Autonomous Car Progress

[Daniel in SD]

Thanks for the California regulations.

However, you've selected regulations pertaining to the testing of autonomous vehicles, not full public deployment. The authority statute CVC § 38750, is similarly limited in scope to testing by manufacturers and their "employees, contractors, or other persons designated by the manufacturer of the autonomous technology." CVC 38750(b)(1). That is not Joe Public.

There's a difference between testing and actual deployment. As I said, liability assignment during operation of autonomous vehicles by members of the public has not been litigated.

No I quoted the section covering deployment. The first part of the document covers testing. The second part covers deployment. There are many rules regarding the manufacturers responsibility to owners in the regulations.

You seem to be saying that we can't interpret regulations until they have been litigated?
Why do you think your interpretation is correct when it is contradictory to what every company who has announced an L3 vehicle is saying?

 

[Microterf]

I highly doubt we will ever see level 3 systems deployed because of the liability factor.
L0-2 is all about functionality.
L3-5 is all about liability.

Obviously, a company would be bankrupt very quickly if it released a half baked L5 system that caused a bunch of at fault accidents.

 

[rxlawdude]

No I quoted the section covering deployment. The first part of the document covers testing. The second part covers deployment. There are many rules regarding the manufacturers responsibility to owners in the regulations.

You seem to be saying that we can't interpret regulations until they have been litigated?
Why do you think your interpretation is correct when it is contradictory to what every company who has announced an L3 vehicle is saying?

I'm wrong with respect to your cited document.

You're correct that there is a section on deployment, but in my defense the document is entitled "Testing of Autonomous Vehicles."

However, the regulations merely require certification that the manufacturer is capable of paying up to $5,000,000 for wrongful death, injury, or property damage caused by their technology, as authorized by CVC § 38750 (c)(3). That's the extent of state-mandated financial responsibility. $5,000,000 is chump change with respect to indemnifying thousands of autonomous vehicles. Obviously, this is analogous to the ridiculously low minimums for auto insurance liability in CA, and judgments in excess of the $5M would come out of the pockets of the corporation and shareholders.

As to litigation predictions, if you're aware of any case law where an autonomous vehicle manufacturer was found liable for damages while their product was being used as documented (including where there is a warning that driver attention is needed at all times), I'm all ears. I'm obviously not talking about robotaxi companies with a "safety driver."

I think we can agree that the future will promise to be a very interesting time for personal injury and corporate defense lawyers.

 

[stopcrazypp]

Volvo has stated two things about their intent for their L3 (highway self-driving) system:

• You do not have to pay attention in this mode. In fact you can do other things or take a nap, per Volvo. I assume it will work increasingly hard to wake you up as you are a few minutes from the planned exit.
• Volvo will assume liability for accidents that occur while in L3 driving. (Comment: they are promoting subscription models including insurance, and I wouldn't be entirely surprised if the liability guarantee were tied to an overall Volvo insurance package, to avoid messy resolution debates between insurers)

Noting of corse that the L3-ready and Lidar-equipped vehicle is to be introduced next year, but the L3 highway software is for future download when available.

Hopefully we'll get updates on all this at the June 30th event.

I will have to look at things myself later, but are you certain that Volvo is saying their system will be L3? I'm pretty sure a few years ago Volvo was adamant that L3 was unsafe and that they wouldn't pursue it, so this is a pretty significant about face.

Here's a statement by Volvo's CEO on L3 in 2017:

Mr Samuelsson expressed his concern about the so-called Level 3 autonomous driving modes. “In this mode the car is in charge of the driving, yet the driver must still be prepared to take over in case of emergency, which could be a matter of a few seconds. Volvo considers this Level 3 driving mode unsafe and will thus skip this level of autonomous driving,” Mr Samuelsson said.

Volvo Cars CEO urges governments and car industry to share safety-related traffic data

Volvo Cars, the premium car maker, believes governments and car makers should join hands in sharing traffic data in order to improve global traffic safety, Håkan Samuelsson, president and chief executive, told a conference at the European Commission in Brussels on Monday.

www.media.volvocars.com

 

[Daniel in SD]

I'm wrong with respect to your cited document.

You're correct that there is a section on deployment, but in my defense the document is entitled "Testing of Autonomous Vehicles."

However, the regulations merely require certification that the manufacturer is capable of paying up to $5,000,000 for wrongful death, injury, or property damage caused by their technology, as authorized by CVC § 38750 (c)(3). That's the extent of state-mandated financial responsibility. $5,000,000 is chump change with respect to indemnifying thousands of autonomous vehicles. Obviously, this is analogous to the ridiculously low minimums for auto insurance liability in CA, and judgments in excess of the $5M would come out of the pockets of the corporation and shareholders.

As to litigation predictions, if you're aware of any case law where an autonomous vehicle manufacturer was found liable for damages while their product was being used as documented (including where there is a warning that driver attention is needed at all times), I'm all ears. I'm obviously not talking about robotaxi companies with a "safety driver."

I think we can agree that the future will promise to be a very interesting time for personal injury and corporate defense lawyers.

Correct, every company currently capable of manufacturing autonomous vehicles has orders of magnitude more than $5 million in assets so that number doesn't have much relevance.
If driver attention is needed at all times then the vehicle is L2 (or a prototype L3-L5 which is a whole other can of worms). I am confident that for L3 vehicles the manufacturer will be held liable if there is an at-fault accident while the system is engaged. They're all saying they will be liable. There would be no point suing the owner when you have a multi-billion dollar corporation accepting responsibility!
Of course there are no examples because L3 vehicles don't exist except for maybe a few leased Hondas in Japan and they probably haven't crashed yet.

 

[dgatwood]

So you think Tesla will be liable when its L3 system does something untoward. Interesting, and now this is becoming more of a legal question that will be more fully resolved when L3 systems are deployed.

If L3 systems are deployed. I have serious doubts about whether it makes sense to bother with L3 at all.

 

[Daniel in SD]

I will have to look at things myself later, but are you certain that Volvo is saying their system will be L3? I'm pretty sure a few years ago Volvo was adamant that L3 was unsafe and that they wouldn't pursue it, so this is a pretty significant about face.

Here's a statement by Volvo's CEO on L3 in 2017:

Volvo Cars CEO urges governments and car industry to share safety-related traffic data

Volvo Cars, the premium car maker, believes governments and car makers should join hands in sharing traffic data in order to improve global traffic safety, Håkan Samuelsson, president and chief executive, told a conference at the European Commission in Brussels on Monday.

www.media.volvocars.com

Sounds like they read the document and saw that it says at least a few seconds, not exactly a few seconds. Practically speaking I don't think there will be much difference between L3 and L4.
I'm still waiting for someone to give an example of a situation where the few seconds would be a safety issue. It seems like designing the vehicle to properly handle the situation during those few seconds is so difficult that it might as well be L4. And of course the vehicle has to recognize the situation in the first place which seems even more difficult.

 

[Daniel in SD]

If L3 systems are deployed. I have serious doubts about whether it makes sense to bother with L3 at all.

Because there aren't legal parking spots on the side of the freeway.

 

[JHCCAZ]

I will have to look at things myself later, but are you certain that Volvo is saying their system will be L3? I'm pretty sure a few years ago Volvo was adamant that L3 was unsafe and that they wouldn't pursue it, so this is a pretty significant about face.

Here's a statement by Volvo's CEO on L3 in 2017:

Volvo Cars CEO urges governments and car industry to share safety-related traffic data

Volvo Cars, the premium car maker, believes governments and car makers should join hands in sharing traffic data in order to improve global traffic safety, Håkan Samuelsson, president and chief executive, told a conference at the European Commission in Brussels on Monday.

www.media.volvocars.com

Well, to be clear the L3 identification is mine, not Volvo's (and along with many here, I think L3 has its problems in definition and practical deployment). Perhaps the quote above means that Volvo doesn't agree with a minimal L3 implementation, with sub-minute hand-off and allowable to engage in the traffic lane. I could agree with that objection as being an L3 that is arguably unsafe.

I searched, and most of the recent spate of articles, and also another collection from mid-2020, do not use the SAE Levels in the press releases or articles. If Volvo means they won't do "L3" at all, and they would call it L4, I think that's questionable - my feeling, which could be mistaken, is that L4 autonomy is an expectation that a full door-to-door trip can be practically planned, i.e. a robotaxi-like capability. Yes there may be a limited ODD (the USA, certain cities, a limited area within Chandler, whatever) but within those, a non-licensed passenger could ride with a high expectation of successful trip completion. If the ODD stops at the highway exit, then it seems like a very marginal L4. It's not L2 with personal liability, constant supervision and readiness for immediate take-over. I didn't think a highway-only albeit unsupervised driving segment would qualify as L4 but maybe so.

(This discussion kind of proves the point that @daniel made in an earlier exchange with me, saying that we need more than the SAE levels to understand what we're talking about. To which I agree conversationally, but I don't support an attempt to further complicate the SAE definitions in regulations because I think it wouldn't solve the confusion nor keep up with future developments).

So I'll happily concede that L3 point and let others explain whether it's L3, L3+, L4- or whatnot. The main news is that on the hardware front, Volvo will apparently be selling a consumer car with a physically well-integrated, highly capable and modest cost (forward-facing) Lidar* and on the ADAS front they claim an unsupervised highway capability is coming to that car Real Soon. I think they do hope/believe the hardware could do door-to-door L4 but they're not ready to claim it's a near-term expectation.

*Personally I am waiting to form my opinion on whether Tesla Vision only will prove L4-capable but I hope so. Meanwhile I like seeing the march of practical productization for sensors like Lidar, better/more cameras, advanced Radar etc. that exceed Tesla's suite. And yes I know that there were some earlier examples of Lidar on commercially-available cars, but this Luminar/Volvo deployment seems to be the first really capable one.​

 

[stopcrazypp]

Sounds like they read the document and saw that it says at least a few seconds, not exactly a few seconds. Practically speaking I don't think there will be much difference between L3 and L4.

There is a huge difference, with L4, the car is expected to be able to fail safely even if the driver does not respond. With L3 there is no such expectation, the driver must respond within the allotted time given (which is not defined, only that it's not immediate, unlike L2).

I'm still waiting for someone to give an example of a situation where the few seconds would be a safety issue. It seems like designing the vehicle to properly handle the situation during those few seconds is so difficult that it might as well be L4. And of course the vehicle has to recognize the situation in the first place which seems even more difficult.

Well that is the argument by people who argue against L3. If you must design the car to be able to handle a few seconds safely, why not go all the way and just make it L4? Why bother with all the legal ambiguity and the likely legal battles that will come up if the driver does not respond within those few seconds?

 

[stopcrazypp]

Well, to be clear the L3 identification is mine, not Volvo's (and along with many here, I think L3 has its problems in definition and practical deployment). Perhaps the quote above means that Volvo doesn't agree with a minimal L3 implementation, with sub-minute hand-off and allowable to engage in the traffic lane. I could agree with that objection as being an L3 that is arguably unsafe.

I searched, and most of the recent spate of articles, and also another collection from mid-2020, do not use the SAE Levels in the press releases or articles.

Volvo does not mention L3 or L4 in subsequent releases, but they do say:
"Cars based on SPA 2 will be updated with software over the air and if customers decide to opt for it, the Highway Pilot feature that enables fully autonomous highway driving will be activated once it is verified to be safe for individual geographic locations and conditions."
I don't think anyone calls a car fully autonomous unless it is at least L4.
Next generation Volvo cars to be powered by Luminar LiDAR technology for safe self-driving

Of course if you do a search you can see all kinds of articles and journalists that claim it is L3 or L4, but I tend to ignore those designations if they don't come straight from the manufacturer given many journalists have a very poor understanding of the definitions and most never even bothered to read the SAE paper (even journalists from very respected publications are no exception).

If Volvo means they won't do "L3" at all, and they would call it L4, I think that's questionable - my feeling, which could be mistaken, is that L4 autonomy is an expectation that a full door-to-door trip can be practically planned, i.e. a robotaxi-like capability.

No, that is not the definition of L4 autonomy at all. The difference between L3 and L4 autonomy is that L3 can't fail safely if the driver ignores the request to take over. L4 can however fail safely even if the driver doesn't respond at all.

However, L4 can be limited to highways only and not finish the whole trip and still be considered L4. The difference between L4 and L5 is simply L4 has limitations on its usage (common example is geofencing, but it can also be weather, or road types). There is no requirement it must finish a whole door-to-door trip.

Finishing the trip has nothing to do with the level. For example, Tesla is aiming for an end-to-end L2, but the system is still L2 even though it can finish the whole trip.

Yes there may be a limited ODD (the USA, certain cities, a limited area within Chandler, whatever) but within those, a non-licensed passenger could ride with a high expectation of successful trip completion. If the ODD stops at the highway exit, then it seems like a very marginal L4. It's not L2 with personal liability, constant supervision and readiness for immediate take-over. I didn't think a highway-only albeit unsupervised driving segment would qualify as L4 but maybe so.

It definitely qualifies as L4 under SAE's definition, as long as the car can fail safely if the driver does not respond. That seems to be what Volvo is aiming for.

 

[daniel]

I'm puzzled at your statement that the law is limiting cars to Level 2 operation. That's absurd. Nothing precludes L3, 4 or 5 operation, other than a presumption the driver is the responsible liable party for any accidents.

I'm saying that if the driver is responsible, then it's Level 2 by definition, even if the car is capable of L5 and operating as if it were L5. Also, if the lawmakers decide to hold the driver responsible, they probably won't grant permits for L5. L4 and L5 allow the driver to take a nap, and L5 allows for there to be no driver at all. If the lawmakers want to hold the driver responsible, they would not allow either L4 or L4 operation. A car might be capable of L4 or L5 but not be allowed to operate as anything but L2. Who is responsible determines in part whether the car maker allows it to be operated without an alert drive in the driver's seat.

I'm still waiting for someone to give an example of a situation where the few seconds would be a safety issue. It seems like designing the vehicle to properly handle the situation during those few seconds is so difficult that it might as well be L4. And of course the vehicle has to recognize the situation in the first place which seems even more difficult.

Which is why a lot of people think that automakers will skip L3 altogether and go directly for L4. My feeling is that the task is so complicated that it's beyond me to even speculate on which parts of it are the most difficult, so I'm not able to form an opinion on whether L3 is a worthwhile goal. I'd pay for L3 when I would not pay for an L2 FSD, because being allowed to take my eyes off the road, even just on the highway, is worth paying for, to me.

I cannot give you an example of a situation where a few seconds would make a difference because the whole driving task is too complicated for me to parse.

 

[Daniel in SD]

Which is why a lot of people think that automakers will skip L3 altogether and go directly for L4.

Of course Hyundai, Honda, BMW, Mercedes, and Audi have all had plans to make L3 "traffic jam pilot" type systems. They've also spent time lobbying governments to put in a legal framework for them. If they every do exist, I'm also skeptical, I think the only L3 systems will be for controlled access highways (so, not very useful in Hawaii!).

 

[hgmichna]

L2 can be as safe in the city as if you were driving yourself, provided that you are as alert and ready to react as you would be if you were driving. My point is that I can relax (while remaining alert) with autosteer on the highway. I cannot relax with autosteer in the city.

So driving with autopilot is still good on suitable city streets. Maybe not as relaxing as on a freeway, but still taking some load off the driver and increasing safety somewhat, because in some situations the car's computer reacts more quickly than a human driver. Why switch it off?

 

[Matias]

I don’t know whether this is relevant metric, but 2022 Volvo will have NVIDIA DRIVE Orin™ system-on-a-chip, as standard. It can do 254 TOPS, source

NVIDIA DRIVE In-Vehicle Computing for Autonomous Vehicles

An end-to-end platform for autonomous vehicle development.

www.nvidia.com

Tesla’s 3.0 FSD computer can do 144 TOPS, source

Tesla claims its latest self-driving chip is 7 times more powerful than its rivals’

At an investor event ahead of its Q1 2019 earnings report, automaker Tesla unveiled its plans for self-driving systems in current and future car models.

www.google.fi

So they’re in the same ballpark

 

[J1mbo]

I don’t know whether this is relevant metric, but 2022 Volvo will have NVIDIA DRIVE Orin™ system-on-a-chip, as standard. It can do 254 TOPS, source

NVIDIA DRIVE In-Vehicle Computing for Autonomous Vehicles

An end-to-end platform for autonomous vehicle development.

www.nvidia.com

Tesla’s 3.0 FSD computer can do 144 TOPS, source

Tesla claims its latest self-driving chip is 7 times more powerful than its rivals’

At an investor event ahead of its Q1 2019 earnings report, automaker Tesla unveiled its plans for self-driving systems in current and future car models.

www.google.fi

So they’re in the same ballpark

That's a pretty big ballpark you got there

Is TOPS a good indicator of overall system performance? The difference in L2 capability between EyeQ3 (0.25 TOPs) and FSD3 (144 TOPs) suggests that we need another (non-subjective) way of measuring the capability of these systems.

 

[FloridaJohn]

L4 and L5 allow the driver to take a nap, and L5 allows for there to be no driver at all.

L4 also allows for no driver at all, as long as the car is operating within the parameters set by the manufacturer. For a real world example, the Waymo taxi service in Arizona is L4.

 

[mark95476]

Oh my. You understand money, right? (1) More is better and (2) there's never enough.

Individual vehicle owners are a relatively easy target. They'll settle for a nominal amount as their insurance probably won't fight.

Big corporations generally don't like bad publicity (confidential settlements), but they'll fight when they have to and can drag it out for a long time especially when there's real money involved. People will go after soft targets time and time again so pushing back makes sense.

Correct, every company currently capable of manufacturing autonomous vehicles has orders of magnitude more than $5 million in assets so that number doesn't have much relevance.
If driver attention is needed at all times then the vehicle is L2 (or a prototype L3-L5 which is a whole other can of worms). I am confident that for L3 vehicles the manufacturer will be held liable if there is an at-fault accident while the system is engaged. They're all saying they will be liable. There would be no point suing the owner when you have a multi-billion dollar corporation accepting responsibility!
Of course there are no examples because L3 vehicles don't exist except for maybe a few leased Hondas in Japan and they probably haven't crashed yet.

Yes, agreed.

As to litigation predictions, if you're aware of any case law where an autonomous vehicle manufacturer was found liable for damages while their product was being used as documented (including where there is a warning that driver attention is needed at all times), I'm all ears. I'm obviously not talking about robotaxi companies with a "safety driver."

I think we can agree that the future will promise to be a very interesting time for personal injury and corporate defense lawyers.

 

[Bladerskb]

That's a pretty big ballpark you got there

Is TOPS a good indicator of overall system performance? The difference in L2 capability between EyeQ3 (0.25 TOPs) and FSD3 (144 TOPs) suggests that we need another (non-subjective) way of measuring the capability of these systems.

It doesn't measure software capability, its simply a measurement of compute power when it comes to DL. And the TOPs of Orin is a measurement of how many deep learning operations it can do, same with HW3.

To mention others, Nio ET7 is using 4x Orin (1,016 TOPS) and Huawei Arcfox Alpha S is using new gen Ascend chip that has 400 Tops.

 

[daniel]

Of course Hyundai, Honda, BMW, Mercedes, and Audi have all had plans to make L3 "traffic jam pilot" type systems. They've also spent time lobbying governments to put in a legal framework for them. If they every do exist, I'm also skeptical, I think the only L3 systems will be for controlled access highways (so, not very useful in Hawaii!).

And there's another weakness of the SAE classification system: A car that allows the driver to take his hands off the wheel and eyes off the road, but only on a controlled-access highway below ten mph qualifies as L3, but is a totally different animal from a car that does the same at any legal speed on any highway. I would pay big bucks, like thousands of dollars, for the latter, but I have no use for the former.

So driving with autopilot is still good on suitable city streets. Maybe not as relaxing as on a freeway, but still taking some load off the driver and increasing safety somewhat, because in some situations the car's computer reacts more quickly than a human driver. Why switch it off?

I used to use EAP on city streets. My subjective experience was that it was not relaxing because of the rapidity with which conditions change, compared to the highway. As long as the system is Level 2 I will not use autosteer in the city. In an environment where split-second reaction to sudden situations matters, I find autosteer to be more stressful rather than less. YMMV.

I could imagine a time when the car's autonomy is so good that I trust it more than I trust myself, and the car is L2 only because of regulations. I would use it in the city in that situation. But we're not there yet and I don't think we will be for some years to come. And I think that when autonomy gets that good, regulators will permit it because the insurance companies will support it.

I will continue using autosteer on the highway but not in the city.