The feds have decided that Tesla's Autopilot technology needs to get better:
The issue here isn't how good Tesla's unassisted driving is. The issue is that Tesla is supposed to make sure drivers are paying attention regardless, typically by beeping and alerting when they take their hands off the wheel or their eyes wander. The NHTSA has concluded that Tesla needs to update their software to do better on this score.
That said, it's worth a note that Teslas aren't especially dangerous. Just the opposite. The NHTSA estimates that in a single year there are about 139 fatal crashes per million cars in the US. Among Teslas the number is only 64.
There's no telling why this is. Maybe Tesla drivers use Autopilot a lot and this cuts down on fatal crashes. Maybe Tesla drivers just tend to be safer than average. Or maybe Teslas log fewer miles per car than most models. The available data just doesn't let us know.
Nor do we know anything about the rate of Tesla fatalities with and without Autopilot engaged. That data should be available, I think, but Tesla has never produced it. However, they do make this claim:
In the 4th quarter of 2022, we recorded one crash for every 4.85 million miles driven in which drivers were using Autopilot technology. For drivers who were not using Autopilot technology, we recorded one crash for every 1.40 million miles driven. By comparison, the most recent data available from NHTSA and FHWA (from 2021) shows that in the United States there was an automobile crash approximately every 652,000 miles.
I would naturally prefer data produced independently, not by Tesla itself. For one thing, it's not entirely clear how they measure "using Autopilot technology." Still, if I had to guess, I'd say that their numbers are close to right. Teslas with autopilot engaged are probably safer than Teslas without autopilot and a lot safer than the average human-driven car.
I'm guessing Autopilot is used much more on freeways. There probably are fewer accidents per mile driven on freeways than in city driving. So I'm not sure the Tesla data shows that Autopilot is safer.
They normalize that, by counting 'miles autopilot was available' instead of merely activated, running it in shadow mode so it's ready all the time it can be.
They don't count miles when Autopilot didn't think it could run.
That doesn’t change the argument.
If they're comparing 'miles driven on the freeway' to 'miles driven on the freeway', that's apples to apples.
It means the 'most people use it on the freeway' a meaningless distinction.
Carissa WTF happened to you? 6-7 years ago you were a good commenter at Mother Jones responding to KD. Now when you aren’t shilling for Tesla you come across as a whack a doodle not getting the point when one can discern what you are commenting . Few people respond to you anymore and those that do like PM rue it. I don’t mean to be rude, but have you had a stroke or brain injury? You are no better at LGM.
They are not comparing “'miles driven on the freeway' to 'miles driven on the freeway', that's apples to apples,“ but please keep f’n that chicken.
Thumbs up, and not the Donald kind.
There has actually been a lot of work on this and Teslas are much less safe than equivalent cost cars driven by the same demographic of people. If you are trusting Teslas numbers you are doing this wrong. Also the Tesla “Autopilot” technology is much less sophisticated/safe than the standard Level 2 ASDS on any American, German, Japanese or Korean luxury car these days. Tesla just takes the safety off so people are confused and use it incorrectly.
Yes, this is why you didn't cite anything or any benchmarks.
As a Tesla owner, I don't understand how anyone could put their trust in Full Self Driving, it works decently enough on the freeway, though it still makes questionable decisions, takes corners too fast and is a menace if you let it handle lane changes or merges. Around town, regardless of what Elon and his fans claim it's useless. Anytime it's on hands are on the wheel and I'm paying attention. On the plus side I do find using it on road trips I arrive less tired and more relaxed
Well, first off, Full Self Driving Beta until this last year wasn't running on freeways. That was just the same Autopilot everyone else was using.
And not until the recent version was it taking road conditions into account - but no other ADAS system does that I know of,
I was referring to my experience with it this year, most recently when I rented it for a road trip and during the free trial.
I’m not going to trust auto pilot in my M3 when it can’t even get the auto windshield wipers to work right half the time.
You do realize seeing water drops is a completely different and more difficult problem than picking out lane linesm right?
Besides, the point isn't to trust it, the point is to use it to hold things steady while you can cast your awareness further down the road or keep more 360 awareness.
Wait a few months, it'll act differently.
The Elon mantra is dutifully repeated.
Just wait, it will work later!
The cult is strong.
I have been using the latest FSD all month in my M3. My prior experience with autopilot on highways was ok, but it sometimes slowed down for no reason when going under a bridge. But FSD has been pretty close to perfect all month for normal driving situations that include merging on and off highways, changing lanes, stop signs and stop lights. Where it still needs a lot of work are situations where rules are different temporarily or periodically (school zones with low speed limit on weekdays when school is in session, reduced speed limit in a construction zone even if no construction is visible, signs at a stop light that say NO TURN ON RED weekdays 8am-9:30am, etc); toll booths - it hesitates when it approaches a toll booth and sometimes when exiting; and it has no idea what to do when approaching a gated entry to a community, campus, parking lot, hospital, etc where there is one lane for visitors and another for residents/employees/sticker holders. I live in a gated community and I had to disengage each time I approached the gate because the car was not going to use the lane that would open automatically for my car. And it really doesn’t like parking lots. FWIW, stopping at stop signs is often annoying because regulators forced Tesla to stop allowing the software to deal with stop signs the way good drivers do. Instead, the car comes to a full stop just before the sign, so of course it can’t always see what’s coming so it inches ahead until it has a clear view of the intersection, whereupon it stops again before proceding. But all in all, the thing performs way better than I ever thought it would in 2024. And now that it is end to end AI, improvements should happen more rapidly. Is it safe? It has a lot more eyes than I do, and I can now see it potentially getting to level 3 by year end. But I have no idea how it will perform in a snowstorm. And driverless robotaxis are another thing entirely.
There's no way to have this data from a third party, Kevin, because only Tesla can produce the data. That's a bit of a problem with any data about autonomy, and why Cruise got in big trouble when they weren't forthcoming after the accident.
Tesla claims they're normalizing the data by only comparing miles that Autopilot is being used to miles Autopilot thinks it would have worked - which eliminates any times when Autopilot would have errored out from being counted in the miles it's comparing against.
The more interesting thing is the current rate of Full Self Driving city streets Supervised is gaining miles - 16 million per day now.
I would guess that the percentage of drunk drivers is lower with Teslas than with cars in general. I have no data to support this.
No, well, not way to know.
But most of these autopilot-related collision investigations were drunk drivers.
We occasionally run into Teslas weaving around weirdly, and our term for those is "drunk Teslas." Once got trapped behind a Tesla weaving well into the opposite lane on a rural road. Our best guess was the fog made it hard for the autopilot to realize that the mailbox at the end of each driveway wasn't a threat. Or maybe it was just a drunk driver.
I tend to give Teslas a lot of room. You never know who's using autopilot inappropriately.
Unlikely that was Autopilot.
This is all because of a couple factors, none of which reflect badly on Tesla.
1. Only Teslas have a computer sophisticated enough to be a form of airliner "black box" which is why when you go back through these headline grabbing crashes it tends to turn out that the driver assist features weren't even on or had nothing to do with it. Noting what Crissa said, this same feature is the only reason Tesla even has data on how many miles are being driven in each "mode" no other manufacturer can even provide such data.
2. So, since Tesla is the only company with data, (no other company has a system advanced enough to "black box" when someone might be speeding on cruise control or whatever) Tesla is also the only company which has "Nags" based on both pressure on the wheel and now (in the past year or so) the activated interior camera which detects when you are not looking at the road.
Its therefore understandable that NTSA would investigate stuff, after all, that's thier job. Its not really understandable that it can be spun as a negative.
But spun as a negative it will be.
By the way, as I have enjoyed Kevin's blog for what feels like a couple of decades now, all of you non Tesla owners should all hit up a friend or figure out a way to driver a Tesla with current FSD version 12 available.
Then, I suspect you will go figure out whether you have any free cash to make a ten or hundred bagger on the stock. If you don't like Musk, take your millions in profits and donate to the thing Musk hates the most.
Not only is FSD coming, but with it will be a drop in accident rates of Teslas that will require every other car maker to license it.
Again, test drive one. I simply cannot believe the negative press. Its like having an iphone prototype a year in advance and ever day the media is saying it will never work, and you know it will work ahead of time.
If you want details I have five years of them, since I have had FSD since mid 2019.
The other brands level 2 ADAS systems do have nags, though!
I stand corrected on nags. What I was getting at was not only nags but when the nags are given compared to the accident.
That would be an interesting study.
Yikes, you shouldn’t be wasting your time online on a Friday night. You should be out telling these fascinating tales at parties.
I should! You’re right about that. 🙂
And if anyone wonders why Crissa and perhaps me appear triggered by this, Om looking around my house at stuff from Samsung, Daiwoo, Bosch, Phillips, my other car is a VW, had 20 other cars of various makes. The Tesla is far and away the most advanced product as compared to any of them - other than perhaps the first iphone yet, the negative coverage is completely unrelenting.
I would note that such negative coverage completely pre dated Elon’s bizzare reaction to taking 20 years of ADD meds or whatever the f is going on there.
This would be neither here nor there if it wasn’t for the vast fossil fuel industry aligned against any transition to an electric sustainable future. So when here, of all places, it’s reflexive anti Tesla it’s worrying.
By all means buy some other EV if you like it better, but if say, you like any of the current BMWs better, do note that without Tesla eating their lunch, BMW would not even have made them.
I’m not going to trust auto pilot in my M3 when it can’t even get the auto windshield wipers to work right half the time. I have one of the first year models so I’ve been manually overriding the auto wipers for five years now. After a dozen software updates it’s no better than it was in 2019.
Seeing water droplets is a completely different problem than seeing lanes.
I hope Elon doesn’t cut Crissa’s position as Tesla spokesperson to the blog.
Fuck you.
I'm just as likely to advocate good data for any EV or autonomy system.
Wow that escalated quickly. I didn’t think corporate whores had such thin skin. The drug reps I’ve dated could take teasing about how relentlessly upbeat they always were about their company’s drugs, but I guess car reps are more sensitive?
Kevin said we're allowed to swear.
But you started it. Escalation was all yours. Don't walk into a bar where people are allowed to tell you to fuck off and complain after you insult them they told you to fuck off.
Boy, Crissa, for one who loves to deal out the insults you sure are thin skinned.
Pray tell, where have I ever handed out an insult?
"You started it" What are you five?
Crissa, as Saint Joan used to say-
Grow up!
Moving back into the political realm, Crissa is perhaps a bit like all the millions of Donald fanboys/girls. Only about a car. Or maybe Elon.
It's a human thing, or maybe a mammal or animal thing. Some icon gets established in the brain and then everything including facts and logic falls to the wayside. Calling Dr Freud!
Fuck off.
I don't like liars, so I don't like you,
No facts or logic in your post except an assumption I must like a socially ignorant asshole bigot just because I happen to like talking about technology.
...and nothing about having an autism spectrum diagnosis makes you socially ignorant or bigoted. Being bad with social cues is not the same as failing to take account in empathy.
Shill sounds mad
What if all cars had "Nags" that detect if you aren't looking at the road enough, jab you when you fall asleep, or even don't let you start when you are drunk or under some other influence?
How elaborate would these things have to be compared to FSD? Things like that would really cut down on accidents.
People would revolt. The autonomy systems are not there yet, and are wrong more than they're right - until you get to something like Waymo or FSD.
And even then, there's patents keeping them from reading speed signs and whatnot.
My 2023 Prius Prime: a) nags me if I look away from the road ahead for more than a few seconds, b) supposedly will alert me if it thinks I have fallen asleep (I haven't fallen asleep yet, so this feature has not been tested by me), and c) if I'm going fast enough for lane-tracing assist to be in effect (around 30 mph), nags me if I don't keep a reasonably firm grip on the wheel (even though it's doing all of the work of keeping the car centered in the lane).
It also has various obstacle detection features (e.g., pedestrians crossing in front of the car), as well as automatic braking if it thinks a collision is imminent. On the freeway, it will automatically change lanes for me--all I have to do is momentarily move the turn signal level in the desired direction.
If I'm in stop-and-go traffic, it will do all of the stopping and going by itself, and will accelerate to the preset cruise speed once the traffic starts moving.
It's not perfect; for example, if I'm in the right lane on the freeway, and there's an exit, it tends to want to follow the right stripe and lead me off onto the exit. Overall, though, it works extremely well.
That's what I have with my Nissan Leaf+. And our Nissan salesman was VERY clear that their "driver assist features" were absolutely NOT an autopilot in any way. It does have a pretty well chosen set of detectors and nags in it's "arsenal".
It can't tell where I'm looking (yet), but if I take my hands off the wheel for more than a very short time, it will start audible warnings. Lane warnings (if you have good road markings) also let you know if you are crossing into another lane - flashing lights in the side mirror on the side you are drifting towards escalating to a "stick shaker" warning if you get too close or over the line.
It also lights up the mirror warnings if cars are overtaking you on either side - steady at first - flashing and stick shaker if you do the slightest action to even start to turn in the wrong direction. Also very strong braking up to emergency level if a car suddenly slams on the brakes or cuts close in front of you.
The adaptive cruise control is very good - maintaining proper distance for the speed you are at, slowing down if the car in front slows down, speeding back up to set speed when safe. Oh yeah, if the car ahead is at proper distance as judged but the adaptive cruise control, you won't get emergency break response on abrupt braking.
There are also sonar sensors on the front and rear bumpers that are really good at detecting people/pets, and cars even before they show up in your visual field - good side detection even with your bumper barely peeking out of a parking space between two big vehicles on either side.
It definitely makes driving a good bit more relaxing.
Statistics are funny, and often all about the framing.
Are relatively expensive models less prone to accidents than the average or less expensive? If we don't know this, we can't just pose one model against all the others as if they are the same just different model name and number.
As for the accident numbers, are we counting all cars involved in any accident regardless of fault, or only cars driven by those deemed responsible for the accident?
Per ride share services, whose freelance drivers are some of the most keen to avoid accidents, as could impact their employment, by my own experience Teslas are are an inordinate percentage by model, given the easy availability for drivers to lease from Uber/Lyft.
The history of statistics is littered with questionable choices for independent and dependent variables, somehow presuming that overall body of findings will overcome the over-simplistic framing of the more questionable individual case findings (ex. do people who eat cereal for breakfast live longer?).
Ride shares are only like, 10% less likely to get in an accident per mile. But yeah, the effect of expensive versus inexpensive does affect, say, the FSD numbers.
But Tesla is trying to do apples to apples comparing drivers who use Autopilot to those who don't - they're the same mix of cars and expense, since Autopilot is a standard feature.
Of course, you could point out that doesn't mean that drivers who use autonomy features aren't just more natively safer drivers.
So, for anyone who cares, here is a link to the actual report.
https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf
Its six pages, not too bad, and it shows exactly what I pointed out above, (1) its from 2018-2023 for starters, (2) it even breaks the crashes down by how - i.e. frontal vs. side to side yaw, and is fully investigated crashes to where Tesla is cleared as to any AutoPilot of FSD use.
As I said, no other automaker is going to have the data to even allow such a report to be produced, which is really the main thing.
Also, it kind of goes without saying that if Tesla's actually had the crash per miles driven of say, a sports car like a Corvette AP and FSD would have been ordered disabled a long time ago.
I found that using AP and FSD actually makes me more careful. The poor Huang crash the guy was playing a video game on his phone! Not even just texting or dialing a number!
The only thing left out of the detailed chart is miles driven which, like any stat, is the key one here.
One of course can say this set of statistics seems to demonstrate that Teslas are involved in less accidents, but you can't then make the causal claim that they are safer, which would be naive, just like asserting some causal factor to cereal eating eating without knowing what type of cereal, what type of milk, and even if controlling for all of these factors, the evidence would still only mildly suggestive considering all the other factors of longetivity not being considered.
One takeaway I've got from working with Big Data is that nearly all human studies are hopelessly small-sample and mostly naive, social science statistics are mostly meaningless.
One thing is certain, though: I'll never buy a car from a company headed by Elon Musk.
That's fine, he's an asshole. But there are 100,000 other folks who work at Tesla and don't get much of a say in who's the CEO, and they seem to make pretty good cars.
“That's fine, he's an asshole. But there are 100,000 other folks who work at Tesla and don't get much of a say in who's the CEO, and they seem to make pretty good cars.”
Now do the 10 million Germans that served in the Heer in 1944 and rationalize that to your hearts content. They seemed to have a pretty good fighting force.
Neither will I. Tesla is just a car company with more competition than they used to have. So they’re dropping prices to keep making sales.
It’s also interesting that elon has been in Tesla leadership since about 2004. That’s a LONG time to only have a few car models for sale. So they’re not particularly good at that.
The cyber truck is a joke. And I’ve seen plenty in the media about its various shortcomings, such as bricking itself in a car wash because it wasn’t in “car was mode”.
The most recent quarterly results were a disaster, but analysts saw elon on a “war footing”, so the stock is up a bit because of his earnings call performance. The real proof will come in future earnings reports.
I don’t think elons leadership qualifies are particularly great, but maybe he’ll pull it off, or maybe not. I don’t hear much about his other companies except spacex so who knows?
Well, there's the revolutionary nonexistent and nonsensical vacuum tube train thing ("Hyperloop," because of course). And the Boring (haha he is so funny) Company digging small bore tunnels just like every other bored tunnel except skinnier with pods no wait Teslas on skateboards no wait just Teslas transporting people in Las Vegas at a fraction of the efficiency of every automated airport train system on earth anywhere.
It was obvious long ago that we’d reach a state where self-driving cars were safer than humans, but would fail in bizarre ways that no human would…and the lawyers would reap billions.
"Teslas with autopilot engaged are probably safer than Teslas without autopilot and a lot safer than the average human-driven car."
This is trivially easy to believe for anyone who has experienced the joys of contemporary motoring on almost any freeway in or around almost any city in almost any level of traffic. On surface streets too, come to think of it.
And FWIW it's fine to admit this while stipulating that Musk is a thoroughly reprehensible person. Henry Ford was also a thoroughly reprehensible person, and Ferdinand Porsche was a literal Nazi. So it goes.
+1
I seriously doubt the accident rates for Teslas with FSD are substantially different from any other high priced modern car with antilock brakes, adaptive cruise control, and lane centering assistance. I put my wife's ID4 driver assistance features on when driving on the freeway and, at least in good weather, it is a very good driver. It isn't integrated with navigation nor will it ever try to change lanes on its own but that doesn't matter much from a safety point of view. All of these cars should be much safer per passenger mile than older, less expensive cars without assistance features driven by younger drivers.
Yes. My Subaru isn't capable of any actual thinking, but what it does do is really helpful. Particularly the adaptive cruise control. But the driving aids are just aids, so I have to keep on actually driving. This is the problem with any more advanced system - the driver's attention goes elsewhere because it seems to be on top of everything, until it's not and the driver is somewhere else.
I've seen recent videos of guys letting the Tesla (and I think they are actual testers, not just some YouTubers) drive in light simple city traffic and the cars regularly make really stupid inexplicable mistakes. The drivers grab the wheel or brake because they actually are paying attention, but that's not what humans in real life would likely be doing. Our brains want to be engaged with something more interesting than driving, or maybe catching up on a bit of sleep.
That's why Tesla compares Teslas to Teslas - they all have Autopilot.
Not all Teslas -- maybe most of them -- do not have FSD. Including them in the Tesla stat dilutes the effect of FSD. I don't think your comparison is meaningful.
Tesla’s are up market cars, which means they tend to be bought more by people who are richer and older who tend to be safer drivers.
We just had a local motorcyclist who was rear ended and crushed to death by a Tesla owner who put his car in autopilot and then proceeded to look for his cell phone. Users of any self-driving software need to be aware that at the present state of development the systems need continuous monitoring.
Weird. My very nonsmart Subaru has nailed the brakes several times to the point of immediately activating the antilock feature (it makes a kind of buzzing vibration) before my foot got to the pedal when someone was just crossing the street on foot or on a bike.
So far every incident of rear-ending in the last five years the driver was overriding the speed or collision reduction system.
So it's no different than idiots who set the cruise control and fall asleep - except less likely to hit anyone.
So far, you have posted 23 comments in this thread. Maybe take a break, go for a walk or something?
The Tesla cult is intense.
Pingback: How safe are Tesla cars with Autopilot? - Kevin Drum - Tesla News
I am already full self driving and tbh I trust myself and my 28 years with no accidents that were my fault so I’ll keep on keeping on and avoiding idiots in teslas.
Speaking of which, Tesla is now by far the biggest red flag for douches on the road, far outpacing bmw and audi and even big stupid fucking pickup trucks.
I wish there was self sorting for morons that works this well in every walk of life.