driver-less cars are mostly fake news
Part 3: The Real News
Click to listen on Apple Podcasts or Overcast.
For a menu of other apps, click “Subscribe” above. You won't have to subscribe to listen. Or just listen via the player above.
An unedited transcript is below.
Ali Tabibian: Self-driving cars are mostly fake news.
Donald Trump: Fake news for a long time.
Ali Tabibian: But the real parts are big news. In this, part 3 of our 3-part series on self-driving cars are mostly fake news, we focus on the real news. Part 1 was about the big news and part 2 was about the fake news.
Robot Voice: Part 3.
Ali Tabibian: Welcome, welcome, welcome everyone [00:00:30] to Tech Cars Machines where we make you smarter about how technology advances everything from sensors, to wireless connectivity, to artificial intelligence is changing industry and transportation. We appreciate you being with us. Let's keep getting smarter. Here's some really good news: when money and sides, and sides is often referred to as form factor, are less of a problem, a lot of progress has been made. The size of the stuff you need and the cost of the stuff you need to get a car to behave [00:01:00] reasonably autonomously has gone from basically being unmanageable in a normal "vehicle" to sporadically workable and larger vehicles in commercial settings for certain uses. Now, that's multiple layers of caveats, but remember, it's a huge industry. You can have a lot of caveats and still wind up with a pretty big opportunity.
That's why our first interview in the next episode or two will be with Josh Switkes. He's chief executive and founder [00:01:30] of autonomous trucking company Peleton Technologies. A super, super episode, super, super guy that's being edited right now. The episode is being edited, not Josh. Now, other classes of activity that tend to be commercial but also have short predictable routes are shuttles, buses, and taxis, and the short and predictable matters a lot as we'll discuss in a minute. In the taxi space, Waymo the one that's ahead, GM's close behind, thanks to their acquisition of a Cruise. That [00:02:00] transaction has worked out really well, by the way, and the news generally about these two companies and their progress relative to others is accurate.
Waymo's implementation is in a van, GM in a bolt electric vehicle, I believe, and they're transporting passengers with no one behind the wheel but with a chaperon and the vehicle and they generally operate in better climates like Phoenix and Austin. GM was actually here in San Francisco a month ago showing off its taxis over one and a half or two-mile route. In the world [00:02:30] of shuttles, there's French company, Navya, that has about a 10% shuttle and they have several public road but off highway deployments, including in Las Vegas, and they're authorized now for chaperon but unpassengered travel in Paris. Speaking of Vegas, we visited them CES, and they're vehicle was a beautiful, beautiful vehicle and if you look at it, it requires 8 light hours, a handful of them which costs 4,000 each, 11 [00:03:00] cameras, and four radars. That's the cost and form factor thing that we were talking about.
In the Bay Area, a company called Auro, A-U-R-O.ai, if you're interested in the URL, is running autonomous golf carts that are transporting people, as we speak, in Santa Clara University, which is a city close to San Jose, for our international listeners. Michigan based, May Mobility has the same concept, slightly larger vehicles. I believe they're essentially in test mode right now. [00:03:30] The common thing here are slower speeds, good weather, and short routes which allow the autonomy systems to get lots of repeat training and have it happen in reasonable time.
So far, we've taught prototypes, we've taught limited autonomy features as expensive options in expensive cars, and I've talked a little bit about vehicles that can operate essentially fully autonomously, albeit in limited circumstances. Here are some good news and this is big news because it's becoming real news. Some of the cheaper components [00:04:00] that some of the people we mentioned with essentially autonomous vehicles, some of those components on their own can provide substantial benefits to the driver. In other words, you don't need to get to full autonomy to get to what's called advanced driver assistance systems, ADAS.
In early example as radar, you know the same radar that you've seen in World War II movies where the contraption on top of the warship is spinning around and looking for the enemy. In fact, that's what radar was refined for. Detection of enemy equipment, [00:04:30] mainly metal equipment, during World War II by the British. In the U.S., because radar works in the radio frequency part of the spectrum, of course, the government allocated part of its spectrum to vehicle radar in the early '90s and very soon, there were some systems introduced that were pretty good at detecting straight-ahead collisions.
A couple of people, and by that, I mean there are couple of manufacturers in the late '90s and probably most famously and consistently of all, Mercedes in its S-class introduced what was called and [00:05:00] it's still called the Distronic System. Now, the system holds not just a constant speed but also holds a constant distance to the car in front of you. These days, this type of system is called adaptive cruise control and, by the way, note that these systems don't manage the steering. They're just doing some of the pedal work, basically. When used for collision avoidance, by the way, what was strange was that the statistics show that these systems really didn't reduce accidents that much. If all, they were doing was issuing a warning. They [00:05:30] really made a difference when they're actually automatically intervened and, in that case, about a 20 to 30% reduction mainly limited to straight-ahead collision. Stuff coming from the side didn't really work that well.
As the cont and size of these radar components have declined and no, not on the digital cost decline curve, but in the wrong way they've made progress and the components have become smaller and they become cheaper to the point where they can now be installed quite broadly in vehicles and, in multiple instances, in some [00:06:00] vehicles, meaning that some vehicles have more than one radar installed in them. For example, typically, when you buy a blind spot assist system, there's at least one additional radar in addition to the look-forward one that's looking behind you for cars approaching into your blind spot.
But here's a surprising fact: consumers generally don't pay for optional safety equipment. When they're forced to decide due to mainly economic constraints between let's say safety, infotainment, luxury, they always choose infotainment and luxury. [00:06:30] These are the things that they use all the time, see all the time, and touch all the time, and sit on all the time. That's the way consumers behave and not just your average consumer, let's look at the high-end.
Let's use the Mercedes Distronic example again as well as, by the way, for its current offspring, Distronic Plus with Steering Assist, who comes up with these names, which, as the name implies, does both pedal work and some of the steering work. These options, Distronic and [00:07:00] Distronic with Steering Assist, are just not popular options. Go look at your favorite used car websites, search for these features in used vehicles and you won't find that many. They're just not that prevalent and the auto manufacturer is not really helping either. Here's a core for an upcoming interview we have for you from Josh Switkes, chief executive of autonomous trucking company, Peleton.
Josh Switkes: Typically, you go to a dealership, the demonstration vehicles they have don't have the features. So for years, you would have to go and [00:07:30] special order a vehicle based on a feature that was listed in an option book that you have never driven. Very few people adopt that.
Ali Tabibian: That's an expensive option.
Josh Switkes: And expensive, yeah.
Ali Tabibian: Here's another example. Lexus introduced self-parking as an option in Japan in 2003, 15 years ago. By the way, I remember in this country, they brought this equipped Lexus to an Oprah Winfrey show and they had her get in to test it and she pressed the buttons and whatnot and of [00:08:00] course, hilarity ensued. But not so hilariously, Lexus canceled this option for a number of years because there was no uptake here or in Japan.
Now interestingly, blind spot assist, which we just referred to, is the exception in terms of optional safety equipment with meaningful uptake. The cost is reasonable. It's about 500 bucks versus several thousand for autonomy or autopilot or whatever the manufacturer happens to call it and it's constantly interacting with the driver. It's not episodic. You're not waiting for it to save your life [00:08:30] once in a lifetime. You see the light all the time in your rear-view mirror. You hear from it frequently, sometimes nauseam. All of this stuff matters to consumer. When they decide to pay for something, they want to make sure they're actually going to use it.
What's interesting though is that consumers do compare standard safety features when they're making a purchase decision. The big news is that this aid-as-functions - collision avoidance, even advanced radar cruise control that we talked about - cost has come down to the point where these systems are becoming real news and [00:09:00] because they're becoming standard equipment, they're becoming big news. An example is Toyota Safety Sense Program, which in the North American market, these features are standard, more or less come across their product line, and get this, they're even included in the $18,000 [YARS 00:09:15].
Want some more good news? We've just got overwhelming amounts of good news here for you at this point in the episode. The type of high-tech that consumers value or pay for isn't necessarily limited to self-driving components and systems and all the fancy stuff. [00:09:30] Infotainment, which we just mentioned and has very good uptake, can be a really interesting driver of technology and feature adoption. For example, Tesla for years has had this systematic program to secure partners to install content on its very big screen. It's unnatural for them. Toyota has an extensive program with quite a few people focused exclusively on getting similar features, especially ones in the concepts of connectivity to their vehicles as well.
A quick [00:10:00] sidebar. The observation of anyone whose car has some limited autonomy steering functions is that it's much less harrowing to make adjustments to infotainment, air-conditioning, et cetera, when the car systems are first in line with steering input. Add to this the connectivity we talked about and the trend toward upsized infotainment screens and we think the market for features on paid content is underappreciated. By the way, look-forward cameras are a really interesting development and increasingly standard equipment on vehicles [00:10:30] as well and they're interesting enough where we'll talk about them separately in an upcoming episode.
Now, cameras, radars, all sorts of sensors in the car, these all mean lots of data to be processed and monetized. Just so we don't leave cameras out totally, the couple examples I'm going to give you about this data processing, why don't we consider ones that are mainly camera based? One is a company called Nauto. Let's see, the investors include, who's [00:11:00] who, BMW, Toyota, General Motors, and the other one is Carvi, which has actually been on the scene longer and probably has more deployments than a lot of its competition. In both cases, they used a windshield mounted forward-looking camera to provide driver assistance features.
In the case of Nauto, they also point a separate camera at the driver to measure distraction and other things. Of course, whenever there's data, everybody starts salivating. They want to collect it, monetize it, et cetera, et cetera, [00:11:30] especially both in the case of Nauto and Carvi where they can sell their systems efficiently through channels like fleets and insurance. Even in an aftermarket setting, there's a lot that could be done with the data that's collected.
Here are a couple notes of caution because we don't want to spend too many minutes with all good news. One, it's hard to get consumers to adopt aftermarket equipment, which is frequently what Nauto, and Carvi, and others are providing even when they're offered a financial incentive. The insurance companies have tried this [00:12:00] with aftermarket devices that they sent the consumers. We then plug it into their cars that allows some behaviors monitoring on the part of the insurance company in return for a discount on the insurance premium. Not too much long-term stickiness to these programs in terms of customer adoption. Two, again from the insurance industry, the data collected isn't clearly more predictive than current tools that are applied based on driver profiles.
ow, before you, our honorable listener, and I, your humble host, before [00:12:30] we both collapse in exhaustion however satisfying and edifying that exhaustion may be, let's go through a quick and compressed version of how autonomy might make it into production vehicles. We're going to make this very compressed but don't worry, we'll have some gruelingly long episodes on this subject soon. As we dive in, I want you to hold three concepts in mind. LiDAR, remember these are the laser scanners. Two, training of computers, not people. Remember the word training. [00:13:00] The phrase workforce in place. So, LiDAR, training, and workforce in place.
First, LiDAR. The conviction of the auto industry, except Tesla, is that LiDAR sensing is a necessary ingredient for autonomy. On self-driving prototypes, you can see these LiDARs as those big contraptions that are spinning on top of vehicles and more recently, you can see them not so much spinning but as a puck. They're shaped like those little yellow minions in Despicable Me, that great movie [00:13:30] with Steve Carell. You can think of these systems as a radar system, but instead of seeing hard objects and metal, they roughly see what the human eye sees and with the potential for much better resolution than radars, there's an enormous activity. Most of it, of course, is because it's so relevant to autonomy, but also because the cost and the form factor of these systems are fairly stubbornly high when you try to get some pretty good resolution out of them.
Velodyne, which we talked about in the context of the DARPA Grand Challenge, is the [00:14:00] leader by far, by far, in revenues and deployments and just industry stature. Now, big deal in the industry is whether it can convert these optics, these optical components into something that rely more on semiconductors and that allows them to take advantage of the digital cost and form factor reductions that the digital industry offers other industries whenever it penetrates them. That fundamental shift in the foundational technologies is what defines the rationale [00:14:30] behind a lot if not most of the investment in the LiDAR industry.
Quanergy has probably been around the longest and the next best funded after Velodyne in terms of seeking to create what's called a solid-state LiDAR. In other words, make it out of semiconductor stuff and get the benefits of cost and form factor. There are a whole host of other entities that are similarly inclined and funded, although generally to a lesser degree. We'll do a special episode on LiDAR. In terms of LiDAR, volume deployment on commercial vehicles is supposedly beginning in [00:15:00] 2020. But based on what we here, none of the suppliers can make this specs in time and so that deployment is unlikely to happen.
Now, word number two or concept number two: training. Remember, I asked you to remember this? All the data that these systems are collecting need to be used to train the driving software. Remember, we talked about this word training in the context of the shuttle section. The results of training is better and come along faster naturally if you can train the software over [00:15:30] and over again on the same route. Now, you can do this by driving the same route over, and over, and over again with somebody behind the wheel, making sure the car doesn't make a mistake, but that's hardly an industrial scalable process except if you're Tesla because Tesla is giving you all the autopilot hardware and keeping track of what you're doing even if you don't pay them the autopilot fee. They have an advantage that their system is being trained by a couple of hundred thousand vehicles that they've deployed.
[00:16:00] Now, this is an opportunity that other manufacturers don't have because their investors are really sensitive to those manufacturers' profitability like most investors and most of their investments whereas Teslas are obviously not the other auto companies just can't throw in a few thousand dollars' worth of equipment that they're not getting paid for just so they can collect some data. As an example, Ford, as it has announced, is aggressively trying to extract more profits from traditional vehicles so they can reinvest it on the self-driving stuff. [00:16:30] This is one reason that everybody hates Tesla, which will be the subject of one of our upcoming episodes in that they just have a whole lot more latitude than the other manufacturers.
Since not everybody has the latitude that Tesla does, one way of getting around the whole training process and having to do it physically out in the field is through simulations. A couple examples here are RightHook, which is a company which I believe just got about 18 million from venture capital company, Fontinalis. Fontinalis has some association [00:17:00] with the Ford family, more than the Ford Company really at this point. Furthermore, a couple of the larger companies that you wouldn't have really expected like Samsung and Apple are also rumored to be working on the same technologies.
For all of this stuff to mean anything, for all this data, these systems, the training to actually result in anything, you need that third item that I asked you to pay attention to and that's "workforce in place". It's a term I've imported from my world of mergers and acquisitions, which in turn imported [00:17:30] it from the world of accounting and basically, your employees, including their experience education training, are referred to as "workforce in place".
As an indication of how promising this self-driving area is but also how nascent it is despite what you keep hearing, all you need to do is notice that all the big deals in the space are for companies with smart people but not much in the way of working solutions or revenues. So, General Motors acquired Cruise, Ford acquired Argo out in [00:18:00] Boston, a [inaudible 00:18:03] respectively, by the way, with different structures for long-term incentivization, and then Delphi, a tier 1 supplier, acquired a company called NuTonomy for about $450 million. Generally, in terms of valuation and valuation metrics, by the way, actually, we're talking about in excess of $20 million per PHD for these acquisitions. $20 million of PHD is the going rate for AI "companies", but it's actually been exceeded by the [00:18:30] General Motors and Ford examples that I gave you. So, for self-driving cars to become real news, we all better hope that these PHDs are worth 20 or 30 million a pop.
Voice over: Isn't it great to let someone else do the thinking for you? Keep indulging yourself. Click subscribe. Subscribe with the little button in your podcast app or click the three dots in the little circle or visit us at techcarsmachines.com and gtkpartners. [00:19:00] com where our subscribe buttons are much bigger.
Ali Tabibian: Aha! I'm like the Energizer bunny, I just keep going. You're not going to keep me down that easily. Okay, here's a quick rundown on what you'll be hearing next in this podcast series. Interview with Josh Switkes of Peleton, that'll be next. Another episode will be about the weird looks of self-driving cars. We'll take apart one of these cars, figuratively speaking of course, and we'll tell you what the parts do, and what it means for pending breakthroughs and bottlenecks. [00:19:30] We'll cover artificial intelligence, which is common to cars and machines. We'll help you get behind the buzz words.
On another episode, we'll discuss what auto buyers actually value when they purchase an automobile. We'll do one on connectivity. How do you get the data in and out of these machines and once you're able to, what can you do? Everybody hates Tesla, which we referred to earlier. This typical Silicon Valley trope that legacy companies are in trouble could actually become true. We'll talk about regulation and get cute with the title and use the ...
Ronald Reagan: Nine [00:20:00] most terrifying words in the English language.
Ali Tabibian: Which are ...
Ronald Reagan: I'm from the government and I'm here to help.
Ali Tabibian: Thank you so much for listening. We hope you enjoyed it. It's a massive amount of work for us to put this out. We enjoy the process and our sincere hope is that you feel satisfied, feel smarter, and come back.