, so it’s, it’s a fancy car. Right. So it’s very different than what you’re saying where people here say, what’s the big bump? I want the big bump on. Uh, in the, in the US Europe markets. So, you know, it’s not, the aesthetics are not a worldwide, uh, you know, they’re not standard necessarily worldwide, but, but for sure, those are considerations. I mean, you know, I don’t get into the, uh, aesthetic design of the vehicle. Uh, but, but of course, sometimes the customers will come with a constraint like that, and then you have to work around it.

Grayson Brulte: It’s interesting ’cause there are certain manufacturers where a badge means a lot from a a, a perspective That’s interesting about the, the, the, the, the Chinese market. I never thought about that. I was in London last week. And I saw this white van going around and I was like, oh my God, it’s got a lidar. I had to look closely. It was just, it was just something to hold it on there, but it was one of those things that, that caught my attention. It’s really interesting that you bring that up. So let’s stay on the consumer vehicles here, because you’re, you’re build developing this software. You’re prototyping it, you’re, you’re deploying it, if you wanna call it. The brains is the compute is the power that enables the sensors, enables your software to run. How do you think about com compute? Do you say, oh, I gotta have as much compute as possible, I gotta be open AI and I gotta have a gazillion data centers. You say, wait a second, I want the arm model. I want as little as possible. ’cause I want this to run.

David Doria: Yeah, I mean, we’re, we’re, we’re right on an inflection point, uh, of, of how these vehicles work. So, I mean, there’s kind of now what I’ll call like traditional ADAS systems, and then there’s. What we can see, and, you know, starting to, to roll out a little bit of, uh, what we’ll call very, very modern a ADAS as a systems approach. And that’s how much, you know, AI or machine learning that you’re using in the system. Uh, so if you, if you very roughly divide an ADAS system in half, you’ve kind of got the perception half. So that’s, you know, take the input from the sensors and build yourself an environment model, which means, now I know where the road is, where the lines are, where the obstacles are, where the signs are, everything like that. So that’s your environment model. Environment model then is kind of the input to the second half of the system, which we call drive policy. Or, you know, the, the behavior of the vehicle. So what do I want to do and how do I execute that within the environment that I’ve just built for myself with the perception stack? So, you know, for a while now, AI ML has been used in the perception half of the system. This is very, very standard. This has now been with the, uh, convolutional neural network, you know, revolution 10 years ago, uh, you know, we’re getting pretty good in the industry of detecting objects in, in cameras, and now doing some, some sensor fusion with cameras and radars and those kinds of things. What hasn’t yet really taken over is, uh, IML in the second half of the system. So, you know, we kind of have what we’ll call rule-based, uh, operation of, in the drive policy system now, where we say, okay, given this environment model, here’s the, the objective I want and I’m gonna use kind of, uh, you know, an algorithmic approach where we say, we’ll use this type of planning algorithm and then use this type of obstacle avoidance, uh, you know, internal lab knobs and say how aggressive you wanna be and these kind of things. , Where kind of, you know, where it looks like things are heading in the future. And the question is, how fast can we get there? But that’s, let’s now take, uh, a machine learning approach to the second half of the system where we say, here’s your input. Here’s how a human driver behaved, or here’s how some kind of existing system behaved. Uh, and let’s learn how to produce that behavior with a, with an artificial intelligence. Uh, you can now get into all kinds of modern things, and you can use LLMs for some of this. But what my point to answer your question is when you make that jump, now all of a sudden you have a compute problem on your hands. Uh, with these traditional approaches, you know, we’re starting to have enough compute in the car where that’s not necessarily a limiting factor right now.

Grayson Brulte: Interesting. You’re right. So right now the, the drive policy is, it’s set, it gets shipped into the vehicle, and then in the future, I’ll use the word potentially. It could go to more algorithmic. Is that a fair way to summarize that? 

David Doria: Yeah, I mean it’s, it’s the, the kind of holy grail is what they’re calling end-to-end systems. So that’s, uh, kind of sensor data in and vehicle actuation out. Uh, you know, there’s, there’s kind of well-known problems in academics of what I guess they call, like, inspect ability. So it’s kind of understanding what the network is doing. So not always you want to treat it as just a giant network like that. That division in the middle still is very useful, uh, for human understanding, so you can actually see the environment. Before you kind of continue with the system, just, just to try to help understand the, you know, where, uh, the deficiencies are, but in, in spirit what the goal is to have none of the algorithms that have been written by humans, but, you know, let the, let the AI figure it all out. Uh, you know, you got all kinds of big bombs that come in with how do we, uh, guarantee safety and how do we think about testing those systems and, you know, so there’s, there’s a revolution on the horizon. Uh, and, you know, frankly, that’s what I, what, what, what, uh, my teams are looking at. So, in, in the division of MAGNA electronics, where I live, we’re called technology strategy. So we’re trying to identify those trends, you know, get ahead of those trends so it doesn’t happen and we watch it fly right by us because there’s, there’s major things both on the kind of, uh, you know, the vehicle architecture and compute side. There’s kind of the pre vehicle stuff in the sense of what do we need for, as far as machine learning infrastructure and machine learning operations at the company. Uh, so there’s a lot of stuff that we want to be prepared for, for when it finally comes, you know, we can kind of stay ahead of it.

Grayson Brulte: Because EDE is bubbling up. I mean, the research is, is, is, is vast and you’re starting, you’re starting to see companies look to commercialize EDE, but with all due respect, they don’t have your balance sheet and they don’t have your reputation. So they could do things that you’re not able to do. Do you see a point, and I don’t know if this is possible, where you get an EDE, but you kind of put guardrails or or verification around it. Is that even possible? 

David Doria: although we’re not building Level four systems ourself, I mean, I follow that space quite a bit because there’s a lot of good lessons to be learned there. Uh, that’s, that’s definitely what a lot of companies that have been successful there have been doing where they’re. Uh, you know, essentially they, they take a machine learning approach. They output several, uh, you know, kind of strategies, and then they have a couple of kind of traditional strategies and they, they try to look at what’s the difference between those. Uh, if, if they’re very drastically different from each other, they kind of throw flags. So, I mean, that, that’s, those guardrails and checks using traditional methods are, are definitely still, uh, in, in a lot of cases the approach to. The kind of safety problem in, in a sense, uh, you know, I, I, the, the, the trick is as a situation to get more and more complex, then the traditional methods are gonna stop working at some point. And then what do we do then? Uh, so, you know, we, I, I think the thought is that we’ll kind of develop strategies and solve a lot of those problems for how do you treat the, the testing and validation. By, by the time we need that to be the only backstop, right? So the, the traditional backstop will, I mean, it, it has to eventually go away. Uh, all of this is a matter of when.

Grayson Brulte: When do you think is when.

David Doria: I, I think, I think, you know, there’s kind of the running joke, right? E everybody about every five years somebody comes in loudly says it’ll be ready in five years. Uh, it’s definitely been about the last 20 years and I think it’s still happening, literally as we speak. So, uh, so could, can I be first to go on the record and say five years? No, it’s, uh, you know, I bet it it’ll be a good decade. Uh, there’s, there’s a lot of runway left. I think we, we’ve got. You know, you get the SAE levels of automation, right? We’re, we’re still operating in level two. Uh, but, but in fact, I gave a talk recently. We called a Scope Creek bene in level two. So what used to be level two and still is the definable as level two, is adaptive cruise control plus, uh, land centering, right? That’s obviously not what we mean by kind of advanced level two systems anymore. So we’ve got official terms in the industry of level two plus level two plus plus. Then we start to talk about level 2.9 and you know, this kind of, uh, some of ’em are joke. But some of them are, are kind of seriously trying to indicate, you know, how this system is really advanced past what the kind of baseline definition of level two is. Uh, but we’re still not prepared to, you know, be confident enough that we can call it level three. So it’s, it’s, it’s a matter of, uh, you know, how much redundancy you have before you can get into a level three system.

Grayson Brulte: ‘ cause I feel the level three, at least from a policy standpoint, opens a can of worms because then you get into. The, the liability issues, which for the record, in my humble opinion, have not been solved yet. But you’re, you’re right about plus and plus plus it started in the media. Disney plus this service plus everything, got a plus and then an automotive. He said, no, no, no, no, no. We have to have two pluses and then I’m gonna be on, go on the record and say this. We’re probably gonna have three pluses at at some point. It’s just the way that things are moving and the technology is moving so, so quickly. Do you look when you’re developing the internal, because this is simple, the level two ADAS systems, do you look to, to potential partnerships with ADAS startup developers? Or do, or do you build it all in-house or how are you looking at these? Well, you said ED, E, but these new emerging models, these new emerging technologies of ways to, to solve ADAS.

David Doria: Yeah. I mean, we’re, we’re always on the lookout for, you know, which option is best. We don’t, we don’t have a philosophical, we’re gonna do this all in-house because we want to, and we don’t have a philosophy of we’re gonna go make an acquisition or a partnership to, to do it. Right. We’re kind. Of literally day to day, month to month, saying what’s the best strategy. So all, all options, uh, are kind of continuously on the tables. Once in a while we see something come around that looks promising and we try to do some proof of concepts together. , You know, we, we have to always consider the kind of long-term maintenance and, uh, stuff like that too. So, uh, there’s a lot of considerations. Uh, it’s, it’s really a complicated problem. Uh, but that, that’s what we do is try to make recommendations around those types of things.

Grayson Brulte: That’s a, I mean, that’s a great way to be because you can, you’re having the ability to see things and you have the resources, if so, choose to make a decision. And if you, you stand on that trend, you look at major trends that are emerging today, one of the biggest trends now is the licensing. Of ADAS systems. And I say that, and that does include L two. We have a very public partnership between Helm AI and Honda. You have one between Wayve and Nissan I believe, based on the press release that’s for the Japanese market. I have, we have no indications it expands outside of that, but Honda, they’re ramping up in, in 27 with the Helm vehicles. Do you see this trend continuing where you have these global manufacturers are looking to license stacks? 

David Doria: sure. I mean, it’s def that’s definitely a, a valid approach, right? I mean, this is a hard problem. To solve, uh, you know, Magna’s kind of a traditionally, uh, automotive manufacturing company. And now we’re gonna getting into this very high tech space, very rapidly of, of ADA systems, uh, or software systems, right? So, you know, there, there are very different business considerations. There’s a lot of upfront investment. Uh, and you know, sometimes when you look at what these partnerships that you see that you just named, , all these other smaller companies have been able to make those investments. ’cause they come from very different types of, uh, investors and they’ve got different timelines and all these kinds of things. So sometimes once it, you kind of matures then, and it’s ready for primetime, there’s no reason to then compete against it, you know, you might as well use it. Uh, so that’s, you know, we’re, we’re, like I mentioned, still, still. You know, trying to make those decisions, right? We, we have to pick a direction, you know, in the next, uh, half a decade so that we we’re kind of set up for success for the longer term future. Uh, and those, those decisions have to happen now.

Grayson Brulte: No you, but you are making decisions. Magna, in my opinion, has the right lead leadership to make those decisions. And if you look at the course of Magna’s history over time, Magna has continued to make, in my opinion, the right ones. And that’s why I’ll say it for the record. You’re the leading tier one supplier.. Because the decisions that the executives have made over the years. I’m curious now from a software perspective, how long has Magna been developing ADAS software internally? 

David Doria: Yeah, well, I, I’ve, uh, got almost, I think now four years with Magna, so it, it was well underway by the time I got here. So, but I can’t speak necessarily to the, to the longer history than that.

Grayson Brulte: And do you, do you see that Magnus commitment to software, developing software for ADAS, but eventually going potentially to higher levels, accelerating over time as the demands of your customers increases? 

David Doria: Yeah, absolutely. I mean, the customers come with very high expectations these days. You know, even over my tenure here. The sophistication in-house has gotten, you know, substantially better. So we’re definitely on the right trajectory. Uh, you know, we gotta just keep up the trajectory ’cause everybody else is giving up the trajectory. I mean, it’s, it’s really, again, another recent talk I gave and we call it the, you know, the race to autonomy we talked about a little bit. Uh, what we’re realizing though is right. It is, it’s okay. There’s obviously some element of a, of a race in the business then still, but, uh, it’s not a, uh, single person, sorry, single company who’s going to take all the elements and push them all as fast as they can to the goal, right? It’s, it’s really a partnership or an ecosystem that has to develop. Uh, and you know, as a tier one, we’re well positioned to kind of foster those types of relationships and kind of build those ecosystems, uh, from our compute partners. You’ve seen, uh, public news with some big, uh, collaborations with Nvidia recently. You know, that’s, that’s gonna be a huge one. Uh, you know, the software stuff that we’ve been talking about with these end-to-end systems and some of these, uh, AI approaches, you’re starting to get AI in the cabin, right? So, kind of, uh, simplifying the, the user interfaces for a lot of things and more. Natural, uh, discussions with the vehicle in, in natural language. Uh, you know, we, we can’t do all of that ourselves at a kind of level that the, that the customers and the consumers expect. Uh, so, but we still have to provide systems that, uh, offer all of that, right? So we have to pick and choose, whereas it makes sense for us to, uh, kind of become the experts and where does it make sense for us to keep our kind of capacity as an integrator, uh, and, and do that integration in a way that makes a lot of sense.

Grayson Brulte: you hit the nail on the head. It is an ecosystem. You just can’t, um, decide, oh, I’m gonna make a car drives called bitty boppy boop. And, and away it goes. There’s, there’s a hole there, there’s software, there’s infrastructure, there’s hardware, there’s connectivity. You and I can go on for days about all the pieces of the ecosystem that need to make this to work. And I’m thinking a lot about the ecosystem because there’s a lot of, I’ll use the word controversy in the market where. Some firms not gonna name names, think that they can build automotive grade hardware and then you can meet all the asil standards for software. And then there’s other firms say, no, we need to have a, a partner. Do you ever see a point in the future where you have a, a corporate contractual relationship with a startup where you become their all automotive grade partner to meet all the automotive safety standards and also become their integrator? Because do you see anything like that potentially happening in the future? 

David Doria: Yeah, I mean, I, I don’t have any specific examples to name, but, uh, you know, Magna’s very good at that stuff. Frankly, that stuff is, is, is a whole different type of hard, right. I come from a software background, mostly algorithm level. Uh, when you start looking at safety systems, you talked about asil and those kinds of things. And, and you’ve got the, the different Misra standards, even for the software itself. Uh, that’s a whole nother expertise that actually MAGNA does very well. Uh, so you know. Could we do that as kind of a software supplier in some sense? Uh, yeah, I think it’s possible. I mean, everything is possible. Uh, is is just a matter of where we wanna focus our kind of limited, uh, energy and attention, uh, and, and what makes the most sense for the business.

Grayson Brulte: That, that’s a great answer. I I, I like that answer. I was at a conference last year and this individual gave a presentation about all the lines of code in a car and how it’s rapidly grown. Your software engineer, does that trend continue? Do we discontinue to see more and more lines of code in that car or where do you see that going? And then does it get to the point where there’s too much code in the car or is it it, it’s too bulky? When do we hit that? If you wanna call that, that tipping point.

David Doria: Well, I think, I mean, it’s gonna, it is gonna reverse direction for sure with this introduction of ai, right? So anytime you introduce a, a, a neural network or similar that’s, uh, able to do a lot of these functions, you get to remove a lot of code, right? So you, you’re trading off, uh, lines of code for a little bit of uncertainty in the behavior, right? But that’s where I think the testing becomes really important. And that’s another area that we’re looking at really, uh, intensively is kind of. How can we use simulation and synthetic data and all these types of things to, uh, you know, as a way to develop answers to how can we prove this is safe if we can’t audit every single line of code? You know, that that’s a, that’s a huge open question in the industry. And, uh, you know, that’s, that’s something that I think we’re, again, pretty well positioned. We’ve got, you know, really big validation teams and stuff like this, so we’ve, we’ve kind of got the. The, the kind of knowledge and the manpower to adopt some of these new techniques. And, uh, it’s a matter of kind of choosing the right ones at the right time.

Grayson Brulte: Is that where digital twins come into the picture? 

David Doria: Yeah, digital, I mean, that’s a little bit of an overloaded term, so depending on who you talk to, you’ll probably get 10 different definitions. Uh, you know, you, you can talk about digitally twinning, the, uh, the, the hardware where you talk about VCUs and kind of modeling the sensors and all that kind of stuff at the physics level. You can talk about, you know, kind of what I’ll call adas level simulation, where you’re taking video game engine types of things. And, uh, running around, uh, creating scenarios to, to see how the, the ADAS stack would behave. You know, so there, there’s a, there’s a whole depth, deep, deep, uh, stack you can talk about when you talk about digital twins. Some element of that is already happening and all of those elements are, are really, uh. Impactful and, and essential to success in the future? You know, as these systems get more complicated, it’s not a matter of, okay, we’ll take it in the parking lot and drive it around and call it safe. Uh, you know, you’ve, you’ve got so many edge cases to handle when now, when you’re moving into these urban navigate on autopilot, uh, cases, uh, features specifically. You know, did a dog jump out in the road? These are now things that are actually possible and that you have to handle properly, uh, when you’re, when you’re restricted to the highway. You know, again, from my, from my mapping language, we call these limited access roads, where you, you kind of are allowed to make the assumption that nothing’s gonna enter the roadway except from the ramps. Uh, you can absolutely not make that assumption when you’re in an urban environment. Uh, so it’s, it’s, uh. It’s, it’s really the, the road is really a, a wild place. I mean, it’s so much more complex than you would ever, uh, just casually think about when you, when you actually study what’s happening there. Uh, and being able to handle those situations is, is really complicated.

Grayson Brulte: I went for a ride around urban London in the Navy last week. And some of the stuff I saw I’ve never seen here in America. First of all, I’m on the left hand side of the road, so that was cool. But some of the, if you wanna call it pedestrian behavior, it’s like, well, I’ve never seen that before. You have to plan for that. And then I saw something I never saw and the, the vehicle handled it flawlessly was a, um, a double roundabout. It’s like, what the heck? And people come in there, you know, my daughter calls it Mario Kart Drive, and they come in there, Mario Kart Drive, but, and then, and the consumers jump off the curb. You have to prepare for all this stuff.

David Doria: Yep. Yep. And, and that, that’s where, that’s where, you know, having a good testing strategy is really essential. You know, they used to talk about the number of miles and they’d say, okay, if you have a million miles of data and you can, uh, pass it all, then you’re safe. Right? I mean, yeah. To how many nines and to, uh, to wood level of confidence. I mean, so that, that’s kind of the continue we’re gonna. Continually chase those nines. , And you’ve got now, you know, I talked about an ecosystem. It’s not even just an engineering ecosystem now. It’s a regulatory ecosystem. It’s a consumer, uh, trust and consumer thinking ecosystem where you know how safe is safe enough for the regulators. How safe is safe enough for the consumers. You see amazing progress. And then there’s one incident after, you know, hundreds of millions of miles, and then the whole company, not, not, uh, our company, but you’ve seen in the, in the public eye, a whole company loses trust and loses funding because of one incident. Uh, so, you know, to, to hold the industry to that high of a standard. Uh, I think is a, you know, I, I’m not gonna comment on, uh, politics here, but I think it’s, it’s, uh, not necessarily in the best interest of progress, right? We wanna, we, we have to always be cognizant of, of, you know, safety number one for sure. But we, we can’t have such a high bar that we don’t make any progress. And so finding that balance is really essential.

Grayson Brulte: We have, we have to move forward. If you, if you study history and you go back to the railroads or the introduction of. The automobile there, there were incidents and there was, uh, a public trusts was built. There was a really great book. It was called The Vagabonds that was written. It was about Thomas Edison, Henry Ford Firestone, and I forget the other gentleman. And they went around on these, on these famous road trips to introduce the vehicle to the public, and it started to build this trust. Oh, Mr. Ford is coming. Mr. Edison is coming. And they’d have parades. They, we also saw that with the, the airplane with Charles Lindbergh. And when he did the tour of St. Louis. Where they would go out and build public system is an industry that we have to build this trust because at the end of the day, autonomy is going to unlock the ability for individuals to have better qualities of life. I think that you hear these stories over and over again around, I had to have the talk, you know, like, oh, here we go again. You know, dad, mom, no more drive. You gotta take the keys away. Well, if you introduce higher levels of autonomy, you eventually get to full L four. You don’t have to have the talk anymore. And then think about it from your parents’ perspective or your grandparents’ perspective. How much better a quality of life they can have. They’re fully independent. Oh, I wanna go to the beauty parlor. I wanna go play croquet, or I wanna do this. They can go do it. It it changes. Changes everything.

David Doria: Yeah, I mean, that, that has been the vision for literally decades and, and I think it’s still the vision. So I mean, that, that’s, I, I, and we’ll get there, right? I mean, I, I can pretty confidently tell you that, that we’ll get there. It’s again, a matter of when, the, you know, in the, in the meantime, right? When we’re not talking about level four, when we’re talking about level two. Uh, you’ve got still, uh, you know, I think it’s improving over time. You know, we look at the kind of consumer surveys every year and, uh, I think it’s, for the most part getting better over time. Uh, but, but you’ve got still pretty big factions of people that say, the first thing I do is figure out how to turn off all the annoying beeping. Guys, those beeps are supposed to be helping you, right? So, yeah. But, but there’s still, there’s certainly room for improvement there, right? So some, some of that we can take lessons and find, uh, better ways to, to interact with the driver, right? We’re starting to, Magnus got some driver monitoring systems that we can do a lot of, uh, nice things with. We’ve got different kinds of in, in cabin monitoring, uh, for driver and other passengers. So, I mean, there, there’s, with the introduction of now, uh, some of this LLM stuff in the cabin, you know, it’s, it’s, it’s definitely going to improve in the near term. Uh, you know, the, the question is, is, is the, some of the mindsets, are they so dug in that they’re not gonna change no matter how good this is? Or is it really actually the beeps are kind of annoying and we need to do better. Right. So there’s probably, uh, somewhere in the middle there.

Grayson Brulte: I think it’s it. It’s somewhere in the middle. And I think that if you look at, you know, an older individual, say north of 70. That their children could say, oh, you know, dad, mom, you do this, this, and this. And then, oh, okay. And I think it’s just getting over. Once you get past the annoying beep, beep, beep, beep, beep, beep. Oh, okay, this is cool. And so when you put your driver monitoring, or, or you put it, it’s a feature into their screen, oh, oh, okay, this is much better that that’s, you know, that blasted thing. Or you know, you know, they can get with no filter. That kind of goes away. And we do know that autonomy is gonna scale. L two is gonna scale, L four is gonna scale in your opinion. How do you see autonomy scaling across the globe? And do you see perhaps certain pockets in certain markets being primarily L two markets where other markets will be L four? What are your thoughts on the scaling? 

David Doria: I mean, it seems like there, there’s very different markets right now. At least I, I don’t, uh, you know, what the future will bring, but you know, no, nobody’s trying to deploy all four cars to the consumer on, uh, you know, highways and everywhere. Right. So they have these very, very, uh, you know, constrained ods that are per city, basically right now for any of the L four deployments. And there’s a very small number of cities. So, uh, you know, I, I don’t think we wait two years and then they just break down those, those, uh, you know, lines around the cities and it just magically works. That’s kind of like the L five Dream, right? Uh, that almost basically nobody even talks about anymore, uh, which may also come, but I think we’re talking about, uh, you know, multiple decades in that case. So, yeah. Right, right now it’s not, it’s, it’s, that hasn’t really been a, um. There’s, there’s not competition between L four and L two. It’s kind of pick a lane and, uh, you know, operate in your lane. You know, at some point there will be a convergence discussion. But I mean, I think that’s very far in the future. Right now they’re, they’re, the hardware is different, the algorithms are different. The approaches to the system are different. The regulation and testing is different. So it’s, it’s really different. I mean, Magna has had a kind of public, Hey, we’re not operating in L four. We’re gonna continue to watch this space very closely. Right at, at some point it’s gonna be the right time, but right now, uh. L two is where we’re operating.

Grayson Brulte: L2 is a growth market. L two is going to continue to scale. And if anybody has doubts, go look at the public filings of all the publicly traded automotive companies and just go on the EDgar database and just simply type in Command F, level two. And you will see that all these major auto companies. Our investing in L four is gonna scale and autonomy is gonna scale globally and as autonomy scales, Magna will be there and continue to be there. At the center of this. The future is bright, the future is autonomous. The future is Magna. David, thank you so much for coming on the road to autonomy today.

David Doria: Grayson, I appreciate it. It’s been fun.