Mercedes to accept legal responsibility for a vehicle when Drive Pilot is active(roadandtrack.com)
roadandtrack.com
Mercedes to accept legal responsibility for a vehicle when Drive Pilot is active
https://www.roadandtrack.com/news/a39481699/what-happens-if-mercedes-drivepilot-causes-a-crash/
457 comments
>Handing over driving responsibility completely requires extremely particular circumstances. Right now, Drive Pilot can only engage at speeds under 40 mph (60 km/h in Germany) on limited-access divided highways with no stoplights, roundabouts, or other traffic control systems, and no construction zones...The system will only operate during daytime, in reasonably clear weather, without overhead obstructions. Inclement weather, construction zones, tunnels, and emergency vehicles will all trigger a handover warning.
Those are some big caveats that mean that you won't be able to use this in most situations. It is basically only good for stop and go highway traffic which is a situation that other driver assist features handle pretty well.
Those are some big caveats that mean that you won't be able to use this in most situations. It is basically only good for stop and go highway traffic which is a situation that other driver assist features handle pretty well.
Mercedes is perhaps the only auto company that is both supremely innovative and supremely respectable at the same time.
Tesla is innovative, but they get ahead of themselves, and Musk is a glorified internet troll who bullies and arguably lies.
The Big 3 are respectable in that they don’t really overhype or misrepresent, but Mercedes will probably out-innovate them in self-driving like they’ve out-innovated them (and pretty much everyone else) in just about every other aspect that matters over the past century.
Disclaimer: I drive a Toyota Sienna and a Subaru Outback. But I’ve seen a few Mercedes up close (and sometimes driven them) over the years. The only really crappy one in my experience was the ~2006 G500.
Tesla is innovative, but they get ahead of themselves, and Musk is a glorified internet troll who bullies and arguably lies.
The Big 3 are respectable in that they don’t really overhype or misrepresent, but Mercedes will probably out-innovate them in self-driving like they’ve out-innovated them (and pretty much everyone else) in just about every other aspect that matters over the past century.
Disclaimer: I drive a Toyota Sienna and a Subaru Outback. But I’ve seen a few Mercedes up close (and sometimes driven them) over the years. The only really crappy one in my experience was the ~2006 G500.
This is an important step. For the first time there is full self driving where the driver is no longer require to pay attention to what the car is doing. However, this is not a generic solution. This system is limited to a very narrow set of conditions. Specifically, to premapped highways with a divider between directions, no crossings, just on- and off ramps. On top of that, at very low speed, so in normal traffic you unfortunately cannot use it either.
So how does this compare to Teslas FSD (not Autopilot, which really is only an assistant)? The answer is: not much at all. The Mercedes system has limited the conditions to a point, where it basically just has to do collision avoidance with traffic with predictable behavior. It is limited to selected roads. You don't need much "AI" here and for this task, LIDAR excels in the specified weather conditions.
The Tesla FSD system on the other side aims to be a generic driving AI. It has no limitations about on which roads it operates, it uses maps only for navigation to the destination. But the driving operation is done solely by the evaluation of camera footage. Which already works in many more environmental conditions than the Mercedes system.
These systems approach the goal from entirely two directions. The Tesla system aims to be a complete solution, but certainly is not able to do that unobserved in the current state. It is doing reasonably well at that, but the literally billion dollar question is, can it ever get so reliable, that the driver can completely hand over to car ever? On the other side Mercedes has full autonomy, but they achieved that in a very limited and controlled environment - it remains to be seen how much they can extend that.
So how does this compare to Teslas FSD (not Autopilot, which really is only an assistant)? The answer is: not much at all. The Mercedes system has limited the conditions to a point, where it basically just has to do collision avoidance with traffic with predictable behavior. It is limited to selected roads. You don't need much "AI" here and for this task, LIDAR excels in the specified weather conditions.
The Tesla FSD system on the other side aims to be a generic driving AI. It has no limitations about on which roads it operates, it uses maps only for navigation to the destination. But the driving operation is done solely by the evaluation of camera footage. Which already works in many more environmental conditions than the Mercedes system.
These systems approach the goal from entirely two directions. The Tesla system aims to be a complete solution, but certainly is not able to do that unobserved in the current state. It is doing reasonably well at that, but the literally billion dollar question is, can it ever get so reliable, that the driver can completely hand over to car ever? On the other side Mercedes has full autonomy, but they achieved that in a very limited and controlled environment - it remains to be seen how much they can extend that.
> Right now, Drive Pilot can only engage at speeds under 40 mph (60 km/h in Germany)
I just felt the urge to clarify that this is written with the intention to mislead readers, and that 40 mph are also 60 km/h in other parts of the world.
I just felt the urge to clarify that this is written with the intention to mislead readers, and that 40 mph are also 60 km/h in other parts of the world.
I may be too cynical about the car industry, but wouldn't this be a great strategy from Mercedes to hamstring competitors? If they can force stricter regulations that they know/suspect competitors cannot meet, they could prevent these competitors from selling cars with their 'FSD' tech? This would level the playing field, allowing the German car manufacturers to catch up, or at least hold off competition for a while.
“If the car crashes while Drive Pilot is operating, that's Mercedes' problem, not yours.”
It’s still my problem if I’m inside the car whilst it crashes.
It’s still my problem if I’m inside the car whilst it crashes.
It was/is pretty clear that german car makers are going to supersede the cult of Tesla. Better engineering, history, support, network, trust and an actual taste for design. Bonus: Leadership teams that don't post gibberish about Crypto.
This doesn't seem unreasonable.
If you consider that it is the driver who is liable today for road traffic incidents (including failing to adapt to environmental conditions)... then why wouldn't the "driver" still be liable when the driver changes?
https://www.synopsys.com/automotive/autonomous-driving-level...
In this it's plain that Level 3 autonomy is the first level where it's considered that the car is the driver. This is what Mercedes are proposing, to accept that the sum of software and hardware that they provide at Level 3, assuming it's used in the situations and conditions described, is the driver and that they are liable.
Those caveats seem reflected in what they're proposing (if the article is accurate), and it seems reasonable.
If you consider that it is the driver who is liable today for road traffic incidents (including failing to adapt to environmental conditions)... then why wouldn't the "driver" still be liable when the driver changes?
https://www.synopsys.com/automotive/autonomous-driving-level...
In this it's plain that Level 3 autonomy is the first level where it's considered that the car is the driver. This is what Mercedes are proposing, to accept that the sum of software and hardware that they provide at Level 3, assuming it's used in the situations and conditions described, is the driver and that they are liable.
Those caveats seem reflected in what they're proposing (if the article is accurate), and it seems reasonable.
"If the car crashes while Drive Pilot is operating, that's Mercedes' problem, not yours."
Only legally speaking. If the car drives me into an incoming truck, it still is very much my problem.
So currently it is only enabled for very clear situations, driving under 40 mp/h(60 kmh) in trafic jams on a motorway (where there usually is a physical separation to the other lane, except in roadbuilding situations, but I would guess, that the autopilot will be disabled then).
So it should be quite safe. Otherwise they would not accept the risk. German car makers are very conservative.
Edit: correct speedlimit
Only legally speaking. If the car drives me into an incoming truck, it still is very much my problem.
So currently it is only enabled for very clear situations, driving under 40 mp/h(60 kmh) in trafic jams on a motorway (where there usually is a physical separation to the other lane, except in roadbuilding situations, but I would guess, that the autopilot will be disabled then).
So it should be quite safe. Otherwise they would not accept the risk. German car makers are very conservative.
Edit: correct speedlimit
Whatever Mercedes says may not be what the law says. It may not be possible for them to take the liability from you. It may be criminal negligence or something else to not operate a car on a road in a state that says you have to.
Maybe they can agree to pay any civil liability you'd have, but taking a nap after you turn a car on may open you to criminal liability that Mercedes can't intervene in.
Maybe they can agree to pay any civil liability you'd have, but taking a nap after you turn a car on may open you to criminal liability that Mercedes can't intervene in.
Putting your money where your mouth is, is probably the only way to obtain acceptance for semi-autonomous driving systems.
I imagine airplane manufacturers have long been on the hook if their autopilot engages Kill All Humans mode, despite the best efforts of Boeing's lawyers to blame some of the humans their software killed.
I imagine airplane manufacturers have long been on the hook if their autopilot engages Kill All Humans mode, despite the best efforts of Boeing's lawyers to blame some of the humans their software killed.
> you are no longer legally liable for the car's operation until it disengages
Until it disengages. Not until you disengage it. So close, yet so far to go.
Unfortunately since this isn't a full self driving I don't know if they can do much better. If it hits the end of the road it needs to hand control back.
Until it disengages. Not until you disengage it. So close, yet so far to go.
Unfortunately since this isn't a full self driving I don't know if they can do much better. If it hits the end of the road it needs to hand control back.
Actually that is exactly the right thing to do! If something goes wrong when Autopilot is on, I think the manufacturer have to prove that it was a human error and not otherwise.
Self-driving cars will be popularized by people not having to pay (increasingly expensive) accident insurance. Once human driving becomes prohibitively expensive, it'll be reserved for the 'elite' who get a kick out of driving a car themselves.
We can't repair our cars, we have no privacy in when and where do we drive and now we don't even drive them nor are responsible for them. Essentially the car manufacturer plays a role of a transportation service provider now.
Ok, I'm not insisting this is bad. But now what's the point of buying a car into your ownership instead of just calling it like a taxi when and where you need it?
Ok, I'm not insisting this is bad. But now what's the point of buying a car into your ownership instead of just calling it like a taxi when and where you need it?
I used to work on autonomous cars back in the early 2000s. Legal responsibility was considered one of the biggest limiting factor to autonomous cars.
Bold move.
Bold move.
I never understood why others are not required to do so. Even when you stand outside your Tesla and "summon" it to you and it drives into something; you're somehow legally the driver?
Neat, but the article seems a little rosy-eyed. eg:
...the software uses microphones and cameras to detect emergency lights and sirens far enough in advance to issue the full 10-second warning before manual takeover.
That doesn't help when the lights and sirens are switched on a few dozen cars away and reach you in 6 seconds.
...the software uses microphones and cameras to detect emergency lights and sirens far enough in advance to issue the full 10-second warning before manual takeover.
That doesn't help when the lights and sirens are switched on a few dozen cars away and reach you in 6 seconds.
Maybe I'm a cynic, but I'm pretty sure someone at Mercedes just did the math, and concluded that the number they're expecting to pay in damages is a smaller number than the profit from the extra sales.
Nothing wrong with that as it is, but while it can be seen as a signal that "Mercedes is super sure their cars won't kill anyone" it should be seen more as a signal that "Mercedes is pretty sure they'll make a profit even when subtracting the price of killing people."
Nothing wrong with that as it is, but while it can be seen as a signal that "Mercedes is super sure their cars won't kill anyone" it should be seen more as a signal that "Mercedes is pretty sure they'll make a profit even when subtracting the price of killing people."
I love this. I genuinely do. But I can't help but imagine some dystopian future where they regret making this decision and they start fighting legal cases on the technicalities like, oh they crashed after they disengaged self driving, so it's the humans fault. Or even worse, they program in self driving to disengage when a crash is imminent. When companies take on liabilities, they're very unlikely to actually pay for it. Leaving the common folk to struggle.
I'm more concerned about "gray areas" where I'm driving but a proprietary computer system in my car have the real control so at a certain point in time it might decide to break or steer or refuse to for some programmed reasons I do not know.
Just imaging actual emergency braking systems: you see an accidents and you decide that the best option is going against another vehicle or even a wall for instance to avoid crushing on a school group on a trip. Actual system do not and can't know that, they'll try to stop your emergency maneuver and due to the little time you have even if formally deactivate you can't deactivate them. Who will be responsible and how to prove what really happen?
That's the issue of automation. I can't trust back box logs from a proprietary black box designed, built and installed by the same company who sell the car. Who can? Even public authorities how much practical resources can have to really investigate?
Just imaging actual emergency braking systems: you see an accidents and you decide that the best option is going against another vehicle or even a wall for instance to avoid crushing on a school group on a trip. Actual system do not and can't know that, they'll try to stop your emergency maneuver and due to the little time you have even if formally deactivate you can't deactivate them. Who will be responsible and how to prove what really happen?
That's the issue of automation. I can't trust back box logs from a proprietary black box designed, built and installed by the same company who sell the car. Who can? Even public authorities how much practical resources can have to really investigate?
> Right now, Drive Pilot can only engage at speeds under 40 mph...
Ah, so maybe this would be great for getting around town.
> ... on limited-access divided highways with no stoplights, roundabouts, or other traffic control systems...
Oh, so not where the majority of driving happens.
> ... and no construction zones.
And none of Pennsylvania.
Ah, so maybe this would be great for getting around town.
> ... on limited-access divided highways with no stoplights, roundabouts, or other traffic control systems...
Oh, so not where the majority of driving happens.
> ... and no construction zones.
And none of Pennsylvania.
I see this as a major signal to the outside world that Mercedes feels confident in its technology.
Mercedes must really think it has a better system than its competitors.
Either that or these expenses are going to be considered part of its marketing spend?
Another thought: isn't it still usually the responsibility of the at-fault party's insurance company to pay out for crashes anyway? Or, in no-fault states, each insurance company covers their own driver and passengers, so it's not really important who cases a crash.
I wonder if this announcement is legally meaningless in most cases...but I'm not expert in this area.
Mercedes must really think it has a better system than its competitors.
Either that or these expenses are going to be considered part of its marketing spend?
Another thought: isn't it still usually the responsibility of the at-fault party's insurance company to pay out for crashes anyway? Or, in no-fault states, each insurance company covers their own driver and passengers, so it's not really important who cases a crash.
I wonder if this announcement is legally meaningless in most cases...but I'm not expert in this area.
I was imagining a play like this utilizing insurance to smooth out the wrinkles. Seems like a fairly good bet that the restrictions are designed to essentially eliminate fault in the event of an accident.
This really sets a new standard in expectations around how safe and reliable these products are. TSLA probably has a significant head-start in terms of technology - but here's a new (call it marketing) feature by Mercedes that actually leap frogs over Tesla in terms of perception on the product.
I think in 3-5 years everyone else will have caught up to tesla and we'll have a more level playing field, mainly in terms of the technology and AI. And "marketing" features like this help accelerate the closing of the gap.
I think in 3-5 years everyone else will have caught up to tesla and we'll have a more level playing field, mainly in terms of the technology and AI. And "marketing" features like this help accelerate the closing of the gap.
> If the car crashes while Drive Pilot is operating, that's Mercedes' problem, not yours.
And the person you hit. Unbelievable that we value these metal boxes on wheels more than human bodies.
And the person you hit. Unbelievable that we value these metal boxes on wheels more than human bodies.
Civil liability is one thing, but what about criminal liability?
If they go through with this, awesome! I expect that they will walk some of this back before the release, though. (whether through extreme limitations on the situations they will take responsibility for, delaying release, extremely limited volume, etc.)
That said, every material step forward is a great thing. If we started out in the autonomous world that we're going to end up in, the idea of humans getting behind the wheel of a vehicle would seem incredibly crazy.
That said, every material step forward is a great thing. If we started out in the autonomous world that we're going to end up in, the idea of humans getting behind the wheel of a vehicle would seem incredibly crazy.
The next question is how Lloyd's of London will handle the insurance risk at scale. Clearly, Mercedes is going to need one heck of an insurance contract to justify the policy. What happens if the system is hacked and the number of payouts required reaches stratospheric proportions? What quantity of secondary insurance will be needed for the insurer?
Fascinating questions.
Fascinating questions.
While I don't agree with Tesla's strategy of playing it fast & loose, I'm afraid that risk aversion will lead us to demand perfect instead of the good enough.
A car manufacturer might not be willing to accept 100% liability, but their system might still be a lot safer than a drunk driver returning home from the bar.
Realistically we're not going to be able to pick "well, don't drive drunk then!", it'll happen. By demanding that these system are that reliable before they're in common operation we'll have incurred a net loss in terms of safety.
I don't know what the right solution is, probably some mixture of the two. Clearly you don't want manufacturers to be incentivized to ignore the risk associated with their systems, but you also don't want these systems not to be offered when they'd be a net gain in safety.
The point at which they're a net gain is not the same point that they become financially viable to the manufacturer from a liability point of view.