Katherine Forrest: Right? They get really meta. Yeah. Well, that is really, that's actually hysterical, I have to say, it's hysterical. But before we get to the framing of how these things actually work, I want to just say a couple of words about, like, what's really exciting from my perspective about, as you call them, AVs, autonomous vehicles.
Scott Caravello: Absolutely, then I get a shot.
Katherine Forrest: Okay, all right, even Steven. So, you know, here's the way I think about it. You know, when we really get to the point where we have self-driving cars that become, you know, ubiquitous and people are comfortable with them and they're driving them everywhere, there are going to be some really startling advantages. And, you know, I think about some older folks who really don't feel comfortable driving anymore and they may not feel comfortable driving at night. And they can't go to things that might be at their community center or at their local church that they otherwise might want to go to or wherever they want to go, go grocery shopping after, you know, after it's dark. And they're going to have all of this additional mobility to get places and interact with people and to do things that they otherwise might not be able to do. And then I think the same thing is true for people who have other kinds of mobility issues. And then the last thing that I think about is, you know, young folks. And I think that there are situations where you have young drivers and sometimes they make mistakes because they aren't really used to driving, and they can get them into, you know, some really unfortunate accidents. And then you get sort of the drunk drivers. And, you know, there's going to be, for a very long time and for important reasons, restrictions on using these autonomous vehicles when you've actually taken in alcohol because then you can't take over, right? You're supposed to be able to be attentive at all times and take over. But when we get to the point where the autonomous vehicles are really able to do the whole thing and you're not required as a matter of law to be able to take over, the ability of the autonomous vehicles to cut down on drunk driving accidents and all of the terrible things that can come from that, both for the drivers and for the people they may interact with—it's going to be a big, big deal. I mean, I'm very excited about it.
Scott Caravello: Yeah, and I know I wanted my shot, but you got all the great points there. The only thing I'd add is even more broadly, right? I mean, if you have just more efficient vehicle flow because of a majority of autonomous vehicles on the road, accidents generally should go down, you know, irrespective of whether someone's impaired. So that's a pretty exciting, exciting potential as well.
Katherine Forrest: It reminds me of when you asked me whether or not the car drove better than I did and I had to confess that I think it does. And that's with me fully like during the day, 10 a.m., the caffeine's at the right level, and you know, so imagine, imagine how it does compare to me at night. Anyway, let's go back and sort of set the stage by talking about, you know, how some of these autonomous vehicle capabilities are categorized, and you know, it's interesting because already there are a variety of international standards for autonomous vehicles, and there are five levels under these various standards under what they call the National Highway Transportation Safety Administration, which has actually adopted, I should say, the international standards. And there are levels zero and one, which are really sort of run-of-the-mill capabilities that are common in basically every car today, like various kinds of driver assistance systems and things like that—you know, things that sort of tell you if you're running over the line and might sort of, you know, have a steering wheel that sort of pulls you back a little bit. But then we go on to the more exciting pieces.
Scott Caravello: Right, right. So then level two, and that's Tesla's official classification, gets at partial automation, which is where the car can handle steering and speed control but does require a human to remain attentive and be ready to take over at any time. So even the full self-driving capabilities are considered level two automation, for now, at least. And that's because the Tesla level two is still being supervised with the driver's hands on the wheel and their eyes on the road.
Katherine Forrest: You know, and I actually had to look that up because when I was first thinking about level two, I thought, well, the self-driving capabilities that I experienced actually now, with some software modifications that occurred at some point—I don't know when it was, maybe over the summer—now allow you to take your hands off the wheel. And I was thinking, well, how is that level two? Because you've got to have your eyes on the road, but you can have your hands off the wheel. But still, it is level two. That is—that's how they characterize it. But, you know, then I thought, well, now I have to understand how it actually—I have assumed that there's a camera in the rearview mirror because if you have your eyes vary from being on the road to, say, looking down at your phone or something, the car will pretty quickly—the Tesla will pretty quickly beep you, tell you you've got to take control back, and if you don't, it will actually disable the self-driving capability. So there is, in fact, a camera. And that I find both affirming and a little, you know, in terms of—I was right. But it also, you know, it sort of raises the question of where does that data go? And you can actually, in your settings, you can have it deleted or you can have it sent, actually, to Tesla to utilize for training purposes. And so I actually have my settings set to have the data go automatically for all of the Tesla driving that I do to the manufacturing facility for training purposes—or whatever facility it is. Maybe it's like a computer facility. But anyway, finally, when we get to level three—All right. So after we go past level two, we're gonna get to level three, and that's gonna be when the car handles all the driving and the actual human who's supposed to be in the driver's seat will be able to disengage their attention, although they're supposed to be able to still take over if prompted. And then there is a level 4 and there's a level 5, which would get us to the point where the car could operate even without a driver. And Waymos, for example, I think, are at level 4. Is that right?
Scott Caravello: Yeah, exactly.
Katherine Forrest: Yeah, okay. And while level 5s haven't made it here yet, they've been publicly announced, but we're not quite there yet.
Scott Caravello: Right, and so how are these cars actually doing it at these different levels of automation? Well, Tesla and Waymo both take different approaches, different camps. And so Waymos rely on data collected via LiDAR, radar, cameras, and microphones. The cars then use a series of large language models and vision-language models that handle everything from perception to prediction to control and then the ultimate decision making of the car.
Katherine Forrest: And what's LiDAR? Like, I think we should sort of just spend like two seconds on that.
Scott Caravello: Yeah, so LiDAR is basically an imaging technique that uses lasers to determine the ranges and locations of different objects. So it's more advanced than just, you know, a pure radar system that, we know, we all think of in movies that have the line going around the circle with the dot that shows the distance of some sort of object. That's probably an enemy vehicle.
Katherine Forrest: Right, you know, and LiDAR is also used in virtual reality systems all the time. So a lot of folks don't even realize how ubiquitous LiDAR could be. So, Teslas operate in a different way, as you were saying. They operate with vision. And so it's actually interesting to watch a Tesla approach sort of a stop sign where the turn is beyond the stop sign, which is most stop signs. Anyway, so the Tesla will stop at a stop sign, and it then inches forward so that its cameras can look in the various directions that they need to look and then is able to make the appropriate next move, whether it's a turn or going straight forward or whatever. But they do that by using eight to nine cameras that are on the car, and they then use a video transformer model to operate it. And so the Tesla and the Waymo camps claim different advantages, as you would, if you're a competitive, autonomous vehicle. And the Tesla camera approach is supposed to be quicker and more cost-effective to scale up because you can just collect the data from Teslas all over the road and use it to further train the AI model that's behind the whole thing. On the other hand, Waymos are only commercially operating in areas that have been mapped in advance and that are using LiDAR sensors so that they can have some significant upfront investment costs in a given area, but then operate seamlessly once they're there.
Scott Caravello: Yeah, yeah, but that mapping dependency is a big criticism of Waymo's approach, but they've pushed back on that because maps and roads are changing all the time, right? Even just due to a construction zone. So if you're purely relying on the fact that you have a road mapped out, it wouldn't actually be that useful. But so, according to Waymo and their co-CEO, that pre-mapping is really not a significant factor in how Waymo is expanding into a given location. But as the supposed advantages of the Waymo approach—using those other sensors beyond cameras, like the LiDAR and the radar—the cars arguably allow for better driving in variable conditions like fog or bad weather, where a camera, just like any human driver, would have a tougher time seeing the road and what's on it.
Katherine Forrest: Yeah, you know, it's interesting because there are different safety arguments on each side. But I think if you put it together, when we get to the point where these autonomous vehicles are really sophisticated, it's going to be safer than, I think, frankly, humans. The Teslas say that they're concerned about whether or not you can have different sensors disagree about what the car should do. And what they're talking about is like the Waymo sensors. And they think that there's a possibility that you could have LiDAR thinking there's a pedestrian when there's not. But what if the camera also doesn't see anybody? Although, by the way, my Tesla yesterday, because I do use it now and again, I just want to tell you, there was a big, like, oil slick in the road, and it thought that it was something. I guess it maybe just didn't even know what it was. And so it sort of—there was no oncoming traffic, and so it made a very safe move around it. But I was actually sort of impressed. But there's a whole field, anyway, of engineering that's called sensor fusion, dedicated to situations where you're trying to figure out what's happening where with the different kinds of sensory inputs that these technologies are receiving. And it's really pretty complicated. And so there's going to be some hurdles to get these cars sort of on the road all over the place and ubiquitous, as I said. But there's going to also be, on top of that, the regulatory overlay. Right now in the U.S., it's really at a state level. And so we have sort of state-by-state rules as to whether you can drive your autonomous vehicle with full self-driving on, where, and there's just a patchwork of different kinds of obligations in various states. Some allow it to be on the highway, some allow it to be on private roads but not public roads. And so you've really actually got to pay attention to where you're driving before you can actually switch it on. And I am, of course, fully compliant with all state laws. In fact, I carry a booklet of state laws with me while I'm driving. And, if I want to—no, I’m joking.
Scott Caravello: Okay, I was about to ask if this is like self-compiled or if it's, like, purchased on Amazon or—okay.
Katherine Forrest: I don't even think that there is such a thing. But anyway, you know, there you are, you cross a state line and you're like, hold on, pull over. All right. Anyway, so, you know, the point is that there are lots of regulatory laws. Federal safety laws also come into play. And so, you know, we're gonna have to figure out all of this to sort of smooth it out and get—it's like railroad tracks that used to be at different gauges before they finally decided they wanted to be able to have trains go across the entire country. And to do that, they needed to sort of smooth out some of the gauges between tracks. We're gonna have to find some way to actually smooth out some of the state laws between how these vehicles operate.
Scott Caravello: All right, so I know we're probably running a bit short on time, Katherine, so we can't get into the full details of all the different laws and regulations.
Katherine Forrest: No, I got my book here. I've got my book.
Scott Caravello: Well, this will be the sequel to Of Another Mind. Your compilation…
Katherine Forrest: No, no, no—my book of state laws, my book of all of the state laws.
Scott Caravello: But I did want to just mention one sort of fun issue that’s popped up right—so, where there's been a whole issue about who to ticket when there's no driver. So Waymos, for example, will respond to police sirens and pull over, but there have been some anecdotes about police officers not being able to actually cite the cars because there's no robot box check on their form. And so California, where this—this problem was manifesting, Governor Newsom did sign a law that would allow for notices to be sent to the AV companies, though there are no penalties specified for these notices and the law doesn't take effect until next year. But it's just showing how these little issues are going to pop up as AVs get more common. And we sort of figure out how to adjust decades of traffic laws to their presence on the road.
Katherine Forrest: Wait, I want to know why the Waymo is getting cited in the first place. Okay, so we're supposed to, after all of my sort of statements about how this is going to be better than human drivers, etc., etc., etc., do you have any idea, like, what the Waymo problem was while they were being cited and being pulled over?
Scott Caravello: I, to be totally honest—I don't. And, I think that—I think that maybe we can—we can follow up on another episode…
Katherine Forrest: Do another episode, right?
Scott Caravello: On this exact issue—yeah, yeah.
Katherine Forrest: We'll do it on my little handy book of state laws that I always use whenever I'm driving my fully—
Scott Caravello: Exactly, exactly.
Katherine Forrest: And then we'll also talk about why in the world Waymo is being pulled over at all. Or a Tesla, by the way, because Tesla has these different settings that you can put your self-driving capabilities on. You can do chill, standard, or hurry. And hurry will actually go over the speed limit. And so I suppose you could definitely be cited for that. I don't actually like the hurry because I always drive at the speed limit, and so when something is driving faster than the speed limit, it makes me nervous. And so, I do not like that. All right, well, we're gonna be coming back to this because there are a couple of things that are happening in the regulatory space around this. You know, in April, the transportation secretary, Duffy, unveiled an automated vehicle framework that had a couple of different principles: prioritizing the safety of ongoing AV operations on public roads, unleashing innovation by removing unnecessary regulatory barriers, and enabling commercial deployment of AVs to enhance safety and mobility for the American public, which means essentially, so to speak, paving the way—no pun intended—paving the way for trucks to actually have nationwide access to the highways. So, you know, we don't really have a full set of regulations, but we've got some things that are coming down the road, so to speak. I was like, paving and coming down the road all at the same time. Yeah.
Scott Caravello: No, it's great, it's great. I would have been disappointed if we got out of this episode without a few puns.
Katherine Forrest: Alright, okay. Well, that's all we've got time for today. We're going to come back to this topic because it's really a hot topic and we've only just begun to scratch the surface. But thanks for listening. I'm Katherine Forrest.
Scott Caravello: And I’m Scott Caravello. Don’t forget to like and subscribe.
