Thanks Autopilot: Cops stop Tesla whose driver appears asleep and drunk

This is undoubtedly the case here.

BAC is not usually reported in µg/L, and if it were, 340 µg/L equates to a BAC of 0.00%.

BrAC is often reported in µg/L and a BrAC of 340 µg/L equates to BAC of 0.07%, as pointed out by plantagenet. This is below the legal limit in every US state other than Utah (although above the legal limit for everywhere in Europe other than England, Wales, and Northern Ireland).

Plug the numbers in here and see for yourself if you like.

In the US police use linear measurement breathalyzer units that are calibrated for a blood-to-breath alcohol ratio 2,100:1, which assumes that there are 2,100 milliliters of alcohol in the blood for every 1 milliliter of alcohol in the breath. In other words, on average, the alcohol content of 1 ml of blood is 2,100 milliliters higher than that of 1 ml of breath.

In Canada police use linear measurement breathalyzer units that are calibrated for a blood-to-breath alcohol ratio 2,300:1, which assumes that there are 2,300 milliliters of alcohol in the blood for every 1 milliliter of alcohol in the breath. In other words, on average, the alcohol content of 1 ml of blood is 2,300 milliliters higher than that of 1 ml of breath.

In the United States minimum BAC limits are; .08 % (a federal limit) or higher except in Utah where its .05 % or higher - Drivers under 21 its 0.01% to 0.05% (varies by state) - Per Federal Motor Carrier Safety Administration = 0.04% for drivers of a commercial vehicle requiring a commercial driver's license and 0.01% for operators of common carriers (e.g. buses).

The higher the ratio the lower the breath alcohol level. Or to put it another way, for the same breath sample, an instrument calibrated at 2,100:1 will give a lower blood alcohol reading than one calibrated at 2,300:1.

Police in the Netherlands use linear measurement breathalyzer units that are generally calibrated 1.40 milligrams per litre [mg/l BrAC] = 0.32% BAC at the top end and not calibrated directly on a ratio like US and Canadian police units. This may not be the same for all the Neitherlands - police in different areas may use units with different calibration factors and the blood-to-breath alcohol ratio can be 2,300:1, 2,200:1, 2,100:1 or 2,000:1 depending on the police usage in the area. The ratio for these are set after calibration according to the area in which the unit is used in Europe including the Netherlands, unlike US and Canadian police units which use a single standard ratio across the country. The most common unit is the Alcocal (made in the UK) which is common with police across Europe since about 2010, but there are other units in use now also. This is a chart of the Alcocal breath:blood conversion. I use this chart because its the same conversion for all breathalyzers in use by police across Europe including the Netherlands.



Calibration of breathalyzer units is also affected by altitude and a conversion factor is figured into the calibration to account for altitude when necessary. So, for example, a breathalyzer used in Denver would no longer be calibrated correctly if it was used in Florida and would need to be re-calibrated.

BrAC = breath alcohol content
BAC = blood alcohol content

The measure in the Norway case is 340 ug/L which is measured at micrograms of alcohol (in blood - BAC) per 100 milliliters of breath (ug/100ml)

at a 2,300:1 ratio 340ug/L = 78 ‰ (ug/100ml) = 0.078 % BAC (note: also the ratio for Canadian Police breathalyzers)
at a 2,200:1 ratio 340ug/L = 74 ‰ (ug/100ml) = 0.074 % BAC
at a 2,100:1 ratio 340ug/L = 71 ‰ (ug/100ml) = 0.071 % BAC (note: also the ratio for US Police breathalyzers)
at a 2,000:1 ratio 340ug/L = 68 ‰ (ug/100ml) = 0.068 % BAC

Blood alcohol is measured 'weight by volume' (i.e. weight of alcohol per unit volume of blood) mostly. However, in some European countries its measured ‘weight by weight’ (i.e. weight of alcohol per unit weight of blood).

The weight/weight system (w/w) is used in countries which also use promille units. In reference to the chart above, using promille units: To convert from ‰ w/v to ‰ w/w the specific gravity of blood, which is about 1.06, must be used. So a 1.00 ‰ w/w blood sample is about 6% stronger in alcohol than a 1.00 ‰ w/v sample.

There ya go, you have all the information you need to convert from the Netherlands 340ug/L BrAC to US or Canadian BAC measurements. Now solve for X.

I don't know what Tesla does when the driver doesn't pay attention, but I've got the latest VW system, and it's pretty aggressive. I once did a test to see what would happen if I didn't take the wheel when prompted. It goes like that:

1. After 10 seconds: soft beep + the dash shows "please take over steering."
2. 5 seconds later: another soft beep+ the dash changes to "Take over steering!" (Note the lack of "please" and the addition of an exclamation mark. I found that little touch funny)
3. Another 5 seconds later: a really loud, long and angry beep
4. Another 5 seconds later: another angry beep + the car jerks itself by quickly applying and releasing the brakes + the seatbelt is very abruptly tightened.
5. I have no clue what happens then, because I didn't dare to continue the experiment. The last step is freaking scary and would definitely wake me up if I was sleeping (and give me an extra adrenaline boost that would keep me awake). (According to the manual, it's supposed to do the car on the shoulder and call emergency services, but I'm not going to test that)

Making it possible for a driver to stay asleep at the wheel is irresponsible. I'm not sure how Tesla is getting away with it.

I wish Tesla and all mfrs of L2 assist systems would release their actual data. That would let us see what difference if any the systems make wrt safety. It's clear that people like the L2 systems, as demonstrated in these comment threads, but it's not at all clear whether they improve or decrease safety.

What we get instead is comparisons between different scenarios: when Tesla claims there are fewer accidents with AP enabled than without, they're conveniently leaving out the driving context. It seems reasonable to expect that AP is enabled in general in safer situations (e.g. freeway driving) and not used in situations where accidents are more likely. So, yeah, of course, there are fewer accidents with AP enabled than when it's not enabled.

GM's Super Cruise can only be used in those safe situations. So, if GM said that there were fewer accidents with Super Cruise than without, we'd all say, gosh, isn't that amazing? In the safe (well mapped, controlled access) contexts where you let Super Cruise be used, there are fewer accidents than in all the less-safe contexts where you don't let it be used.

The relevant comparison is like-to-like. Is it safer on a controlled access, multi-lane highway with or without AP or Super Cruise? Does AP/SC do better or worse when traffic is heavier? Is it safer on a 2 lane highway without controlled access with or without AP? Is it safer on a city street with or without AP? Do drivers with AP enabled drive faster or slower than drivers with AP disengaged?

Tesla has this data and has either done the analysis and not shared it or hasn't done the analysis. The fact that they don't share this but instead hype meaningless comparisons strongly suggests that AP is not significantly safer to use. That doesn't mean it doesn't improve the driving experience, but it does suggest that such improvements don't result in real safety improvements.

In looking at this data, I'd be happy to filter out the suicidal idiots and just look at "responsible" use of driver assist tech. The suicidal idiots will hopefully continue to just have single-car and single-person fatalities.

This isn't just for Tesla: I'd really like to see comparable info from any car mfr that releases an L2 driver assist package.

Here is something incontrovertible; 100% of drunks who fall asleep behind the wheel of a moving car for a sufficient amount of time, will stop violently - until now. The difference here is that Autopilot gave this man, and the people around him, a fighting chance of survival for long enough for the police to intervene.

Should autopilot have stopped this guy earlier? Yes. Is it designed to? Yes.

Tesla uses a torque sensor to detect hands on wheel. If it doesn’t, it issues a series of escalating alerts before stopping the vehicle and calling emergency services. The fact that the editor has chosen to promote a comment that questions this, comparing it negatively to VW which also 1) uses torque sensors to detect hands on wheel 2) issues a series of escalating alerts before stopping and calling, is ridiculous. The editor either knows, or reasonably should have known, that this is EXACTLY what the Tesla does. Why didn’t it stop sooner? The guy’s hands were still on the wheel when he fell asleep, defeating the abort cascade. Exactly the same failure would have occurred in the VW. GM’s super-cruise, as far as I know, is the only similar system that would have aborted based on eye tracking instead - point GM.

‘So what, autopilot probably encouraged the guy to drive drunk.’

Based on what, exactly? People have been driving drunk since cars were invented, they have never needed an excuse that ‘the car can handle it’. They are filled with rationalizations, even if ‘my car has autopilot, so there’ were one of them, there is absolutely zero evidence that it has any significant contribution to that decision. ‘But the guy on the bridge told the police his car had autopilot...’ well add that to the long list of excuses drunk drivers have offered the police to (unsuccessfully) justify their behavior.

Here is the nail in the coffin for the argument ‘if he hadn’t had autopilot, he wouldn’t have made the decision to drive drunk’;

Here’s a dirty little secret; even if you ignore all the warnings on the internet, the warnings when you take delivery, the warnings in the manual, and click passed all the screen warnings when you engage autopilot, that say it’s not self driving and you need to be alert; absolutely no one who has driven on autopilot for more than 10 or 20 miles could reasonably be under the impression that, in it’s current form, it can drive itself. Sure, Elon has promised it will someday soon, but it is immediately obvious to anyone who has ever used it - soon is not today. It disengages all the time, makes stupid mistakes, and drives like an insecure teenager - no one who has used it could possibly believe it can drive you home from the bar.

‘But what about the Harry Potter guy, or the .... guy’. Yes, there are people who knowingly test the systems limits outside of its design domain. There are idiots doing stupid things in all walks of society, putting themselves and others at risk. Here’s the thing; the responsibility for that antisocial behavior is on them, not the system. The reality is that in the more than 1 billion miles driven on autopilot, the incidence of stupid people doing stupid things is on par with the incidence of stupid people doing stupid things in all walks of life, you just don’t hear about the billion plus miles of people not doing stupid things on news sites and blogs.

There are plenty of things to criticize about autopilot, but in this instance, it probably saved the guy’s life.

Sorry, but that is just factually incorrect. People have defeated the Tesla torque sensor by wedging a grapefruit in the steering wheel. The dead weight of an unconscious driver’s hands could absolutely defeat the abort sequence.

This is undoubtedly the case here.

BAC is not usually reported in µg/L, and if it were, 340 µg/L equates to a BAC of 0.00%.

BrAC is often reported in µg/L and a BrAC of 340 µg/L equates to BAC of 0.07%, as pointed out by plantagenet. This is below the legal limit in every US state other than Utah (although above the legal limit for everywhere in Europe other than England, Wales, and Northern Ireland).

Plug the numbers in here and see for yourself if you like.

In the US police use linear measurement breathalyzer units that are calibrated for a blood-to-breath alcohol ratio 2,100:1, which assumes that there are 2,100 milliliters of alcohol in the blood for every 1 milliliter of alcohol in the breath. In other words, on average, the alcohol content of 1 ml of blood is 2,100 milliliters higher than that of 1 ml of breath.

In Canada police use linear measurement breathalyzer units that are calibrated for a blood-to-breath alcohol ratio 2,300:1, which assumes that there are 2,300 milliliters of alcohol in the blood for every 1 milliliter of alcohol in the breath. In other words, on average, the alcohol content of 1 ml of blood is 2,300 milliliters higher than that of 1 ml of breath.

In the United States minimum BAC limits are; .08 % (a federal limit) or higher except in Utah where its .05 % or higher - Drivers under 21 its 0.01% to 0.05% (varies by state) - Per Federal Motor Carrier Safety Administration = 0.04% for drivers of a commercial vehicle requiring a commercial driver's license and 0.01% for operators of common carriers (e.g. buses).

The higher the ratio the lower the breath alcohol level. Or to put it another way, for the same breath sample, an instrument calibrated at 2,100:1 will give a lower blood alcohol reading than one calibrated at 2,300:1.

Police in the Netherlands use linear measurement breathalyzer units that are generally calibrated 1.40 milligrams per litre [mg/l BrAC] = 0.32% BAC at the top end and not calibrated directly on a ratio like US and Canadian police units. This may not be the same for all the Neitherlands - police in different areas may use units with different calibration factors and the blood-to-breath alcohol ratio can be 2,300:1, 2,200:1, 2,100:1 or 2,000:1 depending on the police usage in the area. The ratio for these are set after calibration according to the area in which the unit is used in Europe including the Netherlands, unlike US and Canadian police units which use a single standard ratio across the country. The most common unit is the Alcocal (made in the UK) which is common with police across Europe since about 2010, but there are other units in use now also. This is a chart of the Alcocal breath:blood conversion. I use this chart because its the same conversion for all breathalyzers in use by police across Europe including the Netherlands.



Calibration of breathalyzer units is also affected by altitude and a conversion factor is figured into the calibration to account for altitude when necessary. So, for example, a breathalyzer used in Denver would no longer be calibrated correctly if it was used in Florida and would need to be re-calibrated.

BrAC = breath alcohol content
BAC = blood alcohol content

The measure in the Norway case is 340 ug/L which is measured at micrograms of alcohol (in blood - BAC) per 100 milliliters of breath (ug/100ml)

at a 2,300:1 ratio 340ug/L = 78 ‰ (ug/100ml) = 0.078 % BAC (note: also the ratio for Canadian Police breathalyzers)
at a 2,200:1 ratio 340ug/L = 74 ‰ (ug/100ml) = 0.074 % BAC
at a 2,100:1 ratio 340ug/L = 71 ‰ (ug/100ml) = 0.071 % BAC (note: also the ratio for US Police breathalyzers)
at a 2,000:1 ratio 340ug/L = 68 ‰ (ug/100ml) = 0.068 % BAC

Blood alcohol is measured 'weight by volume' (i.e. weight of alcohol per unit volume of blood) mostly. However, in some European countries its measured ‘weight by weight’ (i.e. weight of alcohol per unit weight of blood).

The weight/weight system (w/w) is used in countries which also use promille units. In reference to the chart above, using promille units: To convert from ‰ w/v to ‰ w/w the specific gravity of blood, which is about 1.06, must be used. So a 1.00 ‰ w/w blood sample is about 6% stronger in alcohol than a 1.00 ‰ w/v sample.

There ya go, you have all the information you need to convert from the Netherlands 340ug/L BrAC to US or Canadian BAC measurements. Now solve for X.

I don't know what Tesla does when the driver doesn't pay attention, but I've got the latest VW system, and it's pretty aggressive. I once did a test to see what would happen if I didn't take the wheel when prompted. It goes like that:

1. After 10 seconds: soft beep + the dash shows "please take over steering."
2. 5 seconds later: another soft beep+ the dash changes to "Take over steering!" (Note the lack of "please" and the addition of an exclamation mark. I found that little touch funny)
3. Another 5 seconds later: a really loud, long and angry beep
4. Another 5 seconds later: another angry beep + the car jerks itself by quickly applying and releasing the brakes + the seatbelt is very abruptly tightened.
5. I have no clue what happens then, because I didn't dare to continue the experiment. The last step is freaking scary and would definitely wake me up if I was sleeping (and give me an extra adrenaline boost that would keep me awake). (According to the manual, it's supposed to do the car on the shoulder and call emergency services, but I'm not going to test that)

Making it possible for a driver to stay asleep at the wheel is irresponsible. I'm not sure how Tesla is getting away with it.

I wish Tesla and all mfrs of L2 assist systems would release their actual data. That would let us see what difference if any the systems make wrt safety. It's clear that people like the L2 systems, as demonstrated in these comment threads, but it's not at all clear whether they improve or decrease safety.

What we get instead is comparisons between different scenarios: when Tesla claims there are fewer accidents with AP enabled than without, they're conveniently leaving out the driving context. It seems reasonable to expect that AP is enabled in general in safer situations (e.g. freeway driving) and not used in situations where accidents are more likely. So, yeah, of course, there are fewer accidents with AP enabled than when it's not enabled.

GM's Super Cruise can only be used in those safe situations. So, if GM said that there were fewer accidents with Super Cruise than without, we'd all say, gosh, isn't that amazing? In the safe (well mapped, controlled access) contexts where you let Super Cruise be used, there are fewer accidents than in all the less-safe contexts where you don't let it be used.

The relevant comparison is like-to-like. Is it safer on a controlled access, multi-lane highway with or without AP or Super Cruise? Does AP/SC do better or worse when traffic is heavier? Is it safer on a 2 lane highway without controlled access with or without AP? Is it safer on a city street with or without AP? Do drivers with AP enabled drive faster or slower than drivers with AP disengaged?

Tesla has this data and has either done the analysis and not shared it or hasn't done the analysis. The fact that they don't share this but instead hype meaningless comparisons strongly suggests that AP is not significantly safer to use. That doesn't mean it doesn't improve the driving experience, but it does suggest that such improvements don't result in real safety improvements.

In looking at this data, I'd be happy to filter out the suicidal idiots and just look at "responsible" use of driver assist tech. The suicidal idiots will hopefully continue to just have single-car and single-person fatalities.

This isn't just for Tesla: I'd really like to see comparable info from any car mfr that releases an L2 driver assist package.

Here is something incontrovertible; 100% of drunks who fall asleep behind the wheel of a moving car for a sufficient amount of time, will stop violently - until now. The difference here is that Autopilot gave this man, and the people around him, a fighting chance of survival for long enough for the police to intervene.

Should autopilot have stopped this guy earlier? Yes. Is it designed to? Yes.

Tesla uses a torque sensor to detect hands on wheel. If it doesn’t, it issues a series of escalating alerts before stopping the vehicle and calling emergency services. The fact that the editor has chosen to promote a comment that questions this, comparing it negatively to VW which also 1) uses torque sensors to detect hands on wheel 2) issues a series of escalating alerts before stopping and calling, is ridiculous. The editor either knows, or reasonably should have known, that this is EXACTLY what the Tesla does. Why didn’t it stop sooner? The guy’s hands were still on the wheel when he fell asleep, defeating the abort cascade. Exactly the same failure would have occurred in the VW. GM’s super-cruise, as far as I know, is the only similar system that would have aborted based on eye tracking instead - point GM.

‘So what, autopilot probably encouraged the guy to drive drunk.’

Based on what, exactly? People have been driving drunk since cars were invented, they have never needed an excuse that ‘the car can handle it’. They are filled with rationalizations, even if ‘my car has autopilot, so there’ were one of them, there is absolutely zero evidence that it has any significant contribution to that decision. ‘But the guy on the bridge told the police his car had autopilot...’ well add that to the long list of excuses drunk drivers have offered the police to (unsuccessfully) justify their behavior.

Here is the nail in the coffin for the argument ‘if he hadn’t had autopilot, he wouldn’t have made the decision to drive drunk’;

Here’s a dirty little secret; even if you ignore all the warnings on the internet, the warnings when you take delivery, the warnings in the manual, and click passed all the screen warnings when you engage autopilot, that say it’s not self driving and you need to be alert; absolutely no one who has driven on autopilot for more than 10 or 20 miles could reasonably be under the impression that, in it’s current form, it can drive itself. Sure, Elon has promised it will someday soon, but it is immediately obvious to anyone who has ever used it - soon is not today. It disengages all the time, makes stupid mistakes, and drives like an insecure teenager - no one who has used it could possibly believe it can drive you home from the bar.

‘But what about the Harry Potter guy, or the .... guy’. Yes, there are people who knowingly test the systems limits outside of its design domain. There are idiots doing stupid things in all walks of society, putting themselves and others at risk. Here’s the thing; the responsibility for that antisocial behavior is on them, not the system. The reality is that in the more than 1 billion miles driven on autopilot, the incidence of stupid people doing stupid things is on par with the incidence of stupid people doing stupid things in all walks of life, you just don’t hear about the billion plus miles of people not doing stupid things on news sites and blogs.

There are plenty of things to criticize about autopilot, but in this instance, it probably saved the guy’s life.

Sorry, but that is just factually incorrect. People have defeated the Tesla torque sensor by wedging a grapefruit in the steering wheel. The dead weight of an unconscious driver’s hands could absolutely defeat the abort sequence.