Lux Capital - Shahin Farshchi #2

Welcome to the Autonomous Vehicles podcast. Today's guest is Shaheen Farshji, partner at Lux Capital. In this episode, we discuss Amazon's acquisition of Zoox, how the AV market has changed over the last three years, and the ripple effects R and D associated with avs will have on the broader market. Let's dive in. Hi, Shaheen. Welcome to the podcast. Or should I say, welcome back to the podcast.

Great to be back, Ritesh.

For those who haven't heard the first episode, do you mind just giving a quick reintroduction around how you got interested in autonomous vehicles and a little bit about yourself?

So I have been a huge car enthusiast throughout my entire life. It probably started when I was five, watching Knight Rider. And the jury is still out as to whether Knight Rider was actually a kids show. I think it probably was because it got me really excited about cars, it got me really excited about artificial intelligence, and had a huge role in how I got to where I am today, investing in companies that are pushing the frontiers of AI and mobility. So I grew up in the Bay Area, went to Berkeley, undergrad, did electrical engineering, computer science, graduated in the dot com bust, went to General Motors in Detroit, oddly enough, in pursuit of greener pastures. I can tell you how bad it was in the Bay Area at that time, and spent a little bit of time there, realized that I wasn't a big company guy. So I decided to come back to California, did a PhD in electrical engineering in hopes of starting my own company, which I did. And I quickly learned that it's one thing to have a really cool technology, another to have a real business. And I had the former and not the latter. So I decided to get to know a lot of smart investors to help kind of guide my thinking for my next company. And I got to know Lux and they asked me to join and I thought I'd try VC out. And that was back probably 15 years ago. And here I am 15 years later, funding companies across the spectrum, from software and infrastructure to rockets and satellites and driverless cars. And it's a pleasure to be back today.

Yeah, and we'll, we'll dive deep into that. I mean, the episode we last recorded almost three years ago, crazy fun fact. Cheyenne was episode number one of the podcast. It was a very different world back in, let's call it December 2018, January 2019. Much has changed and would love, you know, from your perspective, how do you kind of like, in a nutshell, like, how have things changed? What have been some of the things that have been most interesting for you over the last three years or so in the, in the AV space.

So three years ago the broader automotive establishments became aware and became very active in investing in both electrifying cars and making them autonomous. And we're starting to see the fruits of that today. And we've seen the flurry of M and A and investments that have taken place since then. There's also been a realization of, on the startup side amongst the Zooxes, Waymos, Auroras and cruises of the world, that the autonomous level 4 problem is probably more difficult than most people expected. And it's not the base case that is difficult, it is the corner cases that are almost impossible to predict and build a solution for. So a lot of great progress has been made and we continue to see further progress likely going to be made in the future. And a lot of that focus is going to be on those corner cases to make driverless cars safe before they can be introduced to the broader kind of market.

Yeah, I'm looking forward to diving into that because I remember in our discussion we were kind of projecting out in terms of how things would play out and, and quite frankly a lot of your intuitions and hypotheses around how things will play out came true, whether they were from individual kind of companies or more broad approaches within the market. And everybody over the last few years has been looking at this ever elusive level 4 fleet of autonomous vehicles that I'm going to be able to get into in the cities and here in San Francisco. I, I see Zuks and Waymo and Cruz and, and everybody driving around all the time. But at least to date I haven't been in one in terms of taking my first robo taxi ride from there. I, I'd love to just, you know, spend a little bit of time on Zoox in particular. You know, you were an early investor in Zoox. You, you've known the team there for quite some time in the time since our last recording in this Zoox, you know, had been acquired. Can you share a little bit about that? Just around what that experience was like, some of the broader kind of ideas that you have around the partnership that Zoox and Amazon have.

So the partnership is a fantastic one and I am very happy that Zoox was able to find that long term partner that would assist in taking it to where it needed to go, which was obviously a very, very expensive path and there probably aren't that many partners out there, potential partners out there as committed and as deep pocketed as Amazon. So that partnership was an incredibly positive one and we're very excited about the prospects and what's likely to come out of that partnership. If you zoom out and look at level four as a whole, I like to look at these past, say seven or eight years of investment and talent that has gone into Level four, like the space race. If you look at the space race back in the 60s and 70s, there was a great deal of extremely high quality talent that went into solving these problems and ultimately many industries benefited. And I expect the same to happen here with Level four. Not to say that level four will never happen. We ultimately did go to the moon in the space race and we ultimately will have fully autonomous vehicles. It might take a little bit longer than people expect and cost a little bit more than what people expected. But in the meantime there's going to be outcomes that are going to have a direct impact on other industries. For example, there still is a lot of automation that, that can go into conventional vehicles today. If you look at conventional vehicles, they probably have the same degree of automation, or at least most of the ones that are coming off of assembly lines worldwide. There are some high end models that have some features here in the United States, but if you look broadly across the world, most of these vehicles don't have much more automation that existed probably 30 years ago in the form of cruise control or anti lock braking systems. And so it's my expectation that a lot of the technology that's being developed for level four, which is low cost sensors, high quality software, these will likely make their way into conventional cars. So every single vehicle that's built is going to have anti lock braking, adaptive cruise control, emergency stopping, and perhaps even the higher end vehicles, the kind of mid level luxury vehicles will have, let's say highway autonomous driving, where the driver obviously still has to pay attention. And we may even see slowly in the next few years, point to point level two, or what I like to call level two and a half autonomy, where the driver can program their destination or load their destination through their phones into their vehicles and the vehicle will drive itself to that destination while the driver has to constantly pay attention and take over when necessary. So the driver is still effectively overseeing the vehicle, but the cognitive load associated with doing the driving will be greatly reduced. And hopefully over time, the amount of attention that driver has to put on the vehicle will slowly be reduced to the point where hopefully in the future the cars will be taking over proportions of the ride until the cars take over for the entire ride, which is that level four that we expect. So the level four vision has been Pushed out. But there's a lot of features that will be produced as a result of the work that's gone into level four, that will go into level two, that will hopefully bridge that gap in the next 10 years.

There's a ton there. I'm excited to unpack from the ripple effects that investment in Navy technology will have across other industries. But focusing in on those points you're making in terms of passenger vehicles today, back In December of 2018, when we were discussing this, you very astutely shared that the application of autonomous vehicles you saw to be much more applicable in fleets of service vehicles that are purpose built for, let's call it moving goods or moving people, whatever it may be, and that the. Let's call it level four, Level five autonomous vehicle, it makes less sense as a passenger vehicle that somebody owns and it sits in their garage and then they summon it to wherever it is. So within now, I mean, over the last, I'd say specifically two years, I think there's been a significant focus put now on these ADAs, these driver assistance systems within all vehicles. You're seeing radar and solid state lidar going into vehicles across the board and yeah, like more and more level 2, level 2 plus autonomy going into these vehicles to reduce that cognitive load. I really like the way you put that. Do you still view that the pathway to commercialization in these fleet models to kind of still largely stay intact to that original vision you had where, you know, that's. Well, at least when I see the fleet of Waymo cars outside, that's what it feels like they're aiming for. Is this robo taxi or with Zoox, for Zoox and Amazon Partnership, a fleet of delivery vehicles, for example, does that kind of stay intact, that vision of the fleet application for autonomy, or has that changed at all?

Absolutely. So from a sequencing perspective, I do expect all Level 4 vehicles to be operated in fleets in the early introduction of the technology. I do not expect for the foreseeable future for someone to be able to walk into a dealership and purchase a vehicle that has no steering wheel, that will never be driven by the owner or the operator. So we will have those level four vehicles that are going to be offered in showrooms that don't have steering wheels. But that's going to happen in the very, very distant future. What we will have in the near future is level four technology when it's ready to be applied in fleets. To your point, delivery vehicles, but perhaps even earlier on for long haul trucking, I think it's been pretty much accepted amongst the community that we will see the first incarnations of level four, where the driver can fully disengage and perhaps not even be in the driver's seat. In long haul trucking, where the truck is going in a straight line in relatively good visibility under controlled conditions, meaning that, you know, when and where the technology is used is going to be managed by the operator of the fleet, who are going to be obviously trained professionals as opposed to consumers. And eventually the technology will mature from there into the delivery vans and the fleet vehicles, and again over the very long term into vehicles that can be purchased at retail. And the application obviously is very apparent for long haul trucking because one major constraint, or maybe I should put it this way, the biggest costs associated with long haul trucking are the drivers and the fuel. And there's a limitation as to how long a driver can drive for any given stretch. And so if you can relieve the driver over certain periods of time so the driver can rest and recover, or for the driver to perhaps even leave the vehicle and the vehicle can drive up the road, you know, by itself for another, let's say 10, 15 miles to pick up a new driver, that will greatly reduce the burden and improve the economics of long haul trucking. So again, early applications of level 4, full autonomous long haul trucking, and it would eventually penetrate into the less structured environment still managed under fleets, Amazon, FedEx, UPS delivery, and then after that, perhaps robo taxi fleets that are managed by the likes of Uber and Lyft. And then the long, long term, decades from now, consumer products, consumer vehicles.

Thanks for breaking that down. Interesting that you say, you know, decades from now, I think if even like five years ago, if that kind of timeline was even thrown out there, it would have been from my opinion, kind of viewed as like, probably pretty egregious by some folks and being like, oh no, it's going to be sooner than that. And I think if anything, over the last few years, as we've seen the timelines for this level four autonomy being pushed further and further, and perhaps a more widespread appreciation of the complexities of those corner cases that you mentioned around level four autonomy, it does feel more realistic on that end. And then the question starts to beg, I mean, will passenger, let's call it like the passenger vehicle, the vehicle that sits in my garage, will that ever be autonomous or ever have a need to be autonomous, or have we completely moved as a society to mobility as a service by then?

I do not expect the notion of owning a vehicle to be made entirely Obsolete. At least not in America, where we live in suburbs and things are somewhat distant from one another. I do see a future where there are urban zones where passenger vehicles, conventional passenger vehicles aren't allowed to enter. So your only option is to use public transportation or to use a robo taxi to get to your destination. And then outside of these zones, you would have to, you would be allowed to operate your own personal vehicle. And then the question is, okay, when does that personal vehicle become fully automated? That will be in the distant future. I feel like as if the. Going back to the earlier conversation, the drivers for full autonomy are far stronger in these fleet applications. I was mentioning, for example, I was talking about the economics of long haul trucking. If you look at the economics of ride sharing, again, it is very much driven by the driver, so there is more motivation to automate that. And you're already operating in the construct of a fleet environment. So there are technologies that can be brought to bear. For example, if you're operating a fleet of delivery vans that are autonomous, you can also have a team of tele operators to operate these vans remotely. In the outside chance that these vans get stuck, for example, or need help from a human that is far more difficult to implement for a consumer product. A fleet that is a business that generates revenue from operating these vans or these trucks will be far more amenable to paying the price of, of having that human assistance, that teleoperation assistance, than someone that's going down the street to pick up groceries or pick up their kid, for example, from school. And so I think autonomy will become widespread, but I think it's going to happen a lot sooner with the help of these other resources, for example, like teleoperation for the trucking, for the fleets, for delivering carriers, and then way down the line for consumer.

Thank you for breaking that down and completely agree in terms of that sequencing and that timeline. And really from one makes kind of like the logical sense in terms of application. And two, what makes the economic and financial sense for businesses that are investing in this technology. Totally makes sense there. Wanted to change gears a little bit and go back to something that you alluded to earlier, which was a parallel between the investment in avs and the space race and how the R and D that went into space technology had ripple effects across many other sectors, industries, et cetera, and how that technology really led to quite a lot of societal change even outside of the space domain. I see something similar happening with AVs. And you know, from your vantage point where you're seeing so many companies every day. I'm really curious, I mean, what are some of the ripple effects that you see that are the, the natural kind of continuation of this technology? Some of the areas that have been particularly interesting to me is the increased effectiveness of computer vision, the improvements in simulation, the notion of training artificial intelligence through simulation or through synthetic data versus needing to go out and collect real world data. I'm sure there's many, many more. But you know, from your vantage point, like what are some of the areas, the sectors, markets, anything more broadly that you see either benefiting or just materially being impacted by the R and D that's currently going into autonomous vehicles and has been for many years.

So obviously these autonomous car companies have raised a significant amount of capital. So there was actually an article that showed how much companies like Lordstown and Rivian and Nio had raised since 2019, and each of them were in the billions. And if you think about it, that's a pretty significant number relative to other fields for pure R and D. And so you said it, there are advances in sensor technology that is being geared towards solving this autonomy problem that can be applied to other areas. So for example, ava, which is one of our companies that is focused on sensors perception for autonomous cars, recently announced a partnership where it would be using this technology in manufacturing. And so that was a unintended byproduct of a technology that was developed for autonomous cars. You also talked about computer vision. There are huge advances of computer vision that are being created at these companies that can be applied to manufacturing, to logistics, to interpreting medical images, to security, and to search, for example, in software. So there are many, many other use cases that will come out of this. If you look at a sensor calibration that goes into autonomous cars, you have hundreds of different sensors ranging from cameras to radar, to lidar, to temperature sensors, short range lid, short range radar, like all of these have to be calibrated relative to one another. There's a future where devices, robots, machines will also have a very large number of sensors. And the know how that's created in sensor calibration for autonomous cars can be applied directly to those you mentioned simulation, the use of synthetic data where investors and applied intuition and they also started out building models and simulators and synthetic data for training neural networks and autonomous cars. And they've gotten a bunch of inbound interest from companies outside of transportation that are interested in finding ways to simulate and build data sets for training their models. So, and the list goes on. So I believe that there's going to be and there already, by the way, has been a handful of companies spun out of places like Waymo and Aurora and Zoox that are using the tools that were created at these companies to go out and solve problems for the broader community. One company is called Streamlit, for example, that's building tools for building neural networks and training these models. And so we're starting to see that today. And I do expect there to be many super successful companies that create a huge amount of impact in society, having their roots in solving the problems for level four.

Yeah, it'll be fascinating to see kind of how this plays out from here for all the reasons you mentioned there. I mean, how has your perspective of investing in, let's call it like AVS or associated autonomy technology changed, if at all, over the last few years? Is it still a space that you're looking at pretty closely? Is it a space that you kind of view to be fairly saturated and. And players being quite far along? How has anything changed?

Yeah, so there was a time probably four or five years ago, or yeah, I would say four to seven years ago when there was significant investor interest in companies that are building infrastructure for AI. So when I say infrastructure, I mean core capabilities. And I feel like the emphasis has gone from kind of like core science and core capabilities to real businesses. And so there was a time when talent was relatively scarce and being able to create a computer vision model or natural language model or a speech recognition model was something that was novel and special. But now there is more emphasis on who are you enabling, how are you enabling them, and how is this the basis of an interesting business? So, for example, we're invested in a stealth company right now that is in fact building infrastructure for AI and ML. But there's a very, very clear business case, there's a very clear economic case as to why they exist and why this is an interesting business as opposed to just focusing on just pure core competency. We saw a lot of M and A take place over the past half decade and that was primarily driven from large companies wanting to tap into that talent base. But I feel like a lot of that talent right now already resides at these larger companies. And so for smaller companies, the emphasis is more on, hey, how do we build a real business to monetize these AI based innovations that we're bringing to bear. That's how the thinking has changed and that's obviously as a result of the technology becoming more mature over the past decade as well.

Thanks for sharing that. And to build upon it, if we look back, let's go half decade like you mentioned. So from there, just the notion of a robo taxi business that did not have a driver seem to have sufficient economic incentive or economic opportunity that it seemed to be quite compelling for, let's say like okay, you take Uber, you take the, the existing Uber business model and then you take away the driver and, and now you have a ton of margin and with what feels like a bit of sobering over the last few years or like, okay, like it may actually take quite a long time until there is no driver, line item or safety, safety driver on the books to make this a viable business model. Do you still view kind of the like robo taxi economics to be interesting or on the long haul trucking side? I mean we interviewed Locomation a while back and that's the long haul trucking platooning example as you mentioned, where you can get perhaps 3x the efficiency out of your, out of your trucks and the implications that that has on the cost of moving goods is very clear. Do you still see kind of the, the wide economic incentives for this robotaxi type model or is that something that quite frankly is become, you've become a little bit less interested in. Maybe it stayed the same or maybe you're, you don't really have a strong opinion on it. I'm just kind of curious how your, how your thoughts are.

Yeah. So Guru Tash, going back to the response from the previous question, Zoox was one of those companies, perhaps even an early company, that was combining innovations in AI and robotics with a sound business opportunity or economic model. It may have been a simple one, but it was a real one and that's what enabled them to go out and raise a billion dollars. At that time, most AI companies were, were saying that hey, we're very smart AI people. We're going to build a company that's going to become valuable. And Zoox was one of the early ones that said, hey, here is the economic benefit that we plan to generate as a result of solving this AI problem. Now the thing that happened with Zoox was that it took a lot longer and ended up costing a lot more than what people originally predicted. So now if the question is, okay, well, if an economic model is presented to you, or perhaps more specifically if a robotaxi model is presented to you, would you be supportive of it? And the answer is yes and no, I would not be interested unless there is a fundamental asset or edge that is being leveraged to be able to compete against these other players. Zoox, Aurora, Waymo Cruz, you know Argo and others who have already invested billions and billions of dollars. And so if the pitch is we have been able to solve this problem that nobody else has been able to solve, and we'll get to that, and that's going to help us get to this 99.999% effectiveness or safety, then that is something that I would consider in lieu of that. It would be very difficult to try to take a bunch of other smart guys and say that, okay, like they're going to do it better than these folks that have been added for the past seven, eight years and already have access to billions of dollars. Whether that's going by going through, by through going public or having folks like GM and Amazon backstopping them, it makes perfect sense.

One other thing I'm curious, kind of just how you're thinking about is the data layer behind all of this. So at least you know, that's been something that's been on, on my radar for quite some time where if you have the, all these vehicles, they have, they're collecting tons of information about their surroundings, information about their riders. They could perhaps even like serve up suggestions or one of the ways that was always brought up is that you can have ads in the cars or you know, take somebody to get a hamburger, you know, whatever, whatever it may be. Have you seen any kind of compelling uses for the data behind this? Or is that something that hasn't really been a focus area? And I can kind of tee up with one area that I've been really interested in exploring that seems a bit nascent right now, and that is you have all of these providers. So you have the waymos like execs, et cetera. You listed a bunch of folks, you have all these providers who are now operating these machines around this called geofenced areas. And they're collecting tons of data on their surroundings. Maybe they have some data around road closures, maybe they have other data around. Let's just use like public Works, for example. Is there any kind of existing infrastructure or pathway to infrastructure for this data to be shared and used and, and to be able to provide some sort of either economic or societal benefit? Or is that all kind of just sitting in like a siloed server somewhere? Kind of like almost like a wine bottle that like needs to be corked one day and like there one day, like this data will be useful. Curious on like the data angle if you've seen anything?

Yeah, so there's obviously a huge use case here which is training the models. So these cars are driving around. And there is dogs, there's fire hydrants, there is toddlers, there's. There's adults, there is furniture that's being moved around. And so it's my expectation that this data is going to be used to better train these models and future models, obviously. But then the question is, is there a commercial purpose for this data? The first question that comes in my mind is, who owns the assets that are captured in this data? If you take privacy into consideration. So, for example, if somebody's by the road, you know, having a McDonald's sandwich or is kind of like, you know, whatever, like looking for a place to grab a bite, can the AI determine, okay, this person is hungry, therefore make a recommendation to KFC that KFC should have a location over there, perhaps. But it's worthwhile to keep in mind that there are already many sources of data out there, more so than what we would think. And so rather than have the tail wagged the dog, which is, okay, we have all this data, let's find something to do with it. I would perhaps pose the question to folks that would be consumers of the data, which is, hey, what kind of signals are you looking for that could potentially be captured by cars driving around the street? One example that was very informative to me was the use of geospatial data for consumer behavior. So there was a time, probably in the early 2010s, when financial institutions were very excited about the notion of being able to count cars in parking lots to be able to figure out whether or not Home Depot, for example, is going to beat its earnings expectations. And then credit card data became available, and that credit card data just completely obsolesced the notion of counting cars and parking lots, I think this data would be valuable. And before a company tries to go out and gather all that data and find a place to put it and annotate it, I would probably figure out what problems exist out there, what customers you want to go after and engage with those customers and try to build a useful product for them or have that in mind when you're building stuff like this.

So we talked a little bit around your position on AV investing and kind of the associated technologies there. Curious, more broadly, what interested you today in terms of technologies? What are you most drawn towards? What's exciting? Where do you find most of your energy flowing these days?

I'm very much drawn towards exceptional people. We're in a very exciting time right now where amazing talent is not only being cultivated in traditional institutions like major universities and companies like Facebook, Google and Apple and Amazon, but also small companies outside of educational institutions, bringing with them very different backgrounds and perspectives that could be looking in areas where these giants probably aren't and being able to attract a DNA of engineer and broader team that these larger companies and institutions are not. So I think this diversity will lead to great things and I am very much on the prowl to find these people and get behind them.

Yeah, it's really amazing to see. I mean, it's almost like the promise of the Internet is upon us where you have this democratization of information and you have so many people connecting and sharing ideas and the natural byproduct of that has been, yes, so much innovation at an accelerating pace. Very exciting to see where this all goes. Thank you so much, Shaheen, for taking time out of your very busy schedule. I know. For those who want to learn more about you or Lux, where can they find you?

They can find me on Twitter, they can find me on LinkedIn, and they can find me through friends like yourself.

You got it. There it is. Well, thanks again, Shaheen. Really appreciate it.

Love to Gurtej. Thank you.

If you enjoyed this episode, it'd mean a lot if you could leave a rating and review to help others find the podcast. If you'd like to connect or have any suggestions for future episodes, you can reach me on LinkedIn or send me an email@gautonomous vehicles podcast.com thanks. And until next time.