Does auotpilot present a looming apocalypse for Tesla?

[OPRCE]

Now that the $35K standard model 3 has been introduced and is selling to the non fanboy general public will Tesla soon be facing lawsuits regarding the so called autopilot? I think it could happen.

Tesla is already facing several lawsuits regarding those killed by AP running into stationary objects at highway speeds, and yes, as the fleet size increases these incidents will in all likelihood rise proportionately.

Tesla will have to spend time and money defending itself against what we all know will be frivolous lawsuits.

We don't know this at all and the first one to succeed will kick loose either a recall or an avalanche of further cases.

For example, in the case of Walter Huang, whose AP2.5 Model X went into the gore-point at Mountain View last May, imagine a municipal worker had been repairing the pre-collapsed crash-attenuator at that moment and also been killed, his wife would IMHO have a fairly robust case against Tesla for its contributory negligence leading to the wrongful death, because:

1. The completely blameless 3rd party victim did not agree to participate in any Beta AP testing.
2. Tesla cannot say what level of attention was paid by deceased driver, as nothing actually measures that, just net torque on the steering. Huang could have had both hands balanced on the wheel with nothing detected, while being temporarily blinded by the low morning sun. i.e. he may have been paying reasonable attention but simply failed to brake due to over-trusting AP to make correct decisions.
3. AP followed the wrong line while accelerating across an area where it is not normally permissible to drive, straight into the GP, with no warning tones.
4. Tesla had 28 months since the first AP fatality to fix this known hazard [into stopped objects at highway speed] yet did nothing effective to resolve or safely mitigate it.
5. Musk over the same time greatly equivocated in public about the superior abilities of AP/FSD, tending to mislead the technically naïve as to the current capabilities in their vehicles.
6. Lawyers will preferentially pursue the defendant with deeper pockets, in this case Tesla/Musk.
7. If the plaintiff refuses to settle out of court under NDA then there is ~60% chance of convincing a jury that in this case Tesla's negligence amounts to a >=30% contribution to the wrongful death.
8. Given the current state of HW2.5 & v9 software, it is only a question of time until an analogous case presents in reality.

 

[5_+JqckQttqck]

I don't think so - power tools can decapitate and killed when used unsafely. Same thing with a super power tool such as a car. Unless it's like a air bag failure as the Takata case where people died due to the corporation's incompetence.

Tesla is in the clear with the EULA - of course - if a third party was able to hack the AP system and cause it to fail repeatedly while Tesla knew about it and kept it under the rug; it'd be another story. See diesel gate

 

[OPRCE]

Apart from the distinct possibility of successful retrospective legal action, Tesla may also face some unforeseen hurdles in getting its HW3&FSD on the current sensor suite approved for >=L3 use in certain jurisdictions:

1. The EU has recently mandated AEB in all new cars sold from 2022 and it is imaginable that this directive will be transposed into the domestic law of some member states in such a way that a forward LiDAR sensor is made obligatory to maximise the safety of the function at all operating speeds.

2. The OEM Auto-cartels of Germany for instance can casually achieve tossing this regulatory spanner in the Tesla gearbox, by ganging up to murmur sweetly in Mutti Merkel's ear.

3. The same rule would likely be made applicable to all >=L3 AVs.

4. There will most likely also be a standardised AV Driving Test developed, in which candidate vehicles will have to successfully handle a variety of common tasks/hazard scenarios to gain approval for general public use as >=L3.

5. I have a strong feeling that mere statistics drenched in Musk are never going to cut the mustard here. For a start those stats will have been gleaned from the FSD operating under a L2 regime, during which beta period most crashes will (one hopes) be avoided by the attentive driver having intervened to override the controls, in which case it will not be in FSD, thus heavily skewing the results to favour Tesla's purposes.

6. Musk has projected "FSD ready for sleeping in the back" by Jan 2021, but I would be shocked if L3 approval happens anywhere before Jan 2023.

 

[Daniel in SD]

4. There will most likely also be a standardised AV Driving Test developed, in which candidate vehicles will have to successfully handle a variety of common tasks/hazard scenarios to gain approval for general public use as >=L3.

Any good autonomous vehicle could easily pass such a test today. I wouldn’t be surprised if Tesla could pass such a test later this year. No test can ensure that self driving vehicles will have safety on par with a human driver. The only way to ensure safety is to test on actual public roads with a test driver like many companies are doing in California.
Tesla could easily hire enough test drivers to get the system approved in California. I think once the system proves to be safe it wouldn’t be hard to get it approved in other jurisdictions.

 

[OPRCE]

I wouldn’t be surprised if Tesla could pass such a test later this year.

I'd be very surprised if it could reliably pass the kind of test I am thinking of, which would include a 90mph cut-out into stationary traffic @150m, amongst a wide array of other hazard scenarios to be worked through over the space of a month in the hands of an independent safety institute.

In any case though am very willing to be astounded! The foretaste on Apr.19th should indicate how likely that is.

 

[Big Earl]

Now that the $35K standard model 3 has been introduced and is selling to the non fanboy general public will Tesla soon be facing lawsuits regarding the so called autopilot? I think it could happen.

Are you Lenny?

 

[Daniel in SD]

I'd be very surprised if it could reliably pass the kind of test I am thinking of, which would include a 90mph cut-out into stationary traffic @150m, amongst a wide array of other hazard scenarios to be worked through over the space of a month in the hands of an independent safety institute.

In any case though am very willing to be astounded! The foretaste on Apr.19th should indicate how likely that is.

Now I'm curious what a 90mph cut-out into stationary traffic is. I guess I should have qualified it is that Tesla could likely pass a test if the testing procedure was known to them.
Humans get into accidents about every 150k miles, seems much simpler and more accurate to drive 10 million miles on a variety of roads and conditions and see how well the system works than to try to make an artificial test.

 

[OPRCE]

Now I'm curious what a 90mph cut-out into stationary traffic is

You are in AP @90mph behind a leading vehicle which cuts out [moves aside] to reveal a vehicle parked 150m ahead in your lane. [see link in my sig for example]

As far as I know to date no Tesla passes this test at >=60mph.
Have personally seen it work at 50mph but still do not feel it is reliable.

Here is a recent test using 2019.5.15 with a M3 @72mph:
Reminder: Current AP is sometimes blind to stopped cars : teslamotors

 

[CanadaCarCare]

I think so, just reading the other Tesla forums and once a week you have someone posting something like "Accident while AP engaged"

Those not familiar with how AP works and don't read the manual that says YOU MUST BE PAYING ATTENTION AT ALL TIMES will think this system does more than it should. Once the cheaper cars hit the road and the Model 3 is the new civic, we will see more of these kind of accidents where people are not paying attention and assuming AP will be safe.

 

[Vines]

You are in AP @90mph behind a leading vehicle which cuts out [moves aside] to reveal a vehicle parked 150m ahead in your lane. [see link in my sig for example]

As far as I know to date no Tesla passes this test at >=60mph.
Have personally seen it work at 50mph but still do not feel it is reliable.

Here is a recent test using 2019.5.15 with a M3 @72mph:
Reminder: Current AP is sometimes blind to stopped cars : teslamotors

90 mph is a pretty high bar to set. What sort of braking distance (ignoring reaction and thinking distance) do you assume is required for the average car at 90 mph?

 

[OPRCE]

90 mph is a pretty high bar to set. What sort of braking distance (ignoring reaction and thinking distance) do you assume is required for the average car at 90 mph?

I worked out an estimation for the LR M3 lately and IIRC it was about 91m at 80mph, but anyhow let's just say the distance to occluded stationary target is 10% more than the minimum braking distance for whatever vehicle is being tested at a given speed.

In my own personal experience with S100D [sig link] the stationary jam was about 200m ahead and the car was still accelerating into it when I hopped on the anchors pretty sharply.

90mph is the speed AP operates up to, so it not too high a bar to expect it must safely pass this test.

 

[AlanSubie4Life]

90 mph is a pretty high bar to set. What sort of braking distance (ignoring reaction and thinking distance) do you assume is required for the average car at 90 mph?

At 90mph with 1 second reaction time it's ~126 meters with MXM4s. 125ft*(90^2/60^2) + 1sec * 90mph.

 

[Daniel in SD]

90 mph is a pretty high bar to set. What sort of braking distance (ignoring reaction and thinking distance) do you assume is required for the average car at 90 mph?

I think a Model 3 with MXM4s should be able to stop from 90mph in about 280 feet not counting reaction time.
I’m assuming that FSD will obey all speed limits so 90mph seems a bit far fetched. Obviously any FSD system will have to see stationary objects.

 

[OPRCE]

I think a Model 3 with MXM4s should be able to stop from 90mph in about 280 feet not counting reaction time.
I’m assuming that FSD will obey all speed limits so 90mph seems a bit far fetched. Obviously any FSD system will have to see stationary objects.

In Germany 90 would be considered an annoyingly understated speed for the 3rd lane on an Autobahn, where 125mph (200kph) is more like the going rate.

These days owners are testing NoA there with v.2019.8.3 and I gather it is a pretty hairy experience, as I expected.

 

[caskater47]

Lawsuits involving wrongful death will happen regardless of EULA. No matter good Tesla's lawyers are a EULA isn't a real defense against such a case.

The real question I think is whether Tesla will be sued over false advertising, false claims and fraud over the technology itself. They are selling vaporware to customers and if they can't deliver exactly what Elon keeps promising Tesla may have a hard time justifying themselves. FSD isn't a kickstarter campaign or other type of risky "investment" that can normally be shielded from lawsuits. They are actively taking money, promising a lot and so far have delivered nothing. Eventually that will catch up to them.

 

[Daniel in SD]

In Germany 90 would be considered an annoyingly understated speed for the 3rd lane on an Autobahn, where 125mph (200kph) is more like the going rate.

These days owners are testing NoA there with v.2019.8.3 and I gather it is a pretty hairy experience, as I expected.

At 125mph a Model 3 would have about 75 miles of range. Haha. I think most people would be happy if FSD could drive at the traffic speed in the right lane of the autobahn.

 

[AlanSubie4Life]

In Germany 90 would be considered an annoyingly understated speed for the 3rd lane on an Autobahn, where 125mph (200kph) is more like the going rate.

These days owners are testing NoA there with v.2019.8.3 and I gather it is a pretty hairy experience, as I expected.

Everyone has a death wish, I guess.

I'm all for stress testing systems, but personally when traveling at 125mph, I'll trust myself with the driving rather than a computer. But mostly I probably wouldn't drive at 125mph because I'm a wimp. And I agree a system working at 125mph would be awesome and increase safety at the lower speeds, possibly. But perhaps it is best to have some functionality tested there, but make AP inaccessible to all users at that speed (I thought it already was prohibited at those speeds).

 

[OPRCE]

At 125mph a Model 3 would have about 75 miles of range. Haha. I think most people would be happy if FSD could drive at the traffic speed in the right lane of the autobahn.

That's still probably about 10 miles more than my S100D

But seriously, am not suggesting AP/FSD should handle 125mph but 90mph in AP is within the domain of the legally possible there, so people will do it, hence Tesla has to ensure it works safely.

 

[Vines]

Data about 90 mph stopping distances vary wildly. Link below seems to think that just braking alone on a dry road is 123M, with 1 second thinking distance it calculates 150m exactly. Stopping on a wet road would not happen in those conditions I think.
Car Stopping Distance Calculator

The presumption that Tesla will allow FSD up to the same 90 mph limit as AP is questionable. I imagine FSD will be limited to legal speeds, or something like 70-80 mph. Granted there might be a couple countries where this is a limitation, but mostly this is a reasonable limit. If you want to faster, engage AP, and pay attention.

Legally AP requires you to pay attention and be in control, so designing a test that a human could just barely pass on a dry road and fail on a wet road is a bit of a stretch to expect the computer to beat.

Or you might stay in the right lanes of the Autobahn. I find myself speeding a lot less with EAP, since the driving isn't so bad, there is no need to rush to my destination.

 

[OPRCE]

Everyone has a death wish, I guess.

I'm all for stress testing systems, but personally when traveling at 125mph, I'll trust myself with the driving rather than a computer. But mostly I probably wouldn't drive at 125mph because I'm a wimp. And I agree a system working at 125mph would be awesome and increase safety at the lower speeds, possibly. But perhaps it is best to have some functionality tested there, but make it inaccessible to all users.

Sorry if unclear, but AP in Germany currently only works up to 93mph [150kph], hence NoA is driving at that max speed, though when changing lanes on AB will have to contend with others closing from behind at any speed possible, which with raw software, HW2.5 APE and no rear corner radars I am sure is quite trying for the nerves.

I have managed 255kph [158.4mph] on the German AB at a very quiet time, and got the speedo numbers to light up red. Just for a one-off test though. Generally I do not exceed the AP limits and cruise at 125kph [78mph].