The State of Autonomous Vehicles

myself in that group. I remember seeing the remarkable progress that had come out from the first DARPA Grand Challenge up to about I don't know ten or so, and and it seemed like we were just on the verge of having fleets of robo taxis at our back and call. But now we've gone on for several more years and we're still at a point are only a handful of companies are conducting limited tests. Plus there have been some high profile cases of accidents involving vehicles operating under autonomous

or semi autonomous modes that ended in tragedy. So in this episode, we're going to take a look at autonomous cars and where we stand today. Now, let's start with I think it helps if we run through the levels of autonomy, and not everyone uses these levels to talk about autonomy, and to be honest, the barriers between levels are a bit fuzzy, and sometimes we're not really able to say where we are at as far as levels

of autonomy. We can look back at previous developments and say, all right, well, judging on where we are now, we'd say that this falls into level two or level three, But it can be a little difficult to see what level we are currently in without you know, truly remarkable evidence. But in general, this is a useful way to talk about how far along we are uh as far as getting fully autonomous cars. So technically the levels range from

zero to five, so that means there are really six levels. However, level zero really means there is no autonomy at all. So with that type of vehicle, the human driver is responsible for all operations of the vehicle. Every driving task is handled by the driver alone. So some folks will say that they're really just five levels of autonomy. Zero would refer to vehicles that really, honestly, they don't exist that much anymore. Now you might be thinking but hey,

Mr smarty Pants podcast person. I drive a car and it doesn't have any autonomous vehicle features, but depending upon whom you ask, features like power steering or anti lock breaks or cruise control and other pretty common features fall into the low level autonomous range. It doesn't mean your car is at on this, but it has some of the components that are identified with this concept of autonomy.

So most cars today are actually above level zero if we go by that definition there level one are higher. So level one autonomy would apply to cars where the driver still controls the vehicle. The vehicle is still under driver control, but the car has some driver assistance features like power steering or antilock brakes. The car might have what is called an advanced driver Assistance system or a d a S, and the word advanced makes it sound a bit fancier than it is at this particular level

of autonomy. The car might have systems that help people steer, or it might have systems that help accelerate and or break, but the steering and accelerating or steering and breaking can't happen simultaneously. Either one or the other can be taken over by these systems, but not both. At the same time, not with level one. If we get up to level

to autonomy, then we're talking about partial automation. The A d A s on these cars can do stuff like control steering and breaking, or steering and accelerating at the same time, at least under certain circumstances. But even in those cases, the car's driver still remains primarily in control of the vehicle. With this level of autonomy, a driver would still not remove their hands from the wheel, as the car would need the humans participation to you know,

works safely. So with level to autonomy, you still have to have your attention on the road, you still have to have your hands on the wheel. It's just that the car can occasionally kick in and assist in various scenarios, typically in very restricted cases. Now at level three autonomy, we're getting up to conditional automation. These cars would still require a human driver, but there can be times when the car systems can operate the vehicle on its own

and the driver is essentially a passenger. During those moments. The driver is still supposed to monitor the environment. They're still supposed to be prepared to take over the car should the vehicle indicate it needs to hand over control to the person behind the wheel. So ideally there would be a system where the car would identify a situation in which the driver needs to take over, and then well in advance of that situation becoming imminent, it would

alert the driver to take over control of the car. Uh. This is trickier, right. This is harder to do than it than to say, because the car would have to know far enough in advance to be able to send that alert to the driver, and the driver would have to be able to respond to that. And while we feel like our response time is really fast, in computational terms, we are snails. We move super slow. So this is actually pretty tricky, especially if you're talking about a dynamic

situation where things are changing very rapidly. At level three, autonomous cars are supposed to uh do this seamlessly, and as I said, that is a pretty tricky thing to do technically. Most vehicle systems were looking at now, especially the ones like Tesla's autopilot, falls somewhere in level three.

Level four autonomy is at a point where a vehicle can automatically operate itself at least under certain conditions, but not necessarily all driving conditions, the vehicle would likely include the option for a human driver to take over operations, but under normal, you know, conditions that the car would

pretty much drive itself. So with level for autonomy, you would have self driving cars that could act as a self driving car for most of the time, but also allow a human driver to take over if the human driver wanted to. UM. Level five autonomy is a fully autonomous car. The car can operate itself under all driving conditions, So any condition where a human would be driving a car, a level five autonomous car should be able to operate

in that same situation. There may not be any steering wheel or any controls at all in a vehicle, meaning there's no option for a human driver to take over. Now that's not our prerequisite. You can have a level five fully autonomous car that would still have controls and still allow humans to take over manually if they if they chose to do so. It's just that it's an option, it's not. It's no longer a mandatory thing to have those human based controls with a level five autonomous car.

UM So we don't have any of these yet, So really talking about this is purely in the hypothetical. Arguably, we have some that are in the level four range, but there will get to that they're under very strict parameters, all right, So most experts agree that the versions of autonomous cars we've seen so far are mainly in the level three and level four categories, uh, creeping more toward a firm level four. We're kind of in the early stages of that, and there's several tests programs that are

operating almost as if we're at level five. But there's disagreement about whether or not technology is really sophisticate enough to warrant us calling any existing vehicle a level four or level five autonomous car. And so, while we have some examples of cars and I'll talk about a couple of them that lack control systems for human drivers, they are almost all prototypes and concept vehicles or in very

limited testing situations. Uh. And so therefore they don't really rank as level five autonomous cars because while they lack the controls, they cannot operate in every situation an environment that humans drive in. So it's it's too early for us to talk about deploying cars that have no way to hand over control to human driver in all regions

and in all you know, driving situations. Okay, so let's do a very quick rundown on the history of autonomous cars up to say, like or so, and to see why sub folks like like yours truly we're so bullish on the future of autonomous cars. So the history stretches back a good long ways, particularly if we're looking at stuff like power steering. But that's getting way too granular. I'm not going to do that. And the history is also really complex, and that involves lots of different disciplines

converging into the autonomous car form factor. You have stuff like robotics, you know, sensored development, artificial intelligence, computational processing, power, range finding technology, lots of things that all have to come together. And to really dive into the complex history of all the technologies that are coming together to make autonomous cars possible would require a whole mini series of episodes. So we're not going to jump into all of that

in this one. Instead, I want to focus on things like the DARPA challenges that were created in the mid two thousand's. The first one was in two thousand four and DARPA, as you'll recall, is the research and Development Arm of the depart I'm in the defense in the United States, so it's technically an organization that funds various other groups to do R and D in technologies that ultimately stand to benefit the defense of the United States.

So while there are other uses for those technologies that don't directly relate to defense or military systems, that's the primary purpose for DARBA. So in two thousand four they created this this challenge. They called for teams to build or convert vehicles into autonomous cars that were capable of navigating a long distance desert course is more than a hundred miles long, and there it needs to be no human operators, so it could not be remotely controlled, nor

would there be a driver in the vehicle. And the idea was design a car that would be capable of traveling a predesignated route from beginning to end. For that two thousand four challenge, no team was able to complete eat the challenge. Uh cars failed. Some of them went off road and got stuck, some of them just got confused and stopped. So no one completed it within the time frame that DARPA had set. But it's set the stage for subsequent competitions. In two thousand five, DARPA held

another Grand Challenge again with the desert course. This one was a hundred thirty two miles long and this time five teams were able to complete the route and the winning team was from Stanford Race. Stanford University was the Stanford Racing team. They clocked the shortest time on the course. By shortest time, I'm still talking about a long time. It took them six hours fifty three minutes to make

the one thirty two mile journey. That would mean that the average speed for the vehicle when taken across the whole course was somewhere around eighteen miles per hour or approximately twenty nine kilometers per hour, which isn't exactly tearing up the track, but it was still a very impressive achievement. I don't want to take away from what they achieved. It was incredible, especially for the time, but it's not the sort of speed you would look at and think, oh, well,

this is the replacement for the modern car. The next challenge would happen in two thousand seven, and it switched things up by requiring teams to design a car capable of navigating through a simulated urban environment complete with traffic and traffic laws like you know, traffic lights and stop signs and simulated pedestrians. It wouldn't be enough to design a car that could detect a road and follow it, or even a car capable of managing stuff like how

to how to send torque two different wheels. In order to get out of a tricky situation, the cars would need advanced collision detection and decision making capabilities. They have to obey traffic laws, they'd have to be able to adapt to potentially changing situations, the kind of stuff you might find if you're driving around a city. So in that case, six teams were able to finish the course,

Stanford Racing would actually take second place that time. They clocked in at just under four and a half hours. First place went to a group called Tartan Racing from Carnegie Melon University and they finished in four hours ten minutes. Now, the purpose of these competitions wasn't just to find out which groups of smarty pants engineers were able to build the best car. It was an attempt to kick start serious development in the various fields related to making autonomous

cars a possibility. Engineers worked on all sorts of different designs. Some incorporated lots of optical cameras. Some used lie dar, which is a type of laser based range finding technology similar to radar. So it works by zapping out a laser and then detecting any reflections coming back from that

laser light. It uses an array of sensors looking for any evidence of that laser light coming back to the sensor, and then measures the time difference between when the laser went out and when it picked up the reflection, and then, working with some math, it can figure out how far away an obstacle is from the vehicle. Not only that, it can also figure out whether or not that obstacle is moving, or if it's stationary, or if it's moving away from or toward the vehicle. It can figure out

all that. Uh, and I've talked about that in past episode, so I won't get into the whull technical details here, but it was one of those key components that's used in some, but not all vehicles that are following under this autonomous card development. It's interesting because there are lots of different companies that are working on autonomous cars. They are not all relying on exactly the same technologies to achieve that goal. Some of them are much more heavily

focused on optical cameras. Some of them are more focused on things like lidar and other sensors. Some of them involve a whole, you know, slew of different technologies that are meant to be both uh, you know, primary systems and redundant systems. So it's really interesting and it was. It was really impressive to see these teams complete the Urban Challenge, but again, it didn't immediately make everyone think

driverless cars would be available right away. The challenges, while impressive, didn't compare to what the average human driver deals with on a on a regular day. The competition times were pretty long. The average speeds were all below fifteen miles per hour, so they're all below twenty four kilometers per hour. At that speed, it was clear that these vehicles were just the earliest incarnations of technologies that would power autonomous

cars in the future. So they were airing on the side of caution, which frankly, you want in the first place. You don't want to see a lot of people say let's take some chances, when when you're talking about vehicles, I mean, they're human lives at stake. Meanwhile, another narrative drives home pun intended why a lot of folks got really hyped up about autonomous cars. It's also a sobering line of thought. And I'm talking, of course, about the frequency of fatal car accidents and how many of them

can be traced back to human error. Now, getting global statistics is pretty tough on this, so I'm going to focus on the United States because we have a lot of organizations in the US that track these kinds of numbers, and you can kind of get an idea of how big the problem is. So in tween, the National Safety Council released a report that stated and estimated forty thousand people had died in car accidents in the United States. Uh, that actually amounted to a decline of one percent from

two thousand seventeen. That was when forty one people died. Another four and a half million people on top of that had become seriously injured in car crashes in Meanwhile, the National Highway Traffics say D administration in the United States said that of serious car crashes result due to human error or dangerous choices. So, in other words, mechanical failures only contribute a very small percentage to the overall numbers.

When it comes to serious car accidents, most serious car accidents aren't caused by a tire blowout or you know, a car failing in some way. They're caused by humans doing something wrong, whether it's totally by accident or someone just makes a really bad decision, like they think, oh, sure there's no there's no dashed line here, but I'm gonna go ahead and try and pass this person on this windy rural road because I bet nobody's coming the other way. That's what we would call a bad decision,

So says the tech optimist. If you could create autonomous cars that operates safely, you could eliminate the vast majority of car crashes and thus fatalities on the road. You just remove the human error element, and suddenly you're talking about a staggering result, and that is an incredibly powerful motivator. Tens of thousands of people wouldn't die each year from these car accidents. Millions more would never be injured or affected by the tragic loss of a loved one from

an accident. Then you start moving outward, you go out another circle. You think of this as a ripple effect and you think, imagine all the contributions those people might make in the future that they'll get a chance to make because they wouldn't have had this terrible car crash. These are things we never would see come to fruition if they were to get in a fatal car crash, and it becomes this butterfly effect issue. And of course,

we want to make the roads safer for everyone. Now, I'm sure all of you have already hit upon the major issue here. The whole concept of people being safer in autonomous cars is contingent upon those autonomous cars performing better than humans already do and in every type of situation in which humans find themselves driving in. If we can't get that right, then we haven't made things safer at all. All we would have done is shifted the cause of the accidents from human error to machine error

or computer error. So we must be absolutely certain that the vehicles we make meet a very high standard if our goal is to reduce car accidents. So we have to prove that these machines operate better than people do in all the different situations people find themselves driving in before we can make any sort of declarative statement of

this is the best way forward. Now, when we come back, I'll talk about why this gets super tricky, and talk about thought experiments and things, and and also some real world scenarios that kind of illustrate why this is harder than what it sounds. But first, let's take a quick break. Before the break, I positive that a future with autonomous cars that all but eliminate fatal car crashes hinge upon building driverless vehicles that are much better at driving cars

than humans are in all situations. Now, we could get a bit more lucy goosey here, but doing so brings up some tough ethical issues. So, for example, what if we knew that machines were better. Right, autonomous cars are better than human drivers, but they are by no means perfect. So what if we could be certain that autonomous cars, if widely adopted, would reduce those fatalities by half, for example, but they would still be at fault in the case

of the other half of fatal car accidents. So let's say it's you know, I don't know, and we have level four autonomous cars that are pretty reliably level four and they are better as a whole than human drivers are. So we've seen a vast reduction in vehicles operated by humans. And let's even assume that most cars are now controlled by computers, but let's also assume they're not perfect now.

Using the numbers from if humans were still in control, we would expect to see another forty thousand fatalities due to human error. And I'm just using that number as an example. I realized that in reality we'd be talking about nine of forty thousand. But now that cars are in control, it means half of those accidents are totally prevented. But we still see fatal accidents that claim twenty thousand lives.

On the one hand, we could look at that scenari are you and say, based upon what we know from past experience, we would have seen many more people die in accidents if humans were actually still operating cars. But on the other hand, that's all hypothetical, right, I mean, we can only know anything based on what actually happened, not on what might have happened if things had gone

a different way. We can't be sure. But more than that, though, we're still talking about twenty thousand people losing their lives and all the ripple effects that that makes throughout society, and moreover, we have machines that are at the fault for those twenty thousand lives being lost, and the idea that people have built machines that, through a failure of some sort or another, resulted in deaths is a very difficult proposition to accept. Also, it's just a key to

think of in terms like that. I mean, clearly, one death is too many. We don't want to see anyone die in a car accident. Having a discussion in which you compare a fewer number of deaths and referring to it as quote unquote better is something that's pretty hard for us to process. It's easier to do it the other way, right, I mean, it's obvious that forty thousand people dying is worse than twenty thou people dying, But it's hard to view it the other way because anyone

dying at all is awful. Now. Part of this also really boils down to a fear of handing over control to a machine. I know a lot of people bulk at that idea. They don't like the idea of not being the actual entity making decisions behind the wheel. Confronting them with statistics showing how human error leads to catastrophe, doesn't tend to sway them. I mean a lot of people think, well, yeah, that's other people. I am not

that person. Also, to be fair, we don't have the evidence to show that computers would necessarily be better, so they're something to that right now. Okay, let's let's get back to where we were in our history. The Grand Challenges helped set the stage for the next phase of development, which was mostly the realm of startups and some big companies, namely Google, would hire participants from the Grand and Urban Challenges to come and work in new divisions dedicated to

creating driverless cars. The early pioneering work was now shifting pun intended into a phase of rapid iteration, as engineers and computer scientists and mechanics began to refine technologies to help make them better. So going from the first sort of proof of concept approach to how do we make this a better design so it does the thing it does but more effectively. Google's program began in earnest around two thousand nine, not long after the Urban Challenge. In

twenty ten, publications began to report on the project. So it's been secret for about a year, maybe almost two years. Google had been testing vehicles in and around the Mountain View, California, headquarters for the company. And while the vehicles still had manual controls and they still had a driver behind the wheel, there were at least some segments of some of these test drives that felt totally under the control of the

vehicle itself. It was ranking up miles of autonomous driving experience. It was gathering data, and those people who are working in the division use that data to further refine their approach. By the company had logged more than one hundred forty thousand miles driven by autonomous vehicles, which equals out to around two five thousand kilometers. And that's pretty, you know, respectable distance. But let's compare that against the miles that

were driven by human drivers in the United States. So in US drivers accumulated nearly three trillion miles trillion. So that means if you were to do a percentage and you were to say how many how much percentage of miles did Google cars drive compared to human drivers in the US, in the Google cars would account for about point zero zero zero zero zero four seven percent of all miles traveled vehicle miles. So not you you can call it a fraction of a percent, But even that

is being generous. It's a fraction of a fraction of a fraction of a percent. Now, if you're familiar with the idea of things like conducting surveys, you know that sample size is really important. Right, If you ask five people a question, extrapolating those five answers to try and apply it to the population at large is a bad idea. It's not a good sample size. You don't have enough

data to draw any conclusions. It's definitely bad science. So it makes little sense to compare the results of autonomous vehicles that haven't even come close to accumulating a percentage of the my was driven by the population at large. You cannot compare the two because the experience is so monumentally different. Now, for several years, Google's cars operated without any accidents, at least not any that were the fault

of the driverless car itself. There were a few incidents, but they either happened when the safety driver was operating the car, so a human driver was driving the Google car not autonomous vehicle mode, or there there was the fault of some other driver. Right, someone in a totally different car got into an accident with a Google car, and it wasn't the fault of the autonomous system, but

rather the other driver. Those were really the only two kind of categories of incidents that happened in the early days of Google's testing. So at first glance, it looked like the driverless cars were truly safer than a human operated vehicle. Right, They had a much better record than human drivers did, and it may very well be the case that they were in fact much much safer than human drivers. But we have to go back to the

sense of scale here. So in the United States, drivers travel more than three trillion miles by vehicle per year. I think the most recent one was almost three point three trillion. We're getting ridiculously high in numbers, and there are around forty thousand fatalities per year. And for the sake of this example, will assume all of those fatalities were caused by human error or bad decisions, just to

simplify things. So if we do some rough math, we'll see that that amounts to one death per seventy five million miles driven. Now that's my estimate just based on back of the napkin. The actual estimates even more generous than that. The National Safety Council estimates that there's one point to five deaths per one hundred million vehicle miles driven. So what does that mean for autonomous vehicles, Well, they

haven't driven close to a hundred million vehicle miles. It means those early days when we first learned that Google had launched its project, there were so few miles accumulated that you can't draw any meaningful conclusions. Now. To be fair, I don't think many people were trying to argue that autonomous car technology as it was in two ten, was already clearly superior to human driving. This was still an

early testing phase. This was a point where it wasn't about showing that the technology was already better than humans. It was rather showing, hey, we've created technology that will allow this card to navigate and maneuver through human environments without making it a problem. So it wasn't even that our our standard is higher than human capability. It's more like, can this machine operate at the same level as humans

within certain parameters pretty restrictive parameters. Skip ahead a few years, several companies invested in driverless car technologies that included big car companies, you know Toyota and Chrysler and others GM they've all invested huge amounts of money in autonomous car research and development. UH. It also included startup companies independent startups that either we're working on components for autonomous cars like light our systems, or they were attempting to convert

or build fully autonomous vehicles themselves. And then there were ride hailing companies, most notably Uber, that we're also investing billions of dollars in this technology with an eye on replacing the fleets of human operated vehicles that were the bread and butter of their company to uh turn them all over to robotaxis. So instead of having human drivers over at Uber, you know, Uber at the highest level wants to replace them with autonomous vehicles for reasons that

are complex but mostly come down to money. So meanwhile, consumer vehicles were getting more and more sophisticated, and higher end vehicles started sporting some really nifty features that relate to autonomous cars are semi autonomous in themselves. Some of them are more modest, like lane assist or breaking assist safety features. Some are a little more spectacular, like the self parking capabilities that some cars have where they can park themselves and and pull out of parking spaces all

by themselves, like that's pretty cool. They weren't intended to make consumer cars autonomous, but were rather positioned as sort of value added options for cars, like this is something nifty this car has, other cars don't have it. Don't you want to buy this car? And they give a hint of what might be in days to come. In Elon Musk started talking about an autopilot like feature for cars, and sure enough, the following year, Tesla unveiled a driver

assist suite of features called autopilot. Now, personally, and I've talked about this before, I think naming it autopilot was the wrong move. I feel like the word auto pilot has a loaded meaning to it. It conveys a sense that the car will take care of everything for you, and that's not necessarily the case. In fact, that's not

the case at all. The company tried to walk that back a bit, uh not by renaming it, which I think they needed to do, but they included messages, and this they also need to do, But they included messages that said drivers were not meant to remove their hands from the wheel or to take their attention away from the road, that these systems can assist, but they don't replace the need for a driver, and you have to agree to that before you can enable the autopilot feature.

So the goal was saying, well, you have to acknowledge the fact that no, this is not meant for it to be an autonomous car, and not to go off on too much of a tangent. But I feel as though Elon Musk might be a little too aggressive with his projections about autonomous cars. And I don't mean to suggest that Elon Musk and Tesla are interchangeable. I do see that happening a lot in techt circles, where people will use one or the other interchangeably, and they are

two different entities. But maybe Tesla the company's bravado stems from Elon Musk's own personality. But whatever the case, autopilot has proven to have its own limitations, and we saw that manifest in some rather high profile and tragic accidents. Beginning in ten, there have been several fatal accidents involving

Tesla vehicles operating in autopilot mode. The first one took place on January twenty, two thousand sixteen, in China, and the most recent examples I know about took place on December two, thousand, nineteen, and there are actually two crashes with fatalities that day involving Tesla vehicles reportedly engaged in autopilot.

I say reportedly, because I don't have access to all the data, I don't know if conclusively they've discovered that both of these vehicles were actually operating an autopilot mode. One of these happening California and the other happened in Indiana, both in the United States on December nineteen. Now, Tesla states that autopilot is meant as a driver assist feature

and it's only semi autonomous. But at the same time, Elon Musk has said repeatedly that his goal was to get a fully autonomous vehicle on the road by the end of twenty nineteen, which now has been pushed back to sometime in the first quarter of so there are some conflicting messages coming out. Since a fully autonomous car and I'm talking about something that we would at least classify as level four, if not level five, is well

beyond just a driver assist mode. And I should also add that Tesla drivers have a responsibility to use these features safely and as intended. If someone is taking their attention off the road, or they're sitting back from their steering wheel, or they're taking a nap, or they're watching Netflix or whatever, that's dangerously irresponsible behavior, and they are accountable for it. I don't want to give the implication to you guys that I think Tesla the company is

fully to blame in this case. I actually think it's a shared responsibility, and that you've got some drivers who are eager to test out admittedly really cool and technologically advanced features, and you have a company that might message out these features in a way that isn't perhaps the most realistic or responsible method. It's a really bad combination, right. You've got people who are tech heads who are eager to play with the newest stuff. You've got a company

that's Bill's reputation on creating super cool new stuff. It's only natural that you get when you combine those two, you can get some bad situations if they haven't been messaged properly. And I really feel that Tesla bungledness that the rollout needed to be done in such a way where there was never the implicate atian that this was

an autonomous mode. Uh. Saying hey it's not autonomous after you've already called it autopilot and put the idea in the into people's heads is a little late in the game. So I think that that all parties here share accountability. It's not just Tesla the company's fault, and it's not entirely the driver's faults, although I think it's more their

fault than the company's. Honestly, I mean, we're all adults, right, you should be if you're driving a car, and if you're an adult, you should be able to make the determination of hey, this is a bad idea. I should also add that Tesla is not the only company that has had autonomous or semi autonomous vehicles involved in fatal accidents.

There was a case in Tempe, Arizona, involving Evolvo that had been converted into a semi autonomous vehicle that was being operated under Uber and that car hit a pedestrian while an autonomous mode, and the pedestrian died as a result of that accident. So Tesla is not the only company that has had tragedy befall it due to you know,

failures in autonomous systems. Getting back to the scale argument for a second, when we're talking about autonomous systems allegedly at fault for accidents that lead to fewer than a dozen deaths, you could say, like, well, it's all tragic. You never want to see anyone die. One death is really too many, but still twelve less than twelve, that's

that's so much fewer than you know, forty thousand. And you might be tempted to say these are tragic accidents, but if you look at how many are caused by humans, there's really no comparison. But once again, you have to remember that humans account for way more vehicle miles traveled

by several orders of magnitude. So really the only way you could compare the two is if you had autonomous systems driving as many miles as humans are driving, and then you'd have to see if they still stacked up favorably, if those numbers were still matching up are still mismatched, like if if a ton of this car is still accounted for, you know, uh, significantly fewer accidents. But we can't say that because the autonomous cars are driving far

fewer miles than humans are. So it is true that most accidents involving autonomous vehicles seemed to be the fault of human drivers. You know, it's not like most of the accidents we hear about were caused by the autonomous vehicles themselves. It tends to be that someone else, some other human, caused the accident. But the case of these fatalities, it does look like it was the autonomous system at fault,

and that's truly truly concerning um. And also, you know, when when it's when it's a person who's at fault. We understand that people make mistakes, and we can feel, at least in some cases, we can feel some sympathy for a person where perhaps the situation was truly out of their control, that that situation was was partcularly extreme or unusual, and so we can feel so some sympathy for the person. But when it's a machine, then we've already surrendered control up to it, and that's where it

gets particularly scary. You know, we have to trust in the machine, and when the machine betrays that trust by failing, that's a big problem. So what happens when there are no controls at all? The humans can access more on that in just a moment but first, let's take another quick break. One of the challenges autonomous car companies and engineers have faced is how do you balance between computer and manual control of a car? You know, how should

control switch from one to the other. When should an automated system take over to avoid an accident like a collision prevention system, or when should a driver be able to override autonomous commands and bring the vehicle under manual control. Doing this is not as straightforward as you might think, and and doing it in a way that's safe and has a seamless transition of control is really hard. But

what if there's no question about it at all? Because there are no controls to take See back in twenty four Google showed off a driverless car prototype that had no steering wheel, had no accelerator, no brake pedal, so there were no controls for a human to take over. The car would only operate autonomously because there were no other options. The prototype worked with a smartphone app and acted as sort of a ride hailing or robo taxi service.

Users could summon a car using the app and they would indicate where they were wanted to go within a very restricted range of operation. Like it was geo fenced, so you couldn't go beyond a certain border that was pretty limited, and that meant that the vehicle had a lot of variables reduced, right it It cut back on the types of conditions and routes and situations the car might encounter, and thus made the problems of having an

autonomous car slightly less complicated. There's still opportunities for complications, but you've drastically reduced them because you've reduced the variables. Well. The vehicle used an electric motor that was good for about one miles of driving per charge, and it boasted a top speed of twenty five miles per hour. So this little car would only really be suitable for transportation and restricted situations such as the campus of a big

company like I don't know, Google. It wasn't intended as a practical vehicle for widespread adoption, but rather another iterative step towards fully autonomous cars. The robotaxi vision is one that tends to be the most common across the autonomous

car space. That's largely because the technology used to of cars autonomy, you know, the the sensors, computers, robotic systems, that kind of stuff they don't come cheap, and a vehicle would cost significantly more than a manually operated vehicle a traditional car, So most experts agree that the future of autonomous cars, at least in the near term, will be in fleets that are operated by companies like Uber

or Lift. They will be ride healing vehicles or robo taxis, and they will take passengers to their destinations, and then those cars will then move on to pick up their next fair, or they'll return to some sort of h Q for recharging or maintenance or whatever. It's unlikely that we're gonna see autonomous vehicles offered up for private ownership right away for the most part, due to the prohibitive

expense of this additional technology. The Google's experiment pointed out both the advances of the tech and the limitations of autonomous car technology. Yeah, the car had no controls, which is what you would expect only if you had a level five autonomous car. But it also had very strict geo fencing restrictions and operational restrictions, so it couldn't go very fast, it couldn't venture very far, it wouldn't likely

encounter unusual situations. So because of that, it wouldn't be Level five anyway, because you've you've limited the scenarios where it would be operating in the first place, it would not be driving into all the different situations that a human driver would encounter. A truly autonomous vehicle would need to be able to handle everything, all sorts of unpredictable situations.

The average person isn't likely to encounter a truly unusual experience on any given drive, right, It's not like if you drive down the road you're going to see every single outlier. That's very unlikely. However, when you have a collective three trillion vehicle miles traveled per year, you're bound to get some pretty extreme situations somewhere in those three

trillion miles. So you might have a person who has to drive through a dangerous environment, like maybe mud slides are coming across a road, or when people were evacuating parts of California that were affected by wildfires, or there might be you know, animals in the road. There could be people in the road. Weather effects can be unpredictable, and they can change driving conditions rapidly. There are all sorts of things that humans encounter every year, with varying

degrees of success and maneuvering around or through them. And if we actually do see autonomous cars take up more of the car landscape, those autonomous cars are also going to encounter those situations too. It's just a matter of the odds, you know. And there are a lot of unanswered questions about how these cars are going to deal with those situations when they arise, and that includes the

famous trolley problem dilemma. Now, in the classic trolley problem, you're presented with a hypothetical situation in which a trolley is out of control. It's moving down the tracks, uh and it cannot stop. So if you do nothing, if you do not act, the trolley will continue down the track and it's going to hit a group of five people.

It's gonna there's no doubt it will kill those five people. However, there next to a lever, and if you pull that lever, you will send the trolley down a side track, so it will miss the five people, but it will definitely hit and kill one person. So if you do nothing, five people die, But if you act, one person dies. So does making the choice to pull the lever amount to murdering that one person? Did you just choose to

kill that person. Does doing nothing mean that you've murdered five people or does it just mean that you allowed five people to die? Is there any meaningful difference between those two things. Well, these are all questions and ethics, but with autonomous cars it gets into less hypothetical territory. You have to actually start to answer these questions. Cars may very well encounter, since you, ations in which there

is no way to avoid injuring or killing someone. So in those cases, what do the cars do you know who? How do the cars choose which person is to be put at risk? How do they decide what action to take? Do they try to protect the people who are inside the car at all costs, so in other words, yeah, we're gonna make this decision which will protect the people who are inside the car. By anyone else there they

are fair game. Or do they try to protect people who are outside the car who maybe don't have the benefit of the car's other safety features. Maybe you build it into an autonomous car that the people inside the car are allowed to encounter a bit more risk because your thought is, well, the inside the car is very safe, so we want to make sure we protect say a pedestrian or bicyclist. We don't want the car to hit them because they will suffer way more damage than the

people inside would. So we're going to make that decision. That's a that's a possible choice too, But these are not necessarily answered questions. There are questions that are being answered as people are designing these vehicles. One benefit that autonomous cars might have is that organizations overseeing them could, at least in theory, use the collective information across an entire fleet of autonomous cars to improve performance of each

vehicle within that fleet. So if one car were to encounter a really unusual experience, engineers could take the data from that experience and tweak the behavior of all the cars across the fleet. So when one individual encounters something new, everyone learns from that experience. So it's sort of like

the borg in Star Trek. It's a collective and that's a big advantage over human beings, right because when it comes to humans, the person who experienced something, they might learn from that experience, but that that learning, that knowledge doesn't automatically spread across the population and general So in that way, autonomous cars can have a big advantage over

human drivers. If that is used properly on the flip side, when it comes to something as potentially deadly as a vehicle, it's pretty cavalier to say, well, the cars will learn as they go, and we'll apply that knowledge to all the vehicles. They'll get better the longer they drive, because if learning also includes accidents that could potentially result in injuries or fatalities, that's a really steep price to pay for knowledge. And we're seeing more companies developed vehicles that

have no manual control systems. You know, Google came out with There's in but that's not the only case of it. In January, g MS Autonomous Car division, which is called Cruise. Originally it was an independent startup, but GM gobbled them up. A couple of years ago, they unveiled a driverless car called Origin. And the Origin, like Google's prototype, has no steering wheel, has no accelerator, no brake pedal. It has

seats that all face inward. They're kind of like, you know, think imagine two benches with with backs, but the two benches are facing each other, so the people sitting in what would consider to be the front of the vehicle would have their backs to the windshield and they'd be looking back at the people sitting in the back seats, who'll be looking forward. Uh. Now, it's about the size of a crossover suv, and that means there's a pretty

good amount of space inside the vehicle. So while you are facing the other folks, like if if you're in the front seat, you're facing the folks in the back and they're facing you. Because there's so much space, you're not likely to accidentally kick each other or anything. It looks pretty roomy. On top of that, the car has a cool little keypad on the doors. And the idea is that a production model of this car would be

used like a robotaxi. So you would hail a ride on your smartphone and this little robo car would come driving up to you, and then it would give you a multi digit pass code. You would get one on your app and you would look at that ask God, and you would type the numbers into the keypad and that would open the doors. So that way, you know, some unauthorized person wouldn't just jump into your car and

then go gallivanting off without you. You would be able to unlock the car yourself because you had a one time use key code. That's a decent concept for a working robotaxi, but the fact remains that we haven't hit level five autonomy yet. At best, we have limited level four.

Most of the vehicles we've seen in testing can perform autonomously, but only with pretty tight restrictions like that along specific predefined routes or within very strict geo fencing, or at particular times of the year or even particular times of the day. Again, that helps reduce the variables that the car mine encounter on any given day, and it gives it the best chance to operate safely, but that really

limits how useful the cars are in practical applications. For autonomous cars to work as an alternative to manually controlled vehicles, they need to brand and pretty much all the same conditions that regular cars do without restrictions, and we just aren't there yet, and we might not be for several more decades. The Prognos Research Institute actually identified four factors that are in the way of autonomous vehicles. They include technological maturity, which is what I was just talking about.

Infrastructure development, so having you know, cities that are designed in such a way that they can allow for autonomous cars the inertia of the fleet. This means that you know, there's a ton of manually controlled vehicles out in the world already, right The vast majority of cars that are out there are manual control vehicles. They might have some limited autonomy, but for the most part, they're controlled by humans.

It would take a very long time before autonomous vehicles represent a significant percentage of the overall vehicles on the road, let alone a majority. So it will take many, many, many years to wean off of a human controlled cars and go to autonomous cars, barring any legislation that outlaws vehicles um or human controlled vehicles, I guess I should say. And then finally we have legal hurdles to overcome the regulations that are going to be coming out around driverless cars.

We're seeing a lot of money poured into research and development to push the technological limits further and to establish

the foundation for truly autonomous vehicles. But I wonder if these various companies and their investors are really in it for the long haul, so to speak, because I suspect it's going to take a pretty long time to get to a point where we feel there's really reliable safe level five autonomous vehicles in the world, let alone a world in which governments have also agreed and have caught up and have defined the legal parameters for the operation

of these vehicles. Because you know, it's one thing to prove the technology works. That doesn't necessarily mean that technology will be legal to to operate, right Like, governments tend to move a lot more slowly than technology does. So if investors are willing to play the long game, then I think their investments will ultimately pay off. But it's going to take a long time, which means lots of repeated investments are going to be required to keep these

companies going, to keep them innovating and improving technologies. And meanwhile, there's not going to be an actual market for them to capitalize on outside a few, you know, test programs that don't really count, because there's no way that the revenue they're generating is actually eclipsing the cost of operation.

It's got to be a money losing proposition right now in all the different test cases at scale, with fully legal vehicles that are embraced by the general public shore it could work from a financial standpoint right now, though it's all just proof of concept that hasn't uh seem full fruition. Now. I still believe in autonomous cars. I still believe they will ultimately make the roads safer and reduce the number of deaths and injuries from car accidents. I just think it's going to take a lot longer

that I had previously imagined. And that's not necessarily a bad thing. This isn't important enough issue that we have to make sure we get it right that we can deploy vehicles in ways that makes sense, that are truly safe, that are ethical, that there are in as an ideal implementation as we can manage UH. And we have to

make sure that it makes financial sense too, right. We need to have h make sure that it truly represents an affordable way to get around that eliminates the need for stuff like garages and parking lots and dense urban centers. Those areas could be reclaimed and used for other stuff, and that stuff might be far more productive than just being a storage place for a car when it's not

being in use. Personal ownership could really be on a serious decline in that kind of future, replaced with on demand car service, and the cars that are in service would be used much more frequently rather than just sitting idle and taking up space for the vast majority of their existence. If you think about your average car um it's the amount of time you're actually using it versus the amount of time it's just sitting there doing nothing

is staggering, right. So if you're able to make more use of the vehicle, uh, then it's a more efficient use of the technology. It's a it's a better investment for all the the materials that went into making that vehicle. So you could argue, well, this makes more sense from multiple perspectives if we're able to make better use of this technology and not just have it sitting someplace taking up room. But it's it's a lot a lot of things have to fall into place for that future to

come true. I think it's a future that it makes sense, but only if we can get the tech just right and before then, what we're really risking is making bad decisions that just make it harder to get to the right future. So we have to be careful in how we're testing these things. We have to minimize risk while maximizing our our ability to learn things, which is a very tricky thing to do because Ultimately, you do have to start deploying autonomous cars into populated centers or else.

All you've done is created something that works really well in the lab, but not well in the real world, and that would be useless to us because most of us don't live in a lab. I know I don't, not since two thousand fourteen, but that's another story. Guys. If you have any suggestions for future topics for tech Stuff, reach out to me. You can find me on social media. I'm on Facebook and on Twitter with the handle tech stuff h s W and I'll talk to you again

really soon. Hext Stuff is a production of I Heart Radio's How Stuff Works. For more podcasts from I heart Radio, visit the I heart Radio app, Apple Podcasts, or wherever you listen to your favorite shows.