Subscribe
Share
Share
Embed
Urban Innovate TALKS explores how cities, technology, and people shape the future of mobility. Hosted by Dr. Henriette Cornet, the series features conversations with global experts on autonomous vehicles, public transit, and the human side of innovation — including insights from pulsur, our AI platform for understanding public sentiment and travel behavior.
Steven Shladover (00:19)
Okay, hi, Henriette. It's good to talk to you.
Henriette Cornet (00:22)
Yeah, and so I really want you to introduce yourself and say everything you have ⁓ done in the field of autonomous vehicle or automated vehicles. We'll come back on this type of terms with you. just for the listeners, so that they know for me, you are really kind of someone that I've always seen around since I worked in the field and that I think have carried a lot of important work over the years ⁓ to really help us define and
and make sure we are all talking about the same thing. And that's exactly what I would love to discuss with you today because you contributed a lot to the discussions around ⁓ SAE, the Society of Automotive Engineers. And I would like today to dig into the topic of the levels of automation because for people familiar with it, we kind of agree, although I'm not sure we always share the same definition. So I'd like to go through that with you, but for other people,
It's completely new, so we may need to really explain from the scratch what it is, etc. But before digging into the topic, can you please present yourself and tell us a bit what you did in the past and so far?
Steven Shladover (01:29)
Okay. ⁓ All right. ⁓
Yeah, I'm one of the long time people who's been involved with vehicle automation. actually began when I was a graduate student, which is over 50 years ago. And I did a master's thesis on automatic steering control back in 1973, and a doctoral thesis on ⁓ automated platooning of vehicles in 1978.
have been working on and off in the field ever since then. I've led a variety of research projects at the University of California, Berkeley. At this point I'm partially retired from work there, although I continue to do some work. And I've also been active in the technical standards work both in SAE and in ISO, and actually bringing the SAE and the ISO
together so that we try to use common terminology in that we're actually talking about the same things. So my work has included developing the technology, testing the technology on full-scale vehicles, giving public demonstrations of the technology, and more recently helping our state of California regulators with technical advice on the development and the implementation.
of their regulations to govern the testing and the public operation of automated driving systems in California.
Henriette Cornet (02:56)
Yes, you have a very impressive career because you also a graduate from MIT, correct? So the status you mentioned was...
Steven Shladover (03:04)
Yes, yeah. Right,
I spent 10 years at MIT and then spent 11 years working in a Silicon Valley contract research and development company and then moved to the University of California, Berkeley when the PATH program got started. I actually helped get that program up and running in the late 1980s. So, and then have been with the university since 1989.
Henriette Cornet (03:34)
impressive.
Yeah, thanks for sharing all that. And yeah, I realize you are really the person to talk to with this, with all the work you've done around standards. I myself have been involved with some standards to very, very small scale in comparison to you. When I was working in Singapore and recently, and it matches with what you say, with your work with California regulatory body ⁓ with ⁓ the European Commission, the European Commission has updated also or wants to update all the type of
approval, homologation, regular standards that we have and we need every time you have the word driver or ⁓ like some test that needs to be conducted with a driving wheel, et cetera, what do we do if we don't have a driving wheel and if we don't have a driver? So I helped them to update all that. So I'm kind of used to the tedious work.
of standardization and what it means to sit down and read and very very precise about taxonomy etc. So I see kind of a shared mindset maybe on this topic.
I think we should dig now in the topic of the day, like the levels of automation.
So when you meet new people, people that are absolutely not in the field of autonomous vehicles, that have no idea what it is, that just came to San Francisco for the first time and discovered the Waymo and have never heard about it, and they hear the topic of levels of automation, how would you describe that to someone who has really never heard about autonomous vehicles before?
Steven Shladover (05:09)
Yeah,
I generally describe levels of automation as simply the definition of what are the functions that the human performs and what are the functions that the system performs. And what's the division of roles between a human driver and the system in the vehicle, because those can vary widely. ⁓
And that's just one dimension of the description of any automated driving system. It's of equal importance is the operational design domain for the system because the operational design domain covers the range of conditions in which the system is designed to actually operate. And the third dimension of that is what transportation service is the system performing.
Is it moving people or is it moving goods? Is it doing long-haul trips or is it doing short-haul trips? Is it operating in a fleet environment or is it something that's intended to be used in private personal vehicles? Those are all really different from each other and it's just as important to understand those distinctions as it is to understand what's the role of the person, what's the role of the technology.
Henriette Cornet (06:27)
I love the way you put it in such a concise way. But when you say we need to distinguish people and goods,
Do you mean the levels of automation don't apply the same way if it's for people or goods?
Steven Shladover (06:41)
The levels of automation apply the same way, but ⁓ I'm thinking in terms of a description of a system. For example, if you wanted to describe a Waymo ride hailing vehicle or an Aurora truck, they're really different from each other, even though they're operating at the same level of automation. So they're both level four systems, but they're
dramatically different level four systems. They're doing something quite different from each other.
Henriette Cornet (07:11)
Yeah, I see what you mean. So there is a bigger story than just the levels of automation.
Steven Shladover (07:17)
Right
and the reason I emphasize that is sometimes people try to load everything onto the levels of automation and say that all the systems at a particular level are equivalent to each other and they're not they can be dramatically different from each other and the other thing that's also really important to recognize is the levels of automation apply to features in the vehicles they do not apply to the vehicle
vehicle does not have a level of automation. It's the feature that's implemented on the vehicle that has a level of automation. After we go into the levels, then I can explain how an individual vehicle may have different features that operate at different levels of automation in different operational design domains. Same vehicle, but multiple features that do different things in different kinds of conditions.
Henriette Cornet (08:10)
Yeah, let's go for it. Let's jump into it and tell us more because I do the shortcut to say, this service, I would maybe speak more about the service. this is a L4 service. this is a L2 service. And in my head, it's easier to distinguish what is more like passenger cars.
or car manufacturer kind of the environment and those are more like the ride hailing, the fleets. So I do like shortcuts and I think for the people it's maybe still confusing because they already heard you saying L4 and I mentioned L2 so I think we just need now to go for it and describe the six levels of automation if you don't mind.
Steven Shladover (08:39)
Yeah.
Yeah. ⁓
Well, and and actually in the updated version of document, we're really talking about as five levels of automation because level zero is not a level of automation. That's the absence of automation. And then there are really five levels of automation. And the other thing.
to bring up at the beginning is this is automation on a sustained basis, sustained automation of the dynamic driving task. So for example, an automatic electronic braking system that just might do emergency braking in a hazard situation is not sustained automation. So it doesn't fit within the automation classification or a lane keeping system that will tweak the steering to keep
a vehicle from departing from the lane is not a level of automation. A lane centering system, a system that's continuously maintaining the vehicle on the path towards the center of the lane, would be a level one driving automation system. But a system that just ⁓ keeps you from violating lane boundary occasionally would not count as a sustained driving automation.
No, not even level one. It's because it's not operating on a sustained basis. It only comes into effect intermittently based on, say, a hazard condition.
Henriette Cornet (10:19)
like the parking
when you park that helps you park your vehicles will also fall outside
Steven Shladover (10:25)
Well, if
you're talking about the system that gives you the beep if you're getting too close to something, that's a warning system. So the driver is doing all the driving, you know, the system is just giving warnings. ⁓ So yeah, that's certainly a warning system, but also a forward collision avoidance system, a system that will slam on the brakes to if you're about to bump into somebody.
is not a driving automation system that's because it's not doing that function on a sustained basis it's not changing the division of roles between the driver and the technology in the vehicle so we just say that so a vehicle that might be equipped only with those systems does not have any driving automation
Henriette Cornet (11:16)
I see, see. That's a great clarification to start with because indeed I see and even you see for me even the warning system all the kind of the tech that comes in the vehicle and makes the driving easier for me it was like on the path to automation but you clearly distinguish.
Steven Shladover (11:20)
Yeah.
Yeah.
it
and I think it's okay to think of it as being on the path to automation it's just that it is not changing the division of roles between what the driver does and what the vehicle does so the lowest level of automation which would be level one
is a system that either controls the longitudinal motion of the vehicle, that's the speed and the separation to vehicles in front, or might control the lateral motion of the vehicle, the position within the lane. So like adaptive cruise control is the most widely used level one type of system. It only controls the speed and the separation to the vehicle in front of you. The driver is doing everything else.
So it's only taking that small part of the dynamic driving task and doing that instead of the driver.
And similarly, ⁓ lane centering system, which is controlling the steering, keeping the vehicle centered within the lane, is only taking over that part of the driving task. And the driver is still controlling the speed and the driver is still controlling the separation to the vehicle in front of him or her. So that's the lowest level of driving automation because it's only taking that part of
driving task. So then if we go from level one to level two, that would be the system that covers both the control of the separation between my vehicle and the vehicle in front and controls the lateral motion of the vehicle. But the driver is still doing everything else.
And what that means is the driver is looking out for hazards. The driver needs to continuously supervise the operation of the system because the systems are not intended to operate 100 % of the time under 100 % of the conditions that you encounter. And those are technological limitations.
And they also become economic limitations as well in terms of the cost of implementing the systems. So those are systems that can be of assistance to the driver, but the driver is still doing the driving. He or she is just being assisted by the level two driving automation system. And yeah.
Henriette Cornet (14:05)
So to pause here and to recap,
level 1 will be either longitudinal or lateral, but not combined, and level 2 brings the combination of both, if I simplify.
Steven Shladover (14:16)
rights combination of both
under and the really important thing is the driver is still the driver. So, ⁓ and that's important from the legal perspective as well as from the need for the driver to understand.
that they still need to be fully engaged in the driving. They can't be ⁓ looking at their smartphone. They can't be texting or doing those other things because they need to be continuously paying attention to what's going on in the road environment.
Henriette Cornet (14:51)
Yeah, and you also
say the word legal, which I think is key when you look at what the automotive industry is doing, because at the end it's a lot about liability in case of accident. But maybe we'll come back to that. Maybe it's going already too much in detail.
Steven Shladover (15:05)
Yeah, we probably need to come
back to that a little bit later. But it's also important to recognize that because of the potential for people to misunderstand the limitations of the system, the developers need to incorporate driver monitoring in these systems. And they have to have systems to detect is the driver actually doing what they're supposed to.
is the driver actually paying attention to what they're supposed to be paying attention to so that they are prepared to intervene when they need to in order to maintain safe operation. So the earliest sort of least sophisticated systems at level two
normally did that by trying to detect whether the driver's hands were on the steering wheel or whether the driver was applying torque on the steering wheel and they used that to try to estimate whether the driver was still engaged and users pretty quickly figured out that they could trick that type of a driver monitoring system into thinking that they were driving when they really weren't so
And, you know, there were videos circulating in those earlier years of people strapping beer cans onto the steering wheel so that it would give a torque and would trick the sensor into thinking that they were still steering when they really weren't. So the more modern systems use video cameras and especially infrared video cameras to look at the driver's face and to detect is the driver's
Are the driver's eyes looking forward? Are the driver's eyes open? Has the driver not had their eyes closing? To try to estimate whether the driver is still actively engaged in the driving task. And then normally if they detect the driver is not doing that, the driver's looking somewhere else, if they're looking down at something in their lap, at a tablet or at their phone, or going somewhere else, ⁓
then the system will start giving alerts and they will, the systems will typically then start slowing the vehicle down and bring the vehicle to a stop if the driver does not remain engaged in the driving task. And now that's a case where some manufacturers have started using this designation, a level two plus to indicate that the drivers don't need. Yeah. Which is spurious.
Henriette Cornet (17:35)
wanted to come back to that.
Steven Shladover (17:39)
That's it's total bullshit but we we see a lot of people talking about level 2 plus or level 2 plus plus and That's all rubbish. It's still level 2 because the driver still needs to be paying attention and Continuously supervising the system. It's just they've used a different technology It's a marketing thing. Yeah, but they're they're using a different way of trying to detect whether the driver
Henriette Cornet (17:58)
Some marketing, probably.
Steven Shladover (18:06)
remains engaged in the driving task. then they put a plus on it. And it's really misleading. Yeah. Yeah.
Henriette Cornet (18:11)
Okay, that was on my list of questions, like the L2 plus that we hear
from time to time. also get when people say that, I know a bit what they mean, but I wanted to have your opinion on it. I think you clarified it. You were very clear on it.
Steven Shladover (18:22)
But yeah, but it's just a difference in how the
system is monitoring the driver engagement and it's saying you don't have to have your hands on the wheel for the system to be able to continue operating because it's actually.
looking to see whether your eyes are in the right place and the infrared is really important for that to be able to deal with things like nighttime conditions where you know normal video camera won't be able to see your eyes or people wearing sunglasses or different types of eyeglasses that would make it hard to see with visible light whether the driver's eyes are
pointing in the right direction but with infrared, they're in a much better position to be able to detect that.
Henriette Cornet (19:04)
wow okay okay okay i think for we we got it for level two
Steven Shladover (19:06)
OK. OK.
So then level three is now the big step up from level two. Level three is a situation in which the system is performing the complete dynamic driving task under certain conditions, and it is doing that within conditions where it
may need to call on the driver to take over under certain conditions. level two, it's up to the driver to monitor continuously and take over whenever there's a problem. When you get to level three, the system has enough self-diagnosis to know whether it's okay or whether it has a problem. And then when it has a problem, it gives a warning to the driver saying, you need to take over.
and you need to take over quickly because the system may not be able to continue doing that driving task for very long after it's run into the problem. So that's a case where somebody
Henriette Cornet (20:15)
We can see the problem
here.
Steven Shladover (20:17)
Yes, this is a problem and it's a really challenging human factors issue and it's a challenging human machine interface issue as well.
for example, you can't have the driver going to sleep with a level three system. And indeed, that's where you also have driver monitoring to make sure that the driver's eyes are still open, that the driver has not fallen asleep.
then you would be in really unsafe situation when the system needs that driver to intervene if the driver is not available now you've got a big problem but to get to level three it's necessary to have a lot of built-in redundancy in the system the technical requirements on the level three system are much higher than they are in
level two system because you're not depending on the driver to monitor the system and to look out for unsafe conditions on the road.
The initial systems for motorway use were specifically for use in relatively low speeds in dense traffic conditions. traffic jam assistance with the idea that if the driver doesn't take over in time, the system can still stop the vehicle and stop the vehicle safely without getting into a hazardous situation that might cause a rear end crash, for example.
Henriette Cornet (21:38)
Yeah.
Steven Shladover (21:40)
And
now, as the technology has been improving, the developers are extending the speed range of those systems so they can operate at higher speeds on motorways. They might still typically need to have another vehicle in front of them that they can be following for their longitudinal control.
These systems to particularly also have pretty elaborate displays within the instrument panel and lights on the steering wheel, other things to try to ensure that when it's necessary to regain the driver's engagement, they can get the driver's attention and get the driver.
Henriette Cornet (22:17)
Even with the belt, right? I've heard that you have
even system with the seat belt that there will be like system to like vibration and things like that that would like make sure. Yeah.
Steven Shladover (22:26)
Yeah, and they can use haptic signals or vibrate the seat also. That's another very
effective way of getting the driver's attention. ⁓ Put some vibration on the seat and whatever the driver is doing, that's more likely to get the driver's attention. And each manufacturer has to decide what's the, what do they think is the most salient way of getting the driver's attention back when they need to within
a matter of a few seconds so that the driver can then resume driving. And they also have to have the countermeasures that if, for example, they detect the driver is falling asleep, they need to be able to stop the vehicle and park the vehicle ⁓ so that you don't get into a situation in which the driver needed to intervene and the driver wasn't able to intervene.
Henriette Cornet (23:15)
So it's a super tricky
level and I think we should keep also for later after going through all the level of automation a little bit of view on the market because L3 is a bit kind of where the market is not really going, let's say like that.
Steven Shladover (23:26)
Yeah.
Yes, yeah.
Yeah, it's at a very dynamic state right now. So and one of the challenges there and one of the reasons it is is because the advantage for the driver is not that big compared to a level two system. Level four is a much bigger step in terms of the driver advantage, but
it's still expensive to implement it because it requires redundancy and all the sensing capabilities requires redundancy in the actuation system and requires this more elaborate user interface. So it's a lot more expensive than level two, but it's not necessarily that big an advantage for the user compared to level two. So I don't predict.
a big future for level 3. But again, it's more of a stepping stone to get towards level 4. And then, so when we get to level 4,
Now we have a system that does perform the complete dynamic driving task. But the really important thing is that distinguishes it from level three is that it can achieve a mitigated risk condition under whatever kind of hazard or fault condition it encounters. And that means that if any bad road condition comes up or any failure on board the vehicle comes up.
the system has to be able to bring the vehicle to a safe stop. And with that level of automation, it's now possible to allow the driver to go to sleep, for example, so the driver could actually sleep on a long trip.
Or you can have somebody in the vehicle who's not a qualified driver. So that's now a vehicle can be used by people who are, whether they're really young or really old, that people are not qualified as drivers. And that's what enables the automated ride hailing services and the trucks that can operate without a driver on board. So that's the big step that enables
actual driverless operation but still within a bounded operational design domain so and obviously they're really
Henriette Cornet (25:49)
Conditions, can we
summarize a bit, ODD are more or less conditions.
Steven Shladover (25:54)
Yes, but the ODD is a huge set of conditions for any real system. again, this is a different SAE standard that's in the ballot right now. There's a standard J3259, which defines the range of ODD attributes. And we'll just tell you that there's a big part of this is a big table that goes on for 20 pages listing the different types of
attributes that go into the operational design domain.
So those are things like the category of roadway that you're on, the weather conditions, the lighting conditions, the traffic conditions in terms of density of traffic, and what other types of road users do you interact with? Are you only interacting with other motor vehicles or do you also have to interact with pedestrians and cyclists and people on micro mobility devices? And ⁓ you have to be able to
In fact, for any Level 4 system, have to be able to recognize emergency vehicles and ⁓ emergency responders and be able to respond correctly to the emergency vehicles and emergency responders.
But a comprehensive description of ODD is really quite complicated. And one of the challenges with that is how do you simplify that to the extent that a normal user can understand what are the limitations in terms of their ability to use the system.
⁓ The other important thing is that any of these systems need to be able to recognize whether any of their ODD limitations have been violated. So for example, if they have weather limitations and the weather suddenly changes so that, let's say it's too heavy a rainstorm or the fog has gotten too dense so they don't have sufficient forward visibility, they need to be able to recognize that right away and then go to the mitigated risk condition
unless the driver is able to take over and resume the driving task. So again, that will vary depending upon whether there's actually a qualified driver in the driver's seat. But in some cases, the driver might be able to take over to complete the trip. But if there's no driver in the vehicle, then the vehicle is going to have to park itself.
Yeah, and or wait for the conditions to change so that then it's able to to continue doing the driving. ⁓
Henriette Cornet (28:29)
Yeah, and one
might be very easy.
like something easy to grasp for the audience, talking of the ODD only one component of everything you mentioned is also just the surface of the area, of the area that is covered by a service, by an L4 service. And if people go online or on their app, if they live somewhere where autonomous vehicles are operating, they would see exactly in which area they can book the service. And Zoox, for example, just extended
their ⁓ surface coverage, let's say there's a geographical coverage, so it doesn't change the road type and many things, but they just go a bit further. And I would say this is a very easy way on one of the parameter within the ODD that is easy to understand, that has something that can expand, so to say.
Steven Shladover (29:04)
Yeah.
Yes, and
as the technology advances, then the companies can expand their ODDs not just geographically, but also in terms of things like weather conditions and lighting conditions and traffic conditions. For example, some ODDs are set up in ways to
avoid particularly challenging intersections. So if they have to go from one location to another, they might not take the most direct route because the most direct route might involve an intersection that's maybe too complicated, or might require go through an intersection where they have to make an unprotected left turn. And if they're not proficient with unprotected left turns, then they might go a different route in which they don't have to take an unprotected left turn.
Henriette Cornet (30:08)
So not
everything is visible, to say. The user may not, the passenger may not always be aware of all the details of the ODD that you mentioned.
Steven Shladover (30:19)
Yeah.
Yeah, they couldn't
be aware of all of them because it would be too complicated. But the intermediate step on that is the regulators who do need to be aware of a lot more of those limitations than, say, a passenger might. But they're still not going to be aware of all of the technological intricacies behind what the manufacturer has developed. there are challenges in identifying what's the right intermediate level of description of the ODD.
because the attributes are they're not all independent they're coupled with each other and that all makes the actual definition of the ODD technically very complicated but ⁓
We should go on to level five though also. level five is now that full automation of the driving task without technological constraints associated with weather conditions or lighting conditions or geography. So the level five would be the system that can drive anywhere.
that humans are capable of driving. So they don't need to be able to drive in a total whiteout blizzards where humans can't drive. They don't need to be able to drive through flooded roads where a human driver would get stuck in the flooded road. But they do need to be able to drive under all of the conditions in which humans are capable of driving.
Henriette Cornet (31:46)
But you know
that's very good the way you mentioned it, I really like it because in my mind level 5 was always kind of an ideal that kind of even humans are not level 5, know like even like normal let's say if you're not a pilot, if you're not a professional pilot there are many conditions under which I would not drive myself and so I was always thinking of level 5 being like an unattainable kind of goal for...
Steven Shladover (32:12)
Well, well, I see
it and I do believe it is an unattainable goal, but it is still bounded by human limitations as well. I mean, again, you can't expect it to operate under a total whiteout where there's no visibility in the snow or super dense fog. Again, think of some of the Tule fog that we get in inland California where you can barely see the front of your own vehicle and you can't see beyond that.
Henriette Cornet (32:29)
Exactly.
Yeah, in Singapore we had
this heavy rain, everybody just stopped and you just wait for the rain to be over. ⁓
Steven Shladover (32:44)
Yeah. And so then, yeah.
So your level five vehicle would need to stop as well because you can't expect it to cover all of those conditions. But I do believe that is an unattainable goal. The reason we defined it and this is kind of important to bring up is because when we were first creating this taxonomy with all the different levels.
There was a tremendous amount of hype about, ⁓ automation is going to take over the world. Nobody's going to need to have to drive anymore. And the automation system is going to take you everywhere you want to go. And ⁓ those of us who are working on this thing, that's not going to happen for the foreseeable future. We need to make a distinction between that vision that has been put out in the general interest media and the reality.
of the level four systems that are being developed and deployed in a variety of different real world applications because the difference between a level four system and a level five system is bigger than the difference between a level four system and a level one system. It's because to get to level five it would be the combination of all possible level four systems.
I think that's going to be both technologically and economically unachievable. the technology is challenging, but the cost of developing it and the cost of proving that it's going to be safe under all of those conditions would not be economically justifiable when you think of all of the very lightly traveled remote areas where it would have to be able to operate.
That's not a good recipe for anybody who's trying to commercialize the system.
Henriette Cornet (34:35)
Yeah, awesome. Thank you very much for all this, for this overview. I think already it helps a lot. I will put some links so that the people see also there are nice ⁓ charts, nice diagrams that could help understand these different steps. I suggest to move to the ⁓ market and we mentioned it just now with the economic value, with the demand.
So a way I've been using also the level of automations in my work and it was to distinguish so L3 would be in the middle and kind of I don't want to say nobody's doing it because I know BMW has permit these days on L3 to test it. I don't think it sells that much but let's say let's look at L2 and L4.
and so for me the L2 represents very much the car industry also we did not mention the topic of ADAS Advanced Driving Assistance System is that the same of L2 or L1?
Steven Shladover (35:36)
Yeah.
No, and actually that term is a really problematic term. And in the J3016 document, we actually say you should not use the term ADAS because it tries to cover too many different things that are very different from each other. And putting the A in front of it,
a lot of what goes in there is not very advanced at all. They're actually pretty basic systems. So, yeah. Well, yeah, parking warning or just sort of really elementary forward collision warning, you know, a system that gives you a beep and flashing light if you're getting too close to the car in front of you. Well, that may have been high tech 20 years ago, but it's not.
Henriette Cornet (36:06)
Back to my packing warning.
Steven Shladover (36:26)
particularly high tech today.
Henriette Cornet (36:27)
Come on, I did
my driving license 20 years ago and I'm happy when it came out.
Steven Shladover (36:33)
yeah,
right. Well, yeah. So it was high tech then, but it's not high tech now. calling it advanced, I think, is really pretty misleading. But the other thing that's really problematic about it is it often is used to include things like level one and level two, which are actually driving automation systems.
Henriette Cornet (36:37)
No.
Steven Shladover (36:54)
So I think that there's an important distinction between systems that are doing driving automation and systems that are only warning the driver or systems that are only intervening in a safety critical situation for a very short period of time. And trying to put all these those things together in one basket, I think is just a recipe for confusion.
So I prefer to think about there's a set of warning systems, there's a set of collision avoidance and mitigation systems, and then there are driving automation systems.
Henriette Cornet (37:33)
In this L1 and L2, let's say L2 where it's kind of more the topic now, we have all the work that the car industry is doing and I would say probably every car makers are developing these L2 systems. Whereas on the L4 side, we would have ⁓ more the service so to say side, but I think at some point it will merge. But in my head, it was always a bit separated like this, having more the fleet services on the L4,
urban, yeah all these topics of ride-hailing services, shuttles, all these aspects of fleet and shared mobility in terms of shared assets.
Steven Shladover (38:11)
Yeah,
and I generally agree in terms of them being more fleet operated for a variety of good reasons, but I think the most important L4 applications are actually in trucking. And I think those are more important. They're more economically viable than the passenger services at L4. They have a better business case. So I expect
Henriette Cornet (38:38)
because
of driver shortages or cost or wages in general, also the wages of the drivers.
Steven Shladover (38:44)
Well, not
only that, but also the benefit to the fleet operator in terms of utilization of their vehicles. you know, because of driver hours of service limitations, know, the drivers can only drive for 11 hours and then they need to take a rest. So if your expensive truck is only being driven 11 hours a day,
but with an automated system, it could drive 20 hours a day. You've almost doubled the utilization of that expensive piece of equipment. And you can now get from one coast to the other in the U.S. in two days instead of four or five days. And when people are talking about things like fresh fruits and vegetables,
that could be a really big advantage in terms of getting them from one coast to the other in that shorter period of time. there's a big economic value to that, which I don't think you have a comparable, something comparable in the urban ride hailing market, for example. And one of the interesting things we're seeing in ride hailing,
is both Uber and Lyft see this as a hybrid market, that it's going to be a mix of automated and human driven vehicles because it would be much too expensive to buy a fleet of automated vehicles big enough to serve the peak period demand. So they have to size the fleet for
long-term demand over many hours of the day and a lot of the peak is going to have to be served by human drivers anyway.
Henriette Cornet (40:42)
You know, it matches very much also when I talk to my clients that are like transit agencies, for instance, it's really how I see the future having this hybrid ⁓ fleet and not like we are replacing all your services with like, with automated driving system, and by the way, ⁓ the Land Transport Authority in Singapore, so back to Singapore, is exactly testing now having a bus line. So a regular route.
a bus route where during peak hours driven by a still human driver and during off-peak hours they want to experience with low number of passengers, smaller vehicles but the same route, same service, etc. And I find it extremely promising this type of model for the future.
Steven Shladover (41:29)
And typically, especially
in thinking of the late night operations, it's less complicated traffic environment as well. So it's a good way of learning, moving the technology forward rather than trying to deal with the most difficult conditions right from the start. So yeah, there are many good reasons for following that model.
Henriette Cornet (41:52)
So you know every quarter I produce a list of use cases, the different use cases and the different services that exist along everything that we have talked today, starting with L3 kind of car models that you say it's not a car related, but like let's say vehicles that have L3 features, if I say it correctly now, and a prompt with the ride heading services, the buses, the shuttles, the truck. What about the delivery robots?
Steven Shladover (42:12)
Yeah. Yeah.
Henriette Cornet (42:22)
Do they fall under the levels of automation, these little robots we see mainly in LA? saw them.
Steven Shladover (42:28)
Yeah, well,
and I think it really varies depending upon the class of vehicle that they're on. if it's for the vehicles that resigned for use on the road rather than on the sidewalks, yeah, they would be under this classification because they're because it's for it's for road vehicles. OK. But if you have the little
Henriette Cornet (42:45)
It has to be on the road to four... Okay. Okay, okay, that's a good distinction.
Steven Shladover (42:53)
you know the the coolers on wheels you know the really little ones that just go on the sidewalks that's a different class of vehicle that we haven't really tried to extended to that type of vehicle
Henriette Cornet (43:06)
yeah yeah yeah and there will be soon there will be the drones and etc air taxi and etc yeah exactly yeah that's it changes
Steven Shladover (43:11)
Well, that's a different category as well, because that's now air vehicles rather than road vehicles.
Henriette Cornet (43:19)
And with everything we have to say today, this levels of automations, you think they are still highly relevant to be used by the ecosystem, let's say the community.
Steven Shladover (43:26)
Yes.
⁓ Yes,
because I think it's important to have some clarity about what is the role of the human and what's the role of the technology. Now we have to be careful about communicating with the general public because these levels of automation were never intended as things for communicating with the general public. They were really intended for use within the specialists in the industry.
and for the industry relative to the regulatory bodies so that industry and safety regulators can be speaking the same language with each other and be talking about the same things. And of course that also means lawyers need to get involved in those discussions as well because that becomes an issue when there's a crash and who
did or did not do what they were supposed to do. Was it the person or was it the technology?
Henriette Cornet (44:28)
Yeah, and that's exactly where the friction kind of happened between the different players and if we talk about that I have to ask you about what do you think where is Tesla in this spectrum? I kind of have my way of seeing where it is but the messaging that they put out it may be a bit confusing for the public as well
Steven Shladover (44:50)
Well, they're trying to have it both ways.
yeah, they basically have a level two system. They have a system that requires continuous human oversight. They do not have the capability to operate the system without the human oversight.
but of course they've got lots of fanboys who want to be out on the bleeding edge and they want to communicate with them in a way that makes it seem like they've got something that's more advanced than it really is so
Henriette Cornet (45:19)
Yeah, when they wrote
the topic of robotaxi, I found it a bit confusing because robotaxi is associated with Waymo, Waymo is level 4. So it created a bit of confusion in my opinion.
Steven Shladover (45:24)
Well.
Yeah.
Well, that's their
business model is to create confusion. so what you really have to do is look at what's in the owner's manual, which is sort of the more legally binding document. You know, if an incident happens, then people always point to what's in the owner's manual. What's the official communication from the company to the person who bought or leased the vehicle?
And that makes it pretty clear it's a level two system because that human needs to continuously supervise it. Now, we also refer to them as a level four wannabe. Yeah, they certainly wannabe level four and they've been talking about it for over a decade now. They've been claiming it's imminent for almost a decade and they're nowhere close to being that. ⁓
I believe that with the camera only approach that they followed, they will not get there. ⁓ In order to get to level four, they're really going to have to have a more comprehensive set of sensors on the vehicle. They're going to have to have a much more robust architecture. So, and that...
cause the way they communicate about what they're doing causes a tremendous amount of confusion and misunderstanding. ⁓
Henriette Cornet (46:56)
Yeah, yeah, yeah, that's what
I try also in these webinars, in these podcast episodes to clarify a bit the terminology so that it's clearer for the public. I hope some people of the public are listening as well. A last question and I know it's in the standard. The distinction between the word autonomous and automated.
And can you say a word about that?
Steven Shladover (47:18)
Well, the yeah,
the word autonomous should not be used, period. We should strike it from the vocabulary because the word autonomous means independent and self-sufficient. The word automated is the word that deals with what's the division of roles between a human and the machine.
So if you have a machine that is taking over some human functions, that's automation. Now, when we get into the vehicle world, autonomous could be applied to vehicles that don't have any dependence on data that they're getting from other vehicles or from humans. They're operating independently, but that doesn't mean that they're operating automated. So, ⁓
And this goes back into basically the robotics world where people first started confounding the two terms. So for years, it's been clear that automation is the substitution of a machine function for a human function. But then in the robotics world, people started talking about, well, now we've got this thing that can operate on its own without
having a person involved so it's autonomous and that's really a misuse of the terminology and unfortunately it sounded fancier so people thought okay all right that's a that's a makes it look like more than what it is but we actually have language in the document saying should not use the term autonomous because
It confuses things rather than clarifying things.
Henriette Cornet (49:07)
Yeah.
Yeah, thank you. It was great and I really like that you brought light on many topics. I hope it was clear for everybody. We will put resources in the description of the episode. Before ending, ⁓ I would love you to say a word about this conference, this symposium to which you are heavily involved. I have been involved a little bit in the past and I've been following it very closely. I go every year and I would recommend everybody to
Steven Shladover (49:32)
Yeah.
Henriette Cornet (49:38)
the experts in the field to join so tell us a little bit about it.
Steven Shladover (49:39)
Yeah.
Yeah, yeah, so this is the
Automated Transportation Symposium, which will be in San Diego the final week of July. We've had it for since the year 2012. The first one was 2012. We didn't even have a name for it in 2012. It was a workshop organized by the Transportation Research Board. And Jane Lappin and I co-chaired that and organized that. And then it
grew and grew under transportation research board it's now being operated under SAE, SAE international yeah yeah it's ⁓ SAE but it's the conferences and meetings side of SAE rather than the standardization side of SAE
We will have keynote talks by representatives from Zoox and from Uber, and we will have the NHTSA administrator giving a keynote talk, and we will have Seval Oz, who's the incoming Assistant Secretary for Research and Technology in the DOT, giving a keynote talk as well.
and we'll have a variety of panel discussion sessions and presentations covering the latest developments in automated driving. So I guess you can post the link for people to check in and look at the program.
Henriette Cornet (51:07)
And I'm very happy this year. I'm in several sessions contributing from different angles of my work. So I'm very glad to be able to contribute a side of learning from others. So that's great.
Steven Shladover (51:14)
Yeah.
Yeah. And it's really nice venue
right on the water in San Diego.
and ⁓ yeah with fingers crossed Waymo may have some service up in San Diego by that time so people might be able to take Waymo rides we don't know yet for sure but ⁓ that's that's one of their developing markets
Henriette Cornet (51:42)
Fantastic. Okay, so we'll follow that closely and we'll meet you and me there for sure and others from the audience. And in any case, we stay connected. Thank you so much for all the insights. yeah, we'll keep in touch. Make sure you come back on Urban Innovate Talks and share more insights.
Steven Shladover (51:53)
Okay. All right.
Perfect.
Okay, alright, very good talking with you
and hope people found it useful.
Okay, right, bye.