Unless there's a very good reason, if National Highway Transportation Safety Administration has it then the taxpayers who paid for it should have access too.
The "We'll wait" isn't really seen as effective discourse here like it on on Reddit/Twitter. Just state your argument and let it stand on its own without the perceived "mic drop".
Crash data for all other ADAS systems is already public [1]. The only manufacturer with heavily redacted information in that data to the point of being useless is Tesla.
When I look at this data, I see the type of self driving in use, as well as the written narrative of every crash, along with several other fields as REDACTED FOR BUSINESS REASONS, only for Tesla vehicles, where every other manufacturer seems to have these fields populated. To me, that information would be crucial to understanding what actually happened in each case, as opposed to only being able to understand some of the ambient conditions around each accident.
Granted there are some other rows with missing or incomplete information, but Tesla appears to be the only manufacturer for which this information is withheld in every single instance without exception.
> Granted there are some other rows with missing or incomplete information, but Tesla appears to be the only manufacturer for which this information is withheld in every single instance without exception.
Again, not true.
I just filtered for BMW, and in every single instance, without fail, the ADS/ADAS Version cell is either redacted or blank.
Did a quick check and yeah, there's a lot of redaction/blanks in `ADAS/ADS System Version`
Rpts RdBl Pcage Entity
48 12 25.00 Ford Motor Company
18 6 33.33 Rivian Automotive, LLC
10 4 40.00 Volkswagen Group of America, Inc.
29 12 41.38 Mercedes-Benz USA, LLC
38 18 47.37 Lucid USA, Inc.
27 15 55.56 Hyundai Motor America
37 22 59.46 Kia America, Inc.
6 4 66.67 APTIV
6 5 83.33 Porsche Cars North America, Inc.
12 10 83.33 Daimler Trucks North America, LLC
159 156 98.11 Honda (American Honda Motor Co.)
126 124 98.41 Subaru of America, Inc.
3003 3001 99.93 Tesla, Inc.
1 1 100.00 Nuro
2 2 100.00 Mazda North American Operations
6 6 100.00 Chrysler (FCA US, LLC)
71 71 100.00 BMW of North America, LLC
7 manufacturers don't have that field populated with useful information. I consider 124 out of 126 reports redacted or blank to be close enough to "every single instance" for this argument, for example. Furthermore, over half have over half blank or redacted, and the lowest is 25% missing info.
I don't own nor do I want to own a Tesla, but stuff like this is what gets reported and the corrections or actual facts get buried in the resulting noise. I don't really even care that this is about tesla, even.
If this was some sort of rendering or CSV error on your part, then that could happen at CBS or msnbc just as easily, and tomorrow the headlines scream "Tesla only automaker shirking reporting responsibilities"
No serious analysis can be done when we can’t even tell if a crash occurred under FSD Supervised or Autopilot because they’re two very different things with different capabilities. Same with withholding software/hardware versions and narrative of events.
Tesla also has a problem of their telematics underreporting crashes. One of the reasons for that is they don’t consider it a crash if airbags don’t deploy. This was called out by the NHTSA in a prior investigation: https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf.
Here’s the relevant paragraph from that report:
> Gaps in Tesla's telematic data create uncertainty regarding the actual rate at which vehicles operating with Autopilot engaged are involved in crashes. Tesla is not aware of every crash involving Autopilot even for severe crashes because of gaps in telematic reporting. Tesla receives telematic data from its vehicles, when appropriate cellular connectivity exists and the antenna is not damaged during a crash, that support both crash notification and aggregation of fleet vehicle mileage. Tesla largely receives data for crashes only with pyrotechnic deployment, which are a minority of police reported crashes.
One is a paid version where Tesla opts to drive safer, and if you don’t pay then Tesla is allowed to drive more dangerously? Seems like a jury would question why Tesla would allow a version that is known to be less safe.
>and the crash involves a vulnerable road user being struck or results in a fatality, an air bag deployment, or any individual being transported to a hospital for medical treatment.
Tesla has rebranded its self-driving ambitions half a dozen times. (I’m honestly currently blanking on which of Robotaxi and Cybercab is the Level 4 product.) It’s fair to point out that “advanced driver-assistance” is another neologism of Musk’s, and not a term to be treated as comparable with other companies’ capabilities.
This mostly just demonstrates that the reporter doesn’t know what they’re talking about. ADAS is a term of art that encompasses everything above dumb cruise control, including things like adaptive cruise control or collision detection.
He's at war with the gov now. He's on a mad tweet frenzy about burning up all the GOP reps that voted for the tax bill. They just doge'd his pick for NASA Administrator on the same day his black eye showed up. Get the popcorn.
This is interesting because the only thing keeping senators in line with Trump's bs before was the threat of a primary funded by Musk, how does he whip their votes now without that threat?
The primary challenge is still a threat. Instead of just turning to the Bank of Musk, he'll/they'll(RNC) will do what they have always done by sending out campaign emails to get people to donate.
Europe is shooting itself in the foot if they're not doing their damnedest to brain-drain as many scientists and engineers as they can while things implode here.
Who wants to come will come, no worries. We dont offer highest salaries but highest quality of life and happiness, and currently arguably highest moral ground. Like all folks who want to study on Harvard lol.
No point trying to get folks who will move again soon when something else changes in US.
Funny he is suddenly 'at war' at the exact same time he was scheduled to leave government from the start (temporary employee limit) AND he had to go back to his companies and rehab his reputation. Super convenient timing and totally not theatrics/lies from the reality TV personality and the guy who said he wasn't donating to either presidential candidates this cycle right before buying Trump the election.
Honestly still waiting for someone—could be Canada, the EU or California—to announce heightened approval standards for (or even a moratorium on) cameras-only self-driving cars on public streets.
Well, my claim was fairly specific, so it's quite easy.
There are ~42k traffic fatalities in the US each year. Cameras-only self driving cars are a tiny fraction of the number of total cars.
The highest estimates I've seen for annual traffic deaths with an ADAS involved (not even implying causation) is in the range of dozens. Cameras-only self driving cars would be a fraction of those. As a result, there are quite possibly more than a thousand traffic fatalities each year caused by human-driven cars for each ONE caused by a cameras-only self driving car.
But my original claim was only that they kill a lot fewer. That seems self-evident.
That's cute, and also totally irrelevant. Nobody cares about absolute numbers, the thing to care about is the rate. Pick your denominator, but I like deaths per million miles.
Polonium ingestion also kills fewer people than self driving cars, so by that token, polonium ingestion is perfectly safe.
> Meanwhile, there are ~42k traffic deaths per year that could be prevented by focusing on human-driven cars
Eh, there are over a hundred thousands deaths attributable to Musk's actions in the White House [1]. I agree with you on the short-term calculus. But trusting Tesla to help reduce those traffic deaths--and furthermore, enabling its position of power--puts those states' sovereignty in jeopardy. Swapping lives for sovereignty is an old (and trusted) trade.
Tesla has roughly 0.3% VIO TOTAL in the US (taking global statistics into account it's barely measurable), and a fraction of that fraction are actually using FSD on a regular basis - so I would sure hope they kill "a lot fewer people".
Indeed. My point is: so why focus on regulation that targets <0.1% of traffic fatalities?
Here's an alternative idea that would save a lot more lives:
Take the camera-based driver attention monitoring that works in my seven year-old Tesla, which notices IMMEDIATELY if I look away from the windshield for more than a second or two, and then require that in the human-driven cars.
Estimates for annual deaths in the US from distracted driving are between 3,250 and 12,400. An in-cabin camera is not expensive or specialty hardware. The tech is there, the costs are low. We could save a lot of lives!
If we're ignoring that to focus on Tesla's FSD, the goal is not sensible regulation or saving lives.
>Indeed. My point is: so why focus on regulation that targets <0.1% of traffic fatalities?
A: Citation? Just because it's less than 0.3% of cars on the road doesn't mean it's less than 0.1% of fatalaties. And citation that doesn't let Tesla pass off any FSD crash as "driver error" which they have a horrible habit of doing. If FSD disengages at impact, they call that driver error, which is absolute bullshit.
And because Tesla is taking 0 accountability for it, they are passing it onto the driver. They want to have their cake and eat it too. If you or I are driving distractedly and kill someone, we face serious criminal and financial repercussions.
If FSD decides to swerve out of the lane and into oncoming taffic, Tesla wants to shrug and say "I guess the driver should have been better". That's trash, and should be banned from our roads summarily.
> Take the camera-based driver attention monitoring that works in my seven year-old Tesla, which notices IMMEDIATELY if I look away from the windshield for more than a second or two, and then require that in the human-driven cars
This. Put it on all vehicles that are driven (exc. waymo, zoox, and the like).
> Take the camera-based driver attention monitoring that works in my seven year-old Tesla, which notices IMMEDIATELY if I look away from the windshield for more than a second or two, and then require that in the human-driven cars.
This exists already in Subaru vehicles, even ones with ICEs. It's called "DriverFocus." It's super helpful. However, I don't believe the technology is mandated in all vehicles yet.
The driver drowsiness alert in Teslas seems to be much more limited than that. It only activates at speeds over 60km/h, when driven for more than 10 minutes, and when Autopilot is not engaged. I wonder why they disable it when Autopilot is on?
AutoPilot is already independently monitoring driver attentiveness. With FSD, if your eyes are visible, it's watching your eyes. If it can't see your eyes (or you don't have FSD), it falls back to requiring the driver to apply a small and specific bit of torque to the steering wheel consistently.
The drowsiness alert would be superfluous when it's watching your eyes. It's already going to yell at you if it can't see your open eyes looking out the windshield for more than a second or two.
Whether or not the "steering wheel torque" method is better than a vision-based driver drowsiness alert is probably debatable, but it would be pretty tough to fall asleep while also applying exactly the right amount of torque to keep Autopilot engaged.
OP's point is a Subaru monitors its driver just as closely as a Tesla but over a broader range of circumstances. You don't need Teslas in your jurisdiction to get this safety win.
It yells at me sometimes when I'm driving down my drive way and looking at my goats instead of the driveway, which is fair. It also yells at me when I look at the mirrors for 'too long' or if I look for 'too long' for potential cross traffic when crossing an intersection (when driving late at night, I try to look for potential red light runners, but you have to spend more time looking). It also tells my spouse to sit up when she already is. Chances are this alert is going to be disabled, because it's a bigger distraction than anything else.
It also likes to alert about cross traffic when I start moving to sequence after traffic I saw that is in motion. Those alerts would be handy if it were about traffic I didn't see though, so I don't want to turn them off, even though so far they've been unhelpful.
Wife has a relative who was just (this weekend) in a major accident where a tesla ran into them and pushed their a ditch where it rolled a few times. Initial report says the Tesla was in self drive mode. Will be interesting to see who was at fault here but so far it is not looking good for Tesla.
A plurality of those are in Texas, as well. I used to say, someone in the US is more likely to die in a car wreck in Texas even if they never go to Texas, that's how skewed they make the statistics. But I stopped looking at the stats a few years ago so I stopped saying and defending that. It's just a new lens to view this information through.
Car wrecks are correlated with all kinds of things from education to poverty and Mississippi is dead last or tied for last in every dimension of quality of life.
also the roads outside of the interstates and US highways are pretty awful, in my (limited) experience driving in MS. one time we had to brake from 60 because a cow was on a bridge over an interstate.
Texas has a plurality of fatal car accidents (for USA), but California is not far behind, and in 2022 California has slightly more deaths. (This page doesn't have the number of fatal car accidents for 2022, which is a bit odd.)
You're not looking at absolute numbers, which is what plurality means. I don't see how "someone in the US is more likely to die in a car wreck in Texas even if they never go to Texas" could make sense.
A driver in the US dies while driving due to a crash/wreck/whatever.
Statistically, that occurred with the highest probability in TX. as i said, this was like 2015-2019 when i used to claim this. There's a sign on freeways in TX that say "highway deaths so far in <year>: <16 bit int>" which led me to start looking in to it, and i think my little quip is just a way to draw attention to how dangerous it is to drive in TX. But it is quite large, Texas.
Happened in north carolina, but the incident is still being investigated according to the relative who is still in the hospital (luckily only major bruising, nothing broken, subarus are really good at rolling safety it turns out).
To me anything less than true level 4 should remain with the driver.
I also believe that marketing it as FSD should be liable and scrutinized as a level 4 system. Because when you hear FSD, the public naturally thinks the abilities marked in level 4 arguably even 5.
Until the car requests intervention and the timer runs out, levels 3 and 4 are supposed to have the same behavior. If that process has not happened, why should the driver's level of responsibility be any different?
(Though a consequence is that levels 3 and 4 are very close together in difficulty. We might not see many level 3 cars.)
Volvo was visionary and dismissed Level 3 in about 2014 for being too dangerous. Basically the car drives until it doesn't and you may suddenly die because the time to get the situation and react is too short. Level 3 way purely for managers to claim it would be a linear progression whereas it is petty much THE gorge of automated driving. If you look at the SAE table it's just a little blue wart in a green column, but it's a lethal one.
> because the time to get the situation and react is too short
The time is up to the manufacturer, isn't it?
Mercedes uses 10 seconds right now and that seems pretty good to me. At that point I know it can't be too dire or the car would have already emergency stopped.
The time depends on how quickly an event unfolds in traffic. You can't guarantee 10s notice for an event that is imminent in 2s and the system might not be able to handle or can't detect.
The car could become temporarily "blind" for some reason with just 4-5s to brake before a collision. It's enough for a human driver even considering reaction time. But it's impossible to guarantee a minimum time without the ability to predict every issue that will happen on the road.
If there isn't a guaranteed minimum time, then it's not level 3, it's advanced level 2. Level 3 needs to be able to handle very rapid events by itself.
If it becomes "blind" because of an unexpected total system failure, that's an exception to the guarantee just like your transmission suddenly exploding is an exception. It had better be extremely rare. If it happens regularly then it needs a recall.
When dealing with unpredictable real life events there are no guarantees, unless we're considering the many carveouts to that definition from a legal perspective. A blind car (fluke weather, blown fuse, SW glitch, trolley problem) can no longer guarantee anything. Giving the driver 10s, or assuming the worst and braking hard could equally cause a crash.
> your transmission suddenly exploding is an exception
As long as the brakes or steering work a driver could still avoid a crash. The driver having a stroke is closer to a blind car.
> When dealing with unpredictable real life events there are no guarantees
The guarantee here is that the human isn't obligated to intervene for a moment.
If you call that guarantee impossible, then what about level 4 cars? They guarantee that the human isn't obligated to intervene ever. Are level 4 cars impossible?
Is this a wording issue? What would you say level 4 cars promise/provide? Level 3 cars need to promise/provide the same thing for a limited time. And that time has to be long enough to do a proper transfer of attention.
> The guarantee here is that the human isn't obligated to intervene for a moment.
Ah, understood. So the guarantee is that the driver is not legally responsible for anything that happens in those 10s. I always took that as a guarantee of safety rather than from legal consequences.
It's more about safety than legality. But with the understanding that nothing is perfect.
The guarantee is that you will be very safe and you can go ahead and look away from the road and pay attention to other things. But at most this is as good as a level 4 or 5 car, not an impossibly perfect car.
I feel like that lack of standardization is part of the problem. Some manufacturers may pick different times to avoid nuisance braking, but that translates to higher risk to the driver. I’d like to see some core parameters like this standardized (whether by an industry body or regulator).
Yes but there is a minimum time (if a bit under-specified)
> "At Level 3, an ADS is capable of continuing to perform the DDT (Dynamic Driving Task) for at least several seconds after providing the fallback-ready user with a request to intervene."
I've had several different cars from a few different manufacturers with different levels of ADAS systems and have used them on many long road trips and short trips. I haven't used any Tesla ADAS system for very long though.
The highest level of ADAS system I use regularly has facial attentiveness tracking. If you spend too much time drinking coffee or even looking out the sides of the car it will alert you and eventually turn off. So you're not spending a ton of time drinking coffee or reading emails.
It's really nice having the car just want to stay in the center of the lane and keep the following distance all on its own. It's less fatiguing on your hands and arms having the car feel like it's in a groove following all the curves for you instead of resisting your input all the time for hours and hours. It's incredibly nice not having to switch between the brake and the gas over and over in stop and go traffic. Instead, the only thing I need to focus on are the drivers around me and be ready to brake.
I've driven between Houston, Dallas, and Austin dozens of times with ADAS systems and another dozen or so times with only basic cruise control. It's way nicer when the only time I have to touch the gas and brake are getting on and off the highways. I'm considerably more relaxed and less exhausted getting to my destination.
Let's assume all these options are either the same price or an immaterial difference to the price of your next car. If you had an option for a car with basic cruise control or no cruise control, which one would you take? If the option was basic cruise or adaptive cruise which kept pace with traffic and operated in stop and go conditions, which would you choose?
You're right cruise control or whatever you want to call it is definitely great on long journeys on the highway. Maybe car makers should concentrate on those systems?
They are, other than Tesla. GM's SuperCruise, Ford's BlueCruise, and Mercedes DrivePilot are the few actually hands-free driving systems made by the legacy automotive companies and they're all largely locked to only fully operate on mapped and approved highways.
I actually think its worse than driving yourself. Humans are OK at doing a repetitive task non-stop. They're terrible at sitting still doing nothing waiting to quickly spring into action. They fall asleep or their mind wanders. This is something a computer is good at yet we've got it reversed. The car drives along doing mundane things but then hands it over to groggy human right when things really get hairy.
And then there's the skill atrophy. How do you learn to perform in stressful situations? By building up confidence and experience with constant repetition in more mundane ones, which this robs you of.
And the part of my sentence that you cut off was all about the circumstances of intervention.
Level 2 requires the driver to choose whether to intervene at all times. This is an unreasonable task for humans.
Level 3 puts the car in charge of when intervention is needed, and even once it wants intervention it still has to maintain safe control for several seconds as part of the system spec.
Level 4 puts the car in charge of when intervention is wanted, but you can refuse to intervene and it has to be able to park itself.
So I will double down on my claim. Until the car requests intervention AND the timer runs out, level 3 and 4 are the same. They require the same abilities out of the car. And that section of time, between wanting intervention and getting intervention, is the hardest part of level 3 driving by far. If you can solve that, you're 90% of the way to level 4.
A level 3 car has to be able to handle emergencies several seconds long, and turning it into level 4 is mostly adding the ability to park on the shoulder after you get out of the initial emergency.
The gap between 4 and 5 is a bunch bigger. A level 4 car can refuse to drive based on weather, or location, or type of road, or presence of construction, or basically anything it finds mildly confusing. 5 can't.
I edited a bit for clarity, but also I'll append a thought experiment as an extra edit:
A level 3 car with an hours-long driver intervention timer is basically identical to a level 4 car.
If you have a 0 second intervention timer, you're barely better than a level 2 car.
How long does the timer have to be before developing your level 3 system is almost as difficult as a level 4 system?
I agree with your thought experiments and also agree that overall it's a valid, technically accurate interpretation. So, this may be where we agree to disagree.
I still stand by level 3 != level 4 in terms of real world liability.
Level 3 allows too much wiggle room and sloppiness to be able to legally shift liability away from the driver. At that point you're playing that "intervention period" length. Manufacturers claiming Level 3 will want to lower it as much as possible, regulators raise it. To me, Level 3 simply shouldn't exist.
Only at Level 4 is the expectation, without a doubt, the machine is in control. A person in the driver seat is optional because the steering wheel and pedals are as well. When people bought "Full Self Driving" they seriously believe "when can I go to sleep?" ability is where it belongs, which always put the expectation at Level 4.
Saying level 3 shouldn't exist makes sense. But I don't think the liability gets very blurry as long as the intervention period is properly documented.
It looked like the Mercedes system is 10 seconds which seems like plenty to me.
And while it would be nice to sleep I'll be pretty happy just looking away from the road.
"A lie", FSD as it stands right now is a lie. A few cars might be able to drive a few geofenced places, but no car anywhere can drive anywhere, even with perfect weather and visibility and I'd even wager no traffic or even no other cars at all. Our Subaru gives up steering if there's no olcar in front, on "suburban" and rural roads about 35% of the time. More on some roads, less on others. I cannot determine, while driving, the cause for half of the self driving disable occurrences. No fog line and a broken center for an intersection on a 1 lane road it'll shut off nearly every time. It's surprising when it doesn't.
I've clocked nearly a half million miles on the road (I'll be there sometime in the next 9 months), and the range of technical ability you need to drive in just the US, no, scratch that, any given state or even county varies so much and potentially so often that FSD is just a lie to sell cars. I'm willing to upload a full hour drive touring a few parishes around here in my quite heavy Lexus, front and rear cameras, just to prove my point. I'd do it in the subaru but the dashcam isn't very good and also it's lineage is rally so it exaggerates how poor the roads are. My YouTube has dashcam footage of drives that I'm willing to bet no automated system could handle, even if it claimed to be "level 5". Driving after a storm or hurricane is another issue. I know the hazards in general and specifically for the areas I'd need to travel during or after an emergency. I cannot fathom the amount of storage and processing that would take, to have that for every location with roads. On board, in the car? Maybe in 20 years.
> I cannot fathom the amount of storage and processing that would take, to have that for every location with roads. On board, in the car? Maybe in 20 years.
Doing some napkin math, with 4 million miles of road in the US, if you wanted to store 1KB of data per meter of road, hundreds of data points, you'd only need 7TB for the entire database.
And the processing to make it shouldn't be anything special, should it? Collection would be hard.
Currently that would probably cost ~$500 per car to implement based on retail pricing of 8TB SSDs. It would need to be updated constantly, too, with road closures, potholes, missing signage, construction. With an external GPS unit like a tomtom, they had radio receivers in the power cord that tuned to traffic frequencies, if available, and could route you around closures, construction, and the like, so you need a nationwide network to handle this. Cellphone won't cut it. Starlink might, but regardless, you need to add that radio and accoutrements to the BOM for each car.
and i'm not talking about the processing of the dataset that gets put onto the 8TB SSD in the car; i am talking about the processing of the data on the 8TB SSD on the car while at speed.
furthermore, i am fairly certain that it would take, on average, more than 1.6MB per mile to describe the road, road condition, hazards, etc. a shapefile of all roads in the US - that which gets one closer to knowing where the lanes are, how wide the shoulders are, etc is 616MB. and it's incomplete - i put in two roads near me with fairly unique names and neither are in the dataset. So your self driving car using these GIS datasets won't know those roads.
I had an idea to put an atomicpi in my car, with two cameras. it has a bosch 9-dof sensor on the board, coupled with the cameras you can map road surface perturbations, hazards, and the like, which i believe will be much more than 1KB per meter, especially as you need "base" conditions and updates and current conditions (reported by the cars in front of you, ideally).
the csv GIS dataset looks like this:
and i ran, for example `awk -F, '/PACIFIC COAST/ {sum += $4} END {print sum}' NTAD*.csv` and it spat out 79.04, which i think is a bit shorter than reality. Looks like the dataset i pulled is only "major roads" as well - but that doesn't explain 79.04 as the sum of lengths of all rows with "PACIFIC COAST" in them. It does show the total length of interstate 10 is 3986.55, which is roughly double what the actual length is (2460mi), so perhaps i'm just not understanding this dataset.
Anyhow 600+ MB for just that sort of information (plus shapes) for only a really quite small subset of roads in the US.
anyhow my thoughts are scattered, this input box is too small, and i'm not really arguing. Maybe it is possible, but it will raise the price thousands of dollars per auto, you need infrastructure (starlink will work) to update the cars, and so on. I'm prepared to admit i am wrong, but your comment didn't move the needle for me.
If you want such constant updates that's tricky to distribute and hard to collect, but let's put that aside for a bit. I want to focus on the amount of data and how the car would use it. With $500 of SSD being nice and cheap.
> i am talking about the processing of the data on the 8TB SSD on the car while at speed.
I'm not worried about that. The actual driving takes such powerful computers that even if there was a petabyte of total data, the amount the car would have to process as it moves would be a trickle in comparison to what it's already doing. Max 50KB per 10 milliseconds. And obviously the data would by sorted by location so there's very little extra processing required.
But you tell me, how many data points do you think you need per meter of road?
I really don't think you need millimeter-level surface perturbations all the way across. Mapping the precise edges of the road and lanes should only need dozens of data points, 4 bytes each. And then you can throw a few more dozen at points inside the lanes to flesh it out. You can throw a hundred data points at each pothole without breaking a sweat. Measuring the surface texture in various ways and how it responds to weather is only going to take a handful of bytes per square meter, in a way that repeats a lot and is easy to compress.
That's an extremely inefficient format. Unnecessary object ids, repeating metadata over and over, way too many decimal places, and all stored as text.
But even then, your database is so tiny compared to the size I suggested that I don't think we can extrapolate anything useful. Even if we 4x it or whatever to compensate for a lack of rural roads.
Suffice to say that the 600MB just lets you draw the roads on a plane, it's like comparing an ascii art drawing of the road (from .csv/shp) to a digital still of the road (the amount of information you'd actually need). you absolutely cannot rely on "a couple of sensor [types]". I mentioned i have nearly a half million miles on the road. All of that prior experience influences my driving when i am driving someplace new. in that 8TB disk, you have to find a way to produce that "experience", except instead of my 0.5mm miles, you are talking about the aggregate "experience" of 0.5mm miles per road per unit of time (day for some places like I-10 through Los Angeles, month for others, maybe a year for some "rural" roads.)
none of this has to do with visual or proprioception. It's knowing "every inch" of road. It's knowing how far i can leave the center of the lane if someone else crowds me or goes over the center divider, because the shoulder is soft through here because logging trucks have been exiting the forest onto the highway. It's knowing what part of I-605 floods - not the whole thing, some lanes, some places, and "flood" means 2+ inches of water on the road surface, hitting it at speed makes a tidal wave flying into other lanes. If someone hits that in front of you, you're blind for a couple of seconds minimum. If we want to have semi trucks be "FSD" it needs to know, for the traffic and other conditions, how fast to go and what gear to be in to climb each hill, and then the hazards that are over the hill - that a trucker would know. Where's the gravel bed on more mountainous passes? Or more simply, what time of day neighborhoods are more likely to have people approaching or going through / out of intersections, blind or otherwise. How many "bytes" is that information, times every neighborhood? If many cars brake at the same place, there's probably a reason, and that needs to be either in the dataset or updated somehow if conditions change. You ever used Waze and had a report of something on the road or a cop parked somewhere, and it's nowhere to be seen? And that's updated much more frequently than the radio-info on the GPS systems i referred to earlier. Some roads become impassable in the rain, some roads ice more readily.
If this was easy/simple/solved, waymo et al would be bragging about it, the tech in their cars. Waymo (or the other one) specifically, because they cover less than 0.1% of road surfaces in the US, in some of the most maintained and heavily traveled corridors in the world. So, if anyone from a robotaxi company happens by and knows roughly how much storage is needed for <0.1% of the road surfaces in the US, then we could actually start to have this dialog in a meaningful way. Also i am unsure how much coverage robotaxis actually have in their service area. A "grid system" of roads makes mapping and aggregate data "simple", for sure.
This reminded me a bit of the idea that somewhere in the US there's a database of every sms sent to or from US cellular phones. "it's just text; it'll compress well" - belies how much text there is, there.
for reference, the map in my lexus is ~8GB, for the US. And that's just "shapes" and POI and knowing how the addressing works on each road. It doesn't know what lane i'm in, it doesn't track curves in the road effectively (the icon leaves the road while i'm driving quite often), and overpasses and the like confuse all GPS systems i've ever used - like in Dallas, TX where it's 4 layers high and parallel roads stacked. furthermore, just the road data on google maps for the nearest metro area to my house is 20MB. i have a recollection it goes real quick into hundreds of MB if you need to download maps for the swaths of areas where there is no cellphone reception, like areas in west...
The part of the computer that knows how to drive is completely separate from the 7TB database of the exact shape and location of every lane and edge and defect.
> knowing how far i can leave the center of the lane if someone else crowds me or goes over the center divider
Experience, not in the database.
> knowing what part of I-605 floods
> Where's the gravel bed on more mountainous passes?
That goes in the database but it's less than one byte per meter.
> How many "bytes" is that information, times every neighborhood?
I don't know why you would want that data, you should be wary of blind traffic at all times, but that's easy math. There's less than a million neighborhoods and time based activity levels for each neighborhood would be about a hundred bytes. So: Less than 1 byte per meter and less than 100MB total.
> If this was easy/simple/solved, waymo et al would be bragging about it
This doesn't happen for two reasons. One they are collecting orders of magnitude more data than road info, two like I keep saying the collection is extremely difficult and I'm only defending the storage and use as being feasible.
> This reminded me a bit of the idea that somewhere in the US there's a database of every sms sent to or from US cellular phones. "it's just text; it'll compress well" - belies how much text there is, there.
Well we know how many meters of road there are. So it's basic multiplication.
I can tell you how many hard drives you need to store a trillion texts. It's five hard drives.
Google thinks the human race sends almost ten trillion text messages per year. So I guess you could store them all very easily? Why do you think it's not doable?
> Your claim is that a mere 1000x more data is enough to autonomously self-drive anywhere in the US, at any time of day or year, etc...
My claim is that 1000x is enough for utterly exhaustive road maps. Figuring out how to drive is another thing entirely.
ohhhh, we're arguing past eachother. I am unsure how to reconcile.
an SMS isn't just "140 characters/bytes" or whatever (i honestly don't care what your definition of "SMS" is). Of course you could fit 140 characters * 1e12 onto 5 hard drives. Where are you going to put the 1PB (for 1e12, but your own cite says it's 1e13, so 10PB) of metadata, minimum? the most barebones amount of metadata you need to actually have actionable "intelligence" is 1KB per message (technically i was able to finagle it to ~1016 bytes.) And that's for every message, even an SMS that is the single character "K".
you need the metadata to derive any information from the SMS. "Lunch?" "yeah" "where?" "the place with the wheel" "okay see you in 25, bring Joel" This is what you propose to save. (quick math shows you went off something like ~32TB of sms data per 1e12 messages)
in the same way that you propose that the shapes of a road and it's direction and distance "plus 1KB of metadata per meter" is enough to derive the ability to drive upon those roads.
It's pretty obvious that just using sensors is not going to get FSD. Maybe in the next 20 years we will develop sensor technology (and swarm networking and whatever else) that will allow us to dispense with the "7TB" of metadata. My argument is that: we need much more "metadata" than 1KB per meter to "know the road baseline, current conditions, hazards", much in the same way a text message is more than 140 bytes. Driving with "only sensors" and rough GPS has killed people. It does not matter if human drivers have more death per million miles or whatever, because i am strictly talking about FSD, what other people are calling level 5 (i'd even concede level 4; although i wouldn't be able to use a level 4 car where i live for roughly 1/4th the year - and other areas would have more than 1/4th the year.)
> the most barebones amount of metadata you need to actually have actionable "intelligence" is 1KB per message (technically i was able to finagle it to ~1016 bytes.) And that's for every message, even an SMS that is the single character "K".
How did you reach that number?
I figure the most important metadata is source and destination phone numbers and a timestamp, and I guess what cell tower each phone was on. A phone number needs 8 bytes, and timestamp and cell tower can be 4 bytes, so that's 28 bytes of important metadata.
> (quick math shows you went off something like ~32TB of sms data per 1e12 messages)
I was going for a full 140TB of data. 20-30TB hard drives are available.
I did consider metadata, but I figured you could probably put that in the savings from non-full-length messages.
> Where are you going to put the 1PB (for 1e12, but your own cite says it's 1e13, so 10PB) of metadata, minimum?
Well for just the US it would be closer to 1PB. But, uh, I'd store it in a single server rack? (ideally with backups somewhere) As of backblaze's last storage pod post, almost three years ago, it cost them $20k per petabyte. That's absolutely trivial on the scale of telecomms or governments or whatever.
> My argument is that: we need much more "metadata" than 1KB per meter to "know the road baseline, current conditions, hazards", much in the same way a text message is more than 140 bytes.
I mean, I agree with you about needing extra information.
But that's why the number I gave is 10000x larger than your CSV. My number is supposed to be big enough to include those things!
> note: the metadata for a meter of road could be:
I really appreciate the effort you put into this. I have two main things to say.
A) That's less than a kilobyte of information. Most of the bytes in the JSON are key names, and even without a schema for good compression, you can replace key names with 2-byte identifier numbers. And things like "critical" and "Active roadwork zone with lane closure" should also be 1-byte or 2-byte indexes into a table. And all the numbers in there could be stored as 4 byte values. Apply all that and it goes down below 300 bytes. If you had a special schema for this, it would be even lower by a significant amount.
B) Most of those values would not need to be repeated per meter. Add one byte to each hazard to say how long it lasts, 0-255 meters, instant 99% savings on storing hazard data.
> each "trick" you do to reduce the size of the metadata necessarily implies more CPU needed to parse and process it.
CPUs are measured in billions of cycles per second. They can handle some lookup tables and basic level compression easily. Hell, these keys are just going to feed into a lookup table anyway, using integers makes it faster. And not repeating unchanged sections makes it a lot faster.
and again - if you use clever tricks to reduce this, you increase the overhead to actually use the data.
get a celltower snooper on your phone and watch the data it shows - that's the metadata for your phone. SMS dragnet would need that for both phones, plus the message itself.
It's not an integer. But you can store it inside 64 bits. You can split it into country code and then number, or you can use 60 bits to store 18 digits and then use the top 4 bits to say how many leading 0s to keep/remove. Or other things. A 64 bit integer is enough bits to store variable length numbers up to 19 digits while remembering how many leading zeros they have.
If you want really simple and extremely fast to decode you can use BCD to store up to 16 digits and pad it with F nibbles.
> JSON
Most of this is unimportant. Routing path, really? And we don't need to store the location of a cell tower ten million times, we can have a central listing of cell towers.
I don't think we really need both phone number and IMEI but fine let's add it. Two IMEI means another 16 bytes. And two timestamps sure.
Phone number, IMEI, timestamp, cell tower ID, all times two. That's still well under 100 bytes if we put even the slightest effort into using a binary format.
> and again - if you use clever tricks to reduce this, you increase the overhead to actually use the data.
No no no. Most of the things I would do are faster than JSON.
Removing the steering wheel and pedals from the robotaxi is Tesla embracing culpability, whether they like it or not. If they are negligent and cannot claim human error they will face huge damage awards.
This is the same attitude that people used to try and avoid any culpability for Boeing in the 737-Max crashes. Even if they was a technical way to avoid a crash, it doesn’t avoid negligent or blatantly bad engineering practices. There’s a reason why engineers are expected to have an ethical duty to the public. Automakers get an industrial exemption on the assumption that the internal processes are sufficient to address the risk…What are we supposed to do when they aren’t?
It seems clear to me at least that Elon did a major pump of FSD, realized he was full of shit so got into politics to try to hack the system in his favor to hide the truth
i think it's fairly easy to get 80+ percent of the way to FSD and it looks like you're on the verge of being able to moat your company with actual FSD. He should and probably did know better - although i've seen lots of videos/articles about how he isn't actually that proficient technically.
even if that 80% was 99%, that last 1% will be the cause of some mishaps.
my subaru is within a few percent of 80% FSD if everything is turned on. I still technically have to hold the wheel, but the steering only shuts off about 20% of the time with that being met.
Isn't that the whole point of levels 2 and 3? Fine print applied to the marketed operating modes of heavy equipment. Surprise, you were supposed to be driving!
"Surprise, you were supposed to be driving" is a level 2 problem. Level 2 requires inhuman levels of constant vigilance. Level 3 requires you to be awake and able to drive, and you will get several seconds of warning to switch from watching TV to looking at the road.
I always forget the levels, and I couldn't tell from a quick reading how much time L3 gave you. So sure, "several seconds" isn't deep in "surprise, you should have been driving" territory, but it's still on the order of a developing situation which a computer could get into and then balk and demand the human get themselves out of.
That amount of time is also nowhere near enough for a human to switch tasks like that. I would say it should be at least a minute or two, and even that is pushing it with how people are bound to use the system (spacing out while watching a TV show, etc).
Are you suggesting it takes people a minute or two to start driving and get up to a basic level of safety? How does that even work? There's no way sitting still in a parking lot of driveway counts as getting into the groove.
Yes. A "basic level" of safety doesn't suffice when you could be expected to take over a car already going 75mph on the highway.
Let's say you were parked, taking a nap in the drivers set - when you woke up, would you immediately start driving or would you wait a minute to get your bearings? How about at a highway rest stop? It feels like trying to push back on that by asserting "the driver is supposed to always be alert and ready to be attentive!" is another bout of fanciful fiction like L2. Being outright asleep would seem to be derelict, but I can imagine many fuzzy mental middle grounds, especially in a droning car.
(Somewhat related, if I haven't driven in a month I would say it takes me tens of minutes (maybe 20?) to get back up to the usual groove. Obviously the only way to do it is to do it, but I drive much differently until then)
> Let's say you were parked, taking a nap in the drivers set - when you woke up, would you immediately start driving or would you wait a minute to get your bearings?
It would take me a minute to wake up. Napping is one of the few things you can't do with level 3.
> How about at a highway rest stop?
Wait, doesn't this go against your argument?
Someone at a rest stop can go from "not driving at all" to "full speed down the highway and merging" in a handful of seconds. And it works fine.
> It feels like trying to push back on that by asserting "the driver is supposed to always be alert and ready to be attentive!" is another bout of fanciful fiction like L2.
Depends on what "alert" means. I can be in a general readiness state, with no particular requirements on my focus, for hours on end.
> Being outright asleep would seem to be derelict, but I can imagine many fuzzy mental middle grounds, especially in a droning car.
"The boring droning car made me zone out" is something that can happen while you're driving. A TV show could actually reduce the risk of falling half asleep.
> if I haven't driven in a month I would say it takes me tens of minutes (maybe 20?) to get back up to the usual groove. Obviously the only way to do it is to do it, but I drive much differently until then
How much differently? Also an autonomous car asking for takeover is probably driving super cautiously too.
I've actually driven many thousands of miles with level 2 ADAS. There is no "not paying attention" in the system I'm using. If you even spend a few seconds too much gazing out the sides the system will beep at you and eventually pull over. If I take a few too many sips from my cup it'll beep at me and eventually pull over. I have to actively be shifting my gaze around out of the front, constantly being attentive to the cars to keep it in level 2.
It's really not inhuman levels of constant vigilance, not anymore than actually driving the car regularly. I just don't have to actively be keeping the pedals just right to maintain the following distance myself and I don't have to constantly fight the wheel.
I hate Tesla as much as the next sane man, but this rumor is just a rumor. Tesla counts FSD (and Autopilot) as being "in use" during an accident if it was enabled at any time in the 10 seconds before the accident.
Do they? Given that Tesla can make those logs say anything they want, I would like that code reviewed by a 3rd party.
If we're going to allow companies to write code in which human safety is in danger if that code misbehaves, that code should be auditable by a 3rd party, and those audits should regularly happen.
Code which affects the safety of humans should be reviewed with AT LEAST as much rigor as code for slot machines.
While Tesla could manipulate their logs, serious car crashes frequently result in lawsuits (in the US). The insurance industry and the surrounding specialties - liability attorneys, engineering consulting firms - would very quickly notice a pattern of "driver says 'FSD', Tesla says 'Not FSD'".
Why wouldn't every driver with no compunctions about lying^, knowing that tesla has autonomous features, say that "it was on, then it shut off with no time to react!" At some point you have to go off the logs, at which point the code, i agree, should be audited (sibling or a parent in this thread mentioned.) It's dueling disagreements otherwise.
^ reading anecdotes about accident scenes, someone if not everyone is always lying about what happened.
Initial reports always claim Tesla was in self driving mode. I have seen that a number of times.
In one case there was a claim the driver was in the backseat. This got widely published in all media outlets. And turned out to be complete nonsense, it wasn't even in autopilot.
But of course it could be true but I would wait for the data.
While I admit I shouldn't be defending Tesla for free - I've come to realize a lot of these "FSD crashed into me and Elon is hiding it!" claims usually come down to the user driving recklessly then using FSD as a get out of jail free card.
FSDs failures are either far more boring (imagining a stop sign) or put's the user in danger (driving onto train tracks).
It is possible that Tesla wouldn't want positive data released. If their approach is trending positively, releasing the data would suggest to competitors that they should adopt the same approach.
Unless the data literally sings it from the tree tops less than honest people will pretend it says whatever they want it to say for clicks and eyeballs.
With how popular Musk is these days I can 100% where Tesla is coming from here.
If your airbags don't deploy, Tesla doesn't consider it an accident for the purposes of reporting (modern safety systems don't blindly deploy airbags, they evaluate g-forces, speeds, angles of impact, etc., so you can hit something at 25mph and the vehicle decides your seatbelts are sufficient. Tesla decides "that's not a reportable collision"). Know when else your airbags might not deploy? Very serious accidents, when hardware or controllers are damaged.
Speaking of which, fatalities are not included in that report. "It was a collision where someone died, but doesn't merit inclusion in a safety report" is a weird position to take.
I looked up the American regs and they say record event if "non-reversible deployable restraint" was deployed ... or if the vehicle accelerated over a delta-V greater than 8km/h in 150ms. The regulatory record from 20 years ago is available at https://www.federalregister.gov/documents/2006/08/28/06-7094...
Yes. Been standard since 1990s. IDK if it's actually required but everyone does it. At a minimum they store data of the last N seconds before deployment. It's mostly for debugging problems and preventing insurance fraud. The data can be pretty sparse depending on year/make/model. Like in the 90s it was little more than speed and throttle position.
> The records must be reviewed to redact “confidential business and trade secret information of Pfizer or BioNTech and personal privacy information of patients who participated in clinical trials,” wrote DOJ lawyers in a joint status report, filed Monday.
...
> But we can’t show you data
Tesla shows the data to the NHTSA whose experts look at it and can force recalls so your analogy and argument make no sense.
As an anecdotal data point, I picked up a '24 Model 3 precisely for the self-driving capabilities. The difference between a Tesla running hardware/software HW3/v11 vs HW4/v12 was night and day.
Literally felt like the difference between flying a helicopter (actively trying to kill u lol) and an airplane.
I honestly did not get the hype until this specific HW4/v12 combination which didn't exist until last summer or so. It's the first time FSD felt like a safety feature for just $99 a month.
>I honestly did not get the hype until this specific HW4/v12 combination which didn't exist until last summer or so. It's the first time FSD felt like a safety feature for just $99 a month.
That's exactly the problem. It's great right until it isn't, at which point it's likely to make a decision that will kill you or someone else if you aren't lucky.
(most) Humans are REALLY good at paying attention to something that will actively kill them at any moment - you don't see a lot of people running a chainsaw while sending a text to their friend about drinks later in the day.
Humans are REALLY bad at stopping something they trust (IMO foolishly), with less than a half a second of notice, from killing them or someone else. It is completely natural to get lulled into a sense of security when something mostly works exactly as you'd expect.
Meanwhile Tesla wants to act as if it's the driver's fault anytime there's a crash without acknowledging they are actively perpetuating the myth of: "this thing drives itself". It's literally called "Full Self Driving" and Telsa expects the average person to look at that name and think: you need to be vigilant anytime you turn this on because it is a beta feature that may drive into oncoming traffic at any moment.
Agreed, but you can't not pay attention with the new Vision Attention Monitoring. Not sure if it's HW4/v12 specific but it watches your eyesight specifically.
So for example, if I look at the screen, my phone, or start day-dreaming for even a few seconds, it'll beep and quickly strike me out from using FSD. "FSD (supervised)" is how it shows up in the UI too at least giving some expectation of it not being autonomous.
So in practice, I'm picturing the right driving inputs and watching what it's doing.
Quite frankly sounds super boring. I prefer driving than supervising. Only true unsupervised autonomous driving would be interesting for me (e.g Waymo).
An unsubstantiated claim given that there are many, many safe human drivers who have neither LIDAR sensors nor hyper-accurate pre-mapping at their disposal.
"today"? Are there not humans on the road today? There have been a number of safety issues with Waymos, certainly too many to describe them as the one true safe option, today.
Entertaining a No True Scotsmen is a bit of a silly exercise anyway, but this semantic game is extra silly.
The person you replied to was talking about how we can achieve safe autonomous driving today. When I remind you of that context, that your rebuttal is not actually rebutting what they said, I am not performing No True Scotsman.
The gap between humans and computers is enormous, not some weird gotcha tactic.
I literally rebutted that context. Waymos are not safe autonomous driving today, they have caused various safety issues in the era of "today". I didn't include "today" in my original comment because none of the available options are "the only true safe way to do it today", but I don't think it is constructive to just say that.
No True Scotsman was obviously in reference to GC, not you.
"Waymos are not safe" could be a rebuttal to what they said. "There are many safe humans" is not a rebuttal to what they said. Your comment above was the latter.
> No True Scotsman was obviously in reference to GC, not you.
I'm unsure what they said that would qualify. Was it adding "true unsupervised"? I think that's a fair qualification, because most of the point of self driving is lost if I can't look away from the road.
Yes. Once is available more widely to common car owners (either from Waymo or others) that would interest me. Current Tesla supervised style semi autonomous driving I would find either boring or stressful (depending on the scenario). I would rather drive myself.
Note Waymo announced a partnership with Toyota, pretty hand wavy, but at least it seems there’s hope the technology may come to regular car owners at some point.
100% me ... we're on the same page ... I've said it before. I don't want to become my cars manager ... I want to enjoy my driving.
I get that the vast majority aren't car enthusiasts and that's ok, but there is actual pleasure in driving.
And even me - when I want a rest from it - lane assist and cruise control are MORE than enough. I can even add these two to old classics without much bother.
Going all in on autonomy doesn't interest me at all.
When the words are actually spelled out the sheer ridiculousness of "Full Self Driving" having a "(Supervised)" postscript becomes rather readily apparent.
Tesla can't help but know that the supposed non-driver getting constantly nagged to be vigilant as if actively driving destroys most of the value proposition of FSD. Vision Attention Monitoring has quite a bit of potential to be very useful … precisely in situations in which vehicles are not driving themselves.
> That's exactly the problem. It's great right until it isn't, at which point it's likely to make a decision that will kill you or someone else if you aren't lucky.
This should be weighed against the fallibility of human drivers, surely? Our point of comparison is not "perfect", it's "human." Inasmuch, with millions of miles driven, FSD appears to be many times safer than humans: https://www.notebookcheck.net/Tesla-Autopilot-and-FSD-are-no...
Not perfect, and there will be crashes, but much better, and I think that's the yard stick we should be using, because no system will ever be perfect.
Tesla comes across safer when you compare the performance under completely different conditions and then shift blame on humans for not reacting in a second when Tesla suddenly disengage.
Tesla records all FSD accidents when it has been engaged 10 seconds prior to the crash. I don't think your claim is accurate but if you have a source I would be glad to read it.
Those are numbers for Autopilot, not FSD. Autopilot is driving mostly highway miles which have much lower crash rates than the highway/city combined stat Tesla compares against. More importantly, those statistics describe the safety of a supervised system and can't be used to infer the safety of the system on it's own. Tesla refuses to disclose the industry standard "miles per disengagement" for FSD but crowdsourced data puts them at 456, pitiful compared to humans and other self-driving projects. [1]
Also to note, FSD disengagements are probably common enough to still be on the left hand side of the "Valley of Degraded Supervision"[2], where mistakes are common enough that users stay viligant. As mi/de increases to 5,000 or 50,000, the quality of the supervision could degrade to the point that the supervised system is less safe than an unaided driver.
At this point, Tesla's FSD is almost certainly more "baked" than the vast majority of software you've ever used. The amount of engineering and compute time that have gone into it are colossal.
That said, something being excessively baked does not mean it is good.
Thats an irrelevant argument (and unless you work there directly on this just empty baseless words).
The point is - it didnt deliver, and still doesnt. Its a securities fraud out in the open, but clearly from a guy who is above the threshold of applicable law
> It has been designed by brilliant engineers whose only goal is that it works for its intended purpose.
Is this statement actually true? From what I heard, it was designed by highly overworked stressed engineers working in pretty bad workplace conditions. They work there, because there are not many other places to work at and doing similar work.
And their primary goal was to produce as fast as possible.
And it has been limited by cost cutting decisions resulting in not using lidar or radar, meaning it has far less data to work with than other self driving efforts.
I won't trust Tesla FSD for that reason alone. You've got a way to have far more reliable distance sensing and you choose not to use it? That's crazy.
Roads were not built for LIDAR. Building a system that only requires the same sensing that humans use is a system that will work anywhere (if it works).
As soon as you buy into LIDAR (or any other non-human sensing mechanism) you introduce the possibility that your solution could have strict upper bounds on scale.
Which is indeed what we have seen so far with Waymo – it may have fewer issues per mile than Tesla FSD, but every one of those miles is expensively pre-mapped and requires $30k of extra hardware on every vehicle.
And for the use-case of "driverless taxi service in a well-established metro area", the Waymo solution does currently seem to be the better one. But that is also clearly not the ambition Tesla is pursuing.
"Roads were not built for LIDAR" is a strange take. LIDAR detects range between the sensor and some object. Do that in a scanning approach and you have an idea of whether there are objects in front of you, the distance to them, and their rough shape. None of that functionality depends on how a road is manufactured, it's physics.
The $30k figure you're giving for Waymo's LIDAR cost is quite old. As often happens, a new technology got cheaper over time.
"But LIDARS that are $5,000 today are forecast to be under $200 when bought by the millions, meaning the bill-of-materials for the extra hardware should drop below $2,000, and even $1,000, in time."
Fair, I mentioned how roads are built, but I really meant "the entirety of human wheeled transport is built around human sensing capability". But roads matter too. For example have you ever driven through a small Italian town where there are walls on every side? If I was a LIDAR-equipped human I think I would've been overwhelmed and unable to drive at all. I also would've "sensed" various turns a little too late, as LIDAR happens in a straight line, which can occlude facts about the world that are (more) obvious with vision.
In perfect conditions LIDAR is helpful, but what about the million situations where conditions are not perfect? e.g. what about rain/snow/fog? What about telling the difference between a cardboard box and a metal one? If you want a vehicle that can operate in all conditions and situations on the road today, your vision component needs to be incredibly robust to the point of (possibly) obviating the need for LIDAR.
That $30k figure is actually the lower bound on the current Waymo fleet as it exists today. The $5k figure from the article is the estimated cost per LIDAR sensor (there are 4 on a Waymo vehicle) if you bought them today, which is still $20k total. Additionally the vehicle cited in the article isn't on the road yet, and therefore seems a bit premature to cite as a reference for real-world cost. Imagined future costs are just that, imagined. Those are uninteresting in the same way that Tesla's promises of vision-only capability are uninteresting. What matters is what is actually achieved.
If Waymo achieves the range of driving conditions you can currently operate Tesla FSD, I will be impressed. Likewise if Tesla achieves the safety and consistency per mile that Waymo has, I will be impressed. The question is: which hurdle is higher?
I already explained this. There is no such thing as a free lunch, and by focusing on LIDAR (the cost of LIDAR itself, spending valuable onboard compute on processing and fusing LIDAR data, etc) you are by necessity leaving fewer resources for your vision system. This matters a lot if there are situations where you effectively can't rely on the LIDAR at all, which actually seems to happen a lot in a driving context, assuming you want your vision to work on all roads at all times (that a human could drive).
So yes, if you have infinite resources you should obviously use LIDAR. But we don't so its not nearly as clear.
Yes, I have been making that point for the entirety of this conversation.
1. Tesla needs to figure out how to achieve performance of their system without LIDAR. However since humans do so with two eyes, there is strong precedent that this is possible.
2. Waymo needs to figure out how to self-drive in situations when LIDAR cannot be relied upon. If they cannot do this, their system will never achieve Level 5 (but that's OK! Level 4 taxi service seems like a good business). If they can do this, then it is not clear they ever needed LIDAR in the first place.
One thing I wonder is, if BYD was allowed to compete in the western markets, would the scale of deployment yield better data and thus a better FSD experience?
They either have to bake it into the cost of the car or offer it as an option. I appreciate they offer all three options: don't purchase $0, purchase outright for $8,000, or subscribe for $99/month.
It’s honestly hilarious that they think they deserve access to Tesla’s internal data just because users can view the software version on their own car. That’s like saying a public login screen means the whole system should be open-source. Tesla has every right to protect its own data — especially when it’s tied to proprietary tech and competitive edge. If regulators or media want deeper access, it should be done through proper agreements, not by demanding that confidential info be handed over. You can’t just expect to skip the hard work others have done.
Not sure how you came to that conclusion. The "claim", a black and white, "Tesla redacts everything, everyone else redacts nothing" was never really a claim. But what IS not false:
There is a bunch of data that is "missing", both from Tesla and other manufacturers.
Tesla declines/redacts data far more than other manufacturers, to the point where in many cases, the majority of data is redacted for a given incident.
Yes it is a weak argument, but otherwise I disagree.
Data about a hitting a pedestrian or having an accident isn't proprietary tech. They're not asking for source code, but for data that should arguably be made available for people to see in the interest of transparency and this information is sought consistently from other car makers.
Tesla is of course sticking out like a sore thumb, because they have put the most investment into EV's and "autopilot" features the data might show that they stick out.
Given that Elon wants to torpedo this spending bill over his precious EV credits, I imagine the honeymoon phase is assuredly over and he won't be successful in influencing the administration here.
Google is your friend. While there, please try to find the record of Elon saying he wants more subsidies, as presumably you'd also like to see the record of that.
So your assumption here is that Elon is trying to do some sort of reverse psychology trick, or...?
You can't just make up motivations for people. He has said repeatedly that the government should get rid of subsidies. When the government then gets rid of subsidies, people say he is mad about that. A citation is very much needed.
For reference, [1] is the recent UN regulation for road vehicles to have an event data recording (EDR) function which records certain telemetry about a vehicle for -5 to +5 seconds around a crash event. None of these fields relate to ADS/ADAS. This difference is described at [3] but in summary, EDR telemetry describes what the vehicle physically does, not who or how the vehicle was instructed to operate in that way. EDR telemetry doesn't answer if ADS/ADAS applied the throttle input or whether it was the human operator depressing the accelerator pedal.
Countries take time to decide how to implement the UN regulations so in countries such as Australia, there is (from a quick check) still no regulation requiring light passenger road vehicles to record any telemetry. The US already had a form of regulation requiring limited telemetry about a vehicle for -20 to +5 seconds around a crash event to be recorded.[2] This US regulation also did not require recording of fields relevant to ADS/ADAS.[2]
What this article describes is access to telemetry data that manufacturers such as Tesla are voluntarily recording within vehicles that may include some idea of ADS/ADAS operation during a crash event. For example, Tesla may be recording the human throttle input separate from recording of the ADS/ADAS throttle input, showing whether it was the driver or vehicle who caused the car to accelerate dangerously before a crash. But the UN regulation and older US regulation didn't expect Tesla to record more than just a single throttle position field, ignoring whether ADS/ADAS or the driver directed the throttle position.
May I ask why you are a shareholder? IMO Tesla is headed straight into insolvency with sales collapsing all over the world and factories at <60% capacity, all while the global EV market is surging.
AFAICT the stock is also insanely overvalued, especially compared to ”real” car making companies eg Toyota. The Silicon Valley hype valuation based on future exponential growth seems further and further away from reality every day. P/E anyone?
At this point, why anyone would opt to buy a Tesla is beyond my understanding.
The fact that regulation is lacking to such an extent as to allow Tesla to wait for airbag deployment for something to count as a crash is kind of sad.
That may have been true 4 years ago but it’s far from true now. The only feature Tesla has that no other car seems to have is that minimalist “dentist’s waiting room decorated in the 1990s to seem futuristic” energy from the interior, which definitely sets Tesla apart although not entirely in a good way.
I don’t think this discussion makes sense without pointing to specific models. And discussing specific models doesn’t make sense without establishing what are the set of required properties (which is different for everyone). I was just pointing out that for some people (including me) there are no other cars with similar properties that I need.
I don't understand it either. Anybody I know recently in the UK only got one for political reasons or to stick it to the system in some naive way. I wish I was joking. This is even more sad.
These are the same people who are staunchly opposed to regulating emissions in any way.
So in a way it’s great that they’ve been convinced to buy zero-emissions vehicles by giving them a reactionary edgelord option that’s just like every other EV. (Except for the suicide FSD mode which is more like a Darwin awards filter.)
These studies don't really show a decent comparison between EV's, just car makers in general. We aren't trying to argue whether Tesla's are better than ICE car makers, but whether they're consistent or better against other EV's of other car makers.
However, if you look at the latest press release by JDPower (https://www.jdpower.com/business/press-releases/2025-us-vehi...) you'll see that Tesla now ranks right near the average. Significantly better than in previous years and ahead of other common car makers.
The dependability study could do with a segment purely on EV's, given that EV's as a whole are improving by roughly 33 PP100 per year.
I used the same links as the commenter above, of which I cited a more recent study and shows that Tesla isn’t as bad as they perhaps used to be.
If they really wanted to exclude them they wouldn’t have included them all.
I cited the study, because it was the same study commented above where the most recent study wasn’t mentioned.
I’m not bragging at all, but they’re just not as bad of a car as some people are giving them a wrap for. Heck they had the same number of problem reported as Ford, does that mean Fords are bad cars? No.
What I'm saying is just that Tesla are not "excellent cars and are still ranked at the top compared to the rest of the market" as you said. The fact that they move from the bottom to average still don't support Tesla as excellent cars when compared to the rest of the market. Note that I'm not asserting that "Tesla are the worst cars".
> And even then you’re bragging about Tesla being … average?
If you think about issues this way, I think it's fully unproductive. No one was bragging. The fact you see this as either unthinking bragging or unthinking criticism makes it very hard to talk about what matters: the facts of the matter.
They absolutely were bragging when they said this:
> Tesla's are excellent cars and are still ranked at the top compared to the rest of the market.
The actual facts show otherwise, of course, so now he’s trying the squid ink approach of emitting a lot of verbiage trying to say that average and excellent mean the same thing, but that’s just a distraction from the fact that their first statement was based on brand loyalty, not data.
This is what I mean - my impression is you can't tell the difference between someone commenting on a product and someone boasting about something they made or did. The OP (as far as we know) is not in a position to brag about the general status of Tesla cars, unless they are a senior person at Tesla.
If I said "BMWs are excellent cars" would I be bragging?
I wasn't bragging, I can't brag about them because I don't own a Tesla nor serve to gain any validation or benefit if they do well or not.
A study has found that perhaps I was wrong in the eyes of a dependability study, but I still think they're an excellent car. I still think that if a Ford owner said that they think their Ford Ranger is an excellent car people wouldn't disagree because a dependability study put them as average.
I dislike Tesla as a brand; however, they're not particularly lacking considering their size and price point. JD Power's "ratings" are almost entirely based on surveys and are essentially worthless.
Bruh, JD Power is a joke, they said the Chevy Equinox is the most reliable compact SUV of 2024, look at any of their rankings, it’s all basically random, meaningless noise.
I assume they just create hundreds of oveRlapping categories to make sure everyone has something they will pay the license fee to talk about in a commercial
It depends on the segment and time range and market. MotorEasy conducted a survey in 2024 in the U.K. with 29,967 respondents. (https://www.whatcar.com/news/most-reliable-cars/n27337) The Tesla Model Y was the 9th (equal) most reliable car. However Teslas tend to fare poorly in the Consumer Reports survey in the U.S. I suspect one of the reasons for the discrepancy between this market and the U.S. is that the U.K. received Tesla shipments a lot later for new models - years, in fact. This gave Tesla time to iron out first-model issues. Another is potentially the location of manufacture. Most Teslas sold in the U.K. come from China and Germany. Most Teslas sold in the U.S. come from Fremont, California. There were widespread reports of strange manufacturing practises at the Fremont plan during the covid outbreak, like spray-painting cars in makeshift tents.
Interestingly, MotorEasy found that gas and hybrids were the most reliable. Diesel were the least reliable.
I'd chalk up the low diesel reliability to all the emission systems. Before all the Mercedes D's would run forever, million miles. Even the heavy truck engines could reach a million miles. Now the emission systems including EGR, SCR, and DPF, add injectors, plumbing and electronics. The EGR and DPF systems clog up with soot or burn up from regen
Isn’t Tesla Model S Plaid still ranked the fastest (maybe 1 or 2 $300k super cars are faster) while being the safest and have arguably the most advanced self drive available in the market?
My Model Y is the most reliable car I've ever owned, except it eats tires quickly because it's so heavy. Just one anecdote of course.
I'll never buy another one as long as Elon Musk is associated with the company, but I'd be crazy to sell it now because it's paid for and it's a great car.
They're not actually that great of an EV anymore. The build quality is lackluster and the ride on the Model 3 in particular is quite harsh.
Some of the comments I hear almost universally from prior Model 3 owners when they switch to an Ioniq 5 is how much nicer the ride quality is and how nice it is to have buttons on the dash again.
Supposedly. As far as I understand, whether it was fixed or not will be decided by a completely separate group of people. At some point, Tesla can just lie, and don't expect people will be able to verify.
I think Elon found a glitch it tech-bro reasoning: you promise a product, people believe you and buy it, then they realise that it is not what was advertised, but it is OK, because new version has it now, so you can't complain, it is your fault in the eyes of your peers, you should have made a better research or whatever.
Your FSD has almost got you killed? Don't worry, this is your fault, and anyway the bug was already fixed in the update. Probably. This time for sure.
Sorry but its clear you aren't thinking logically.
The changes in the new 2024 version, specifically changed the suspension setup and you can mechanically verify the difference between the models because there were mechanical changes.
You need to drop the FSD rhetoric, there are an enormous amount of Tesla drivers who couldn't care about FSD, but simply enjoying the car as an EV.
You are thinking logically only if you are a shareholder. Then it makes absolute sense to think like that to stay hopeful.
From the consumer POV who doesn't own shares in the brand, it makes no sense at all.
> enormous amount of Tesla drivers who couldn't care about FSD but simply enjoying the car as an EV
Are there an enormous amount of disappointed Tesla drivers as well who stuck with it because of a sunk cost fallacy?
Should, for example, CyberTruck owners who spend 150K on a bullet-proof car be happy that in the new 2026 version the panels finally will stop falling off?
Should the few million people who purchased Tesla Model 3 feel better that the new version finally drives like a car?
All cars have pain points and whilst people in an existing Model 3 might be annoyed ride quality has been improved in the new model, it is good that the specific pain point has been addressed. Afterall you wouldn't stop improving your car simply because you didn't want to annoy existing customers.
Either way, in prior comments you couldn't come to terms with them possibly fixing something like that in the first place.
You should relax a bit. Model 3 ride was a bit 'sporty' compared to others. Some people like that, other prefer it softer. There really isn't a clear 'right' and 'wrong' here. But on avg. people seem to prefer it a bit softer, so they adjusted it slightly.
Successive version always do slight changes like that over time, that how car companies work. Old owners don't need to be happy about it, or think about it at all.
No it's pretty good now. This isn't a software update. It's either softer or it's not. Of course it is subjective. There are plenty of cars with even softer rides, but they tend to feel a lot more floaty, especially in corners, and I personally don't like that.
Still ugly af. Wouldn't touch it. The only well designed car ever made was the model S, and even that is long in the tooth now and was due a refresh 5 years ago. But CYbeRtrUck ...
Combine that with FSD rubbish bait and switch, nutjob CEO, and the cratering of any brand goodwill ... I won't touch those cars if you paid me (seriously).
And honestly, I recognise the innate progression of the technology ... but others have caught and surpassed to the point that I don't have to worry about buying the lesser car anymore when I go elsewhere.
I have a Model 3 2024 and the "ride quality" is beyond any other vehicle I've been in. I genuinely get a little excited every time I drive it. Best car I've ever been in or drove, and it's not close.
If you want to provide anything other than your isolated impression it'd be helpful to know what cars you're comparing it too. Civics? Audis? Ford F150s?
On the whole seems a little... over enthusiastic...
Not that it should matter but those downvoting your comments have high karma and long histories of making positive contributions here.
Please just make an effort to use HN as it's intended. It's only a place where people want to participate because so many people put effort into keeping it that way.
But they aren't. What Tesla has going for them primarily is the Supercharger network.
The Cybertruck is a complete disaster of a vehicle with so many issues (eg [1]) that the only reason people buy them is to make a political statement from a group that 3+ years ago wouldn't have been caught buying an EV.
Teslas are drivable iPads. Many people (myself included) not only hate this (because it's hard to use without looking) but it's also lazy design. By this I mean, it allows manufacturers to say "we'll fix it with a software update" (and then probably never get around to it) whereas haptic controls require more thought and effort to be put into the UI/UX during manufacturing.
For other Teslas, there have been a host of other issues, some small, some not. For example, the seats were unreliable if adjusted too often so Tesla made an OTA update to limit how much you can adjust the seats to avoid failure [2].
The only thing propping up Tesla sales now are trade restrictions on BYD.
As someone who recently bought a Tesla, nothing comes close for the price to the Model Y in terms of range, performance, trunk space, and software. If you want an EV and value those things, the Model Y is the clear choice.
That sounds like unbelievable hyperbole but I'm not interested in getting into a culture war fight today. It's going to be hard to find a brand which hasn't been associated with bad things before but I wish you luck :)
But also it’s not hyperbole, though I agree it sounds like it.
Association comes in different strengths and forms. But currently owned and operated by a guy actively destroying my country is much more direct and urgent than most of these associations.
But the larger point is that it seems correct that the only people willing to buy a Tesla anymore have conservative politics.
I'm the same. I loved my 2020 Model Y when I bought it new and I love my 2020 Model X that I bought used. Both are the best cars I've ever owned. I know that everyone likes to complain about fit and finish but honestly I never noticed a single thing wrong with either of them. They are performant, silent, efficient, and the MX doors make my grandkids giggle (that's worth quite a bit :).
I keep looking for the next EV to buy but every one that I test drive seems lacking, mostly in the software area. Tesla has the software nailed and nobody else can seem to figure it out. I recently drove a Rivian R1S and really liked it, but the software was not great (for a $100K car) and it really disappointing me.
I really feel like Tesla does great work but their leader needs to go. He's smeared the name so badly that I fear it might never recover.
Yeah I’m hoping the R2 is an option but FSD has become very critical.
I don’t see how Tesla could even dissociate with him, he owns the board and would have to divest also. I think the only path back is if people forget, conservatives suddenly like EVs, or they strike gold with the robots or something.
Yes I think Musk is personally responsible for the cuts he did as part of the thing he loudly and proudly claimed to be in charge of, enabled by the president he claims to have gotten elected.
I’m not sure why you’re so easy to write off 1000s of deaths, but 300,000 so far is one model, but even if it’s very wrong it’s going to be hundreds of thousands sometime this year in all likelihood. https://www.washingtonpost.com/politics/2025/05/28/rubio-aid...
What do you want to sync?
I was able to use my phone book contacts and addresses. Use my phone's WiFi for data and play music or videos from any app on my phone.
I was able to send destination POI to the car GPS to navigate there. Ticked all my boxes.
Bidirectional charging would be great, everything else is already far up there.
I'll compare the top spec Ioniq 5 N line with the Model Y AWD Long Range as they're priced the same here in Denmark. The Ioniq has 520 liters of trunk space vs 854 in the Model Y. The Ioniq has 495km of range vs the Y's 586. Software isn't even close - I have tried both. The Tesla wins hands down on software. The only metric the Ioniq wins on is acceleration. Specifically when using special boost mode from stopped. It can do 0-100 in 3.4s vs the Y's 4.8s, which is fast.
For me this is a clear win for Tesla. Anything under 5s is crazy fast to me anyway, so the other things I mention are worth a lot more.
The top spec actually has a smaller boot and a lower range due to using dual motors. If you care about such things, good news because you can pay less money, surrendering acceleration which you don't care about.
Subjectively, the boot of the Ioniq 5 on the normal models is huge, and I say that as someone for whom boot size was the primary requirement. Moving the seats forward creates epic amounts of space at the cost of reduced leg room for rear passengers.
According to their website the RWD long range (84kw battery) has the same 520L trunk space, 570km range, and only 7.5s 0-100. It’s cheaper but it’s a big downgrade on performance, trunk space, and software.
FWIW, Hyundai seem to totally hide the false floor in the boot, which turns the so-so boot space into a much more substantial and usable space. The AWD variant does not seem to have anything like as much space under the false floor. Perhaps they don't want to bring attention to the deficiencies of the more expensive variant.
AFAIK, the whole industry is really inconsistent on this making comparison of factory #s nearly impossible. Things like Bjorn's banana box test are better as practical measures, IMO. The Ioniq 5 and Y are virtually the same, with the winner being decided by whether the seats can be down or not.
I just watched both videos. I think the difference is in the shape. The Y is very curved. The Ioniq is much more square. This is going to favour a test using very square and rigid boxes. It’s useful to understand that fact but most of the time I’m not hauling only very square suitcases.
That's my point, it isn't "63% extra space". Unless you measured everything yourself. Tesla measures it all the way to the ceiling, Hyundai doesn't. I don't know how much the Hyundai doesn't measure but fairly sure it isn't 63%.
Companies used to play these tricks in the 90s. I am NOT happy that Tesla is trying to bring them back.
That is a fair point. I own a Y and have had both sides to this. Sometimes I carry very irregular cargo that fits perfectly and the space is really incredible.
Other times, I have something that should fit, but can't readily fit through the oddly shaped rear hatch and it becomes a problem.
I'd agree with you with regard to luggage. That's a scenario that really hits the strength of the Y setup, especially with the relatively large frunk. I often end up with a few backpack sized items up there on longer trips.
>If I put all the seats down the boot is bigger in the Y than with the seats up in the Ioniq
That’s great but you’re missing the fact that the Ioniq has a HUD. What does the Tesla have? Everything you’ve mentioned is irrelevant. The Ioniq has CarPlay so you don’t touch the software.
The quote you made is not from me. The Y has substantially more trunk space both with the back seats up and down. If you value a HUD more than range, trunk space, and software, more power to you. I don’t. CarPlay only controls the entertainment and navigation. There are many controls hidden behind a crappy UX in the Ioniq.
I purchased a 2026 Model Y in April. I love it and it’s the best vehicle I’ve even owned. My Apple phone pairs with the car just fine. I can call my contacts, play audio from my phone etc.
Well TMY is really efficient for the size, while I've seen hilarious consumption figures for Ioniq 5. Doesn't matter if you charge from solar, but at European electricity prices, TMY is probably significantly cheaper to run.
Boot-space is literally all it has going for it, you can equal or surpass the other categories for less here in Europe, and if you buy used since used EV's are dirt cheap you could get something significantly better with far better badge-appeal for about the same price.
I have a family and a wife who doesn't pack light so it's a pretty important one for us. There are options if I'm willing to drop some of those things, but the Y gives me all of it for a reasonable price.
Do you mean I could get something significantly better if I choose not to get the things which are important to me? Because that would be worse for me, not better. I have never cared about the badge.
Out of interest, if I reduced the trunk space a little, which similarly priced EVs could you recommend with more range and better performance than the Model Y AWD Long Range?
What gets at least equal mileage and is also not tiny? A friend is looking to buy and EV, and his problem specifically is nothing comes close to Teslas range.
Nothing comes close to Tesla's published range because the numbers they post on their marketing aren't real. They're by far more inflated than any other automaker.
My Model 3 Performance is rated for I think 300 miles. In the real world, it can be as low as 230 miles if it's 30F outside and I'm going 75 mph. If it's warm, I'm still looking at ~270 miles at 75 mph. The colder it gets, the lower your range because cold air is denser (increasing drag) and higher usage of the heater drains the battery.
460 comments
[ 3.0 ms ] story [ 298 ms ] thread[1] https://www.nhtsa.gov/laws-regulations/standing-general-orde...
The nice thing is we can look for ourselves to what extent that is true by downloading the CSV: https://static.nhtsa.gov/odi/ffdd/sgo-2021-01/SGO-2021-01_In...
For example, in the case of BMW, in every single case the field for ADS/ADAS Version is either blank or redacted.
Not true. There are many rows for other manufacturers where fields are redacted or blank.
For example:
- Row 7. BMW. ADAS/ADS Version: blank
- Row 8. BMW. ADAS/ADS Version: redacted
- Row 9. Subaru. ADAS/ADS Version: redacted
etc.
Again, not true.
I just filtered for BMW, and in every single instance, without fail, the ADS/ADAS Version cell is either redacted or blank.
I didn't check other manufacturers.
I don't own nor do I want to own a Tesla, but stuff like this is what gets reported and the corrections or actual facts get buried in the resulting noise. I don't really even care that this is about tesla, even.
If this was some sort of rendering or CSV error on your part, then that could happen at CBS or msnbc just as easily, and tomorrow the headlines scream "Tesla only automaker shirking reporting responsibilities"
Tesla also has a problem of their telematics underreporting crashes. One of the reasons for that is they don’t consider it a crash if airbags don’t deploy. This was called out by the NHTSA in a prior investigation: https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf.
Here’s the relevant paragraph from that report:
> Gaps in Tesla's telematic data create uncertainty regarding the actual rate at which vehicles operating with Autopilot engaged are involved in crashes. Tesla is not aware of every crash involving Autopilot even for severe crashes because of gaps in telematic reporting. Tesla receives telematic data from its vehicles, when appropriate cellular connectivity exists and the antenna is not damaged during a crash, that support both crash notification and aggregation of fleet vehicle mileage. Tesla largely receives data for crashes only with pyrotechnic deployment, which are a minority of police reported crashes.
>and the crash involves a vulnerable road user being struck or results in a fatality, an air bag deployment, or any individual being transported to a hospital for medical treatment.
> Tesla is widely known for its so-called advanced driver-assistance systems, including Autopilot and Full Self-Driving (FSD).
(emphasis mine)
It’s not a term Musk dreamed up.
https://en.m.wikipedia.org/wiki/Advanced_driver-assistance_s...
No point trying to get folks who will move again soon when something else changes in US.
There are ~42k traffic fatalities in the US each year. Cameras-only self driving cars are a tiny fraction of the number of total cars.
The highest estimates I've seen for annual traffic deaths with an ADAS involved (not even implying causation) is in the range of dozens. Cameras-only self driving cars would be a fraction of those. As a result, there are quite possibly more than a thousand traffic fatalities each year caused by human-driven cars for each ONE caused by a cameras-only self driving car.
But my original claim was only that they kill a lot fewer. That seems self-evident.
Polonium ingestion also kills fewer people than self driving cars, so by that token, polonium ingestion is perfectly safe.
Meanwhile, there are ~42k traffic deaths per year that could be prevented by focusing on human-driven cars.
Eh, there are over a hundred thousands deaths attributable to Musk's actions in the White House [1]. I agree with you on the short-term calculus. But trusting Tesla to help reduce those traffic deaths--and furthermore, enabling its position of power--puts those states' sovereignty in jeopardy. Swapping lives for sovereignty is an old (and trusted) trade.
[1] https://www.bu.edu/sph/news/articles/2025/tracking-anticipat...
https://wolfstreet.com/2021/12/13/vehicles-in-operation-and-...
Here's an alternative idea that would save a lot more lives:
Take the camera-based driver attention monitoring that works in my seven year-old Tesla, which notices IMMEDIATELY if I look away from the windshield for more than a second or two, and then require that in the human-driven cars.
Estimates for annual deaths in the US from distracted driving are between 3,250 and 12,400. An in-cabin camera is not expensive or specialty hardware. The tech is there, the costs are low. We could save a lot of lives!
If we're ignoring that to focus on Tesla's FSD, the goal is not sensible regulation or saving lives.
A: Citation? Just because it's less than 0.3% of cars on the road doesn't mean it's less than 0.1% of fatalaties. And citation that doesn't let Tesla pass off any FSD crash as "driver error" which they have a horrible habit of doing. If FSD disengages at impact, they call that driver error, which is absolute bullshit.
And because Tesla is taking 0 accountability for it, they are passing it onto the driver. They want to have their cake and eat it too. If you or I are driving distractedly and kill someone, we face serious criminal and financial repercussions.
If FSD decides to swerve out of the lane and into oncoming taffic, Tesla wants to shrug and say "I guess the driver should have been better". That's trash, and should be banned from our roads summarily.
This. Put it on all vehicles that are driven (exc. waymo, zoox, and the like).
It looks like something similar is already happening in the EU (with momentum in the US too.) (See https://spyro-soft.com/blog/automotive/driver-monitoring-sys...)
This exists already in Subaru vehicles, even ones with ICEs. It's called "DriverFocus." It's super helpful. However, I don't believe the technology is mandated in all vehicles yet.
The driver drowsiness alert in Teslas seems to be much more limited than that. It only activates at speeds over 60km/h, when driven for more than 10 minutes, and when Autopilot is not engaged. I wonder why they disable it when Autopilot is on?
The drowsiness alert would be superfluous when it's watching your eyes. It's already going to yell at you if it can't see your open eyes looking out the windshield for more than a second or two.
Whether or not the "steering wheel torque" method is better than a vision-based driver drowsiness alert is probably debatable, but it would be pretty tough to fall asleep while also applying exactly the right amount of torque to keep Autopilot engaged.
It yells at me sometimes when I'm driving down my drive way and looking at my goats instead of the driveway, which is fair. It also yells at me when I look at the mirrors for 'too long' or if I look for 'too long' for potential cross traffic when crossing an intersection (when driving late at night, I try to look for potential red light runners, but you have to spend more time looking). It also tells my spouse to sit up when she already is. Chances are this alert is going to be disabled, because it's a bigger distraction than anything else.
It also likes to alert about cross traffic when I start moving to sequence after traffic I saw that is in motion. Those alerts would be handy if it were about traffic I didn't see though, so I don't want to turn them off, even though so far they've been unhelpful.
that sounds rough; hopefully they're OK! did the car drive into them from the side or from behind?
where did it happen? googling "Tesla ditch self-driving accident" turns up nothing, but I would have thought it would have made the news.
- Mississippi 23.9 (highest)
- Texas 14.7
- CA 11.3
- Rhode Island and DC 4.8 (lowest)
Anecdotally, the huge trucks and driving culture (aggressive, fast) would have made me guess Texas has higher deaths before seeing the data.
Texas has a plurality of fatal car accidents (for USA), but California is not far behind, and in 2022 California has slightly more deaths. (This page doesn't have the number of fatal car accidents for 2022, which is a bit odd.)
2023 here, Texas has plurality again, with California close behind: https://www-fars.nhtsa.dot.gov/States/StatesCrashesAndAllVic...
You're not looking at absolute numbers, which is what plurality means. I don't see how "someone in the US is more likely to die in a car wreck in Texas even if they never go to Texas" could make sense.
A driver in the US dies while driving due to a crash/wreck/whatever.
Statistically, that occurred with the highest probability in TX. as i said, this was like 2015-2019 when i used to claim this. There's a sign on freeways in TX that say "highway deaths so far in <year>: <16 bit int>" which led me to start looking in to it, and i think my little quip is just a way to draw attention to how dangerous it is to drive in TX. But it is quite large, Texas.
Although there's a good argument to be made that Tesla's entire system has fundamental design flaws which they have negligently disregarded.
I also believe that marketing it as FSD should be liable and scrutinized as a level 4 system. Because when you hear FSD, the public naturally thinks the abilities marked in level 4 arguably even 5.
(Though a consequence is that levels 3 and 4 are very close together in difficulty. We might not see many level 3 cars.)
https://www.sae.org/binaries/content/gallery/cm/content/news...
The time is up to the manufacturer, isn't it?
Mercedes uses 10 seconds right now and that seems pretty good to me. At that point I know it can't be too dire or the car would have already emergency stopped.
The time depends on how quickly an event unfolds in traffic. You can't guarantee 10s notice for an event that is imminent in 2s and the system might not be able to handle or can't detect.
The car could become temporarily "blind" for some reason with just 4-5s to brake before a collision. It's enough for a human driver even considering reaction time. But it's impossible to guarantee a minimum time without the ability to predict every issue that will happen on the road.
If it becomes "blind" because of an unexpected total system failure, that's an exception to the guarantee just like your transmission suddenly exploding is an exception. It had better be extremely rare. If it happens regularly then it needs a recall.
When dealing with unpredictable real life events there are no guarantees, unless we're considering the many carveouts to that definition from a legal perspective. A blind car (fluke weather, blown fuse, SW glitch, trolley problem) can no longer guarantee anything. Giving the driver 10s, or assuming the worst and braking hard could equally cause a crash.
> your transmission suddenly exploding is an exception
As long as the brakes or steering work a driver could still avoid a crash. The driver having a stroke is closer to a blind car.
The guarantee here is that the human isn't obligated to intervene for a moment.
If you call that guarantee impossible, then what about level 4 cars? They guarantee that the human isn't obligated to intervene ever. Are level 4 cars impossible?
Is this a wording issue? What would you say level 4 cars promise/provide? Level 3 cars need to promise/provide the same thing for a limited time. And that time has to be long enough to do a proper transfer of attention.
Ah, understood. So the guarantee is that the driver is not legally responsible for anything that happens in those 10s. I always took that as a guarantee of safety rather than from legal consequences.
The guarantee is that you will be very safe and you can go ahead and look away from the road and pay attention to other things. But at most this is as good as a level 4 or 5 car, not an impossibly perfect car.
Yes but there is a minimum time (if a bit under-specified)
> "At Level 3, an ADS is capable of continuing to perform the DDT (Dynamic Driving Task) for at least several seconds after providing the fallback-ready user with a request to intervene."
J3016 Section 3.12, Note 3: https://wiki.unece.org/download/attachments/128418539/SAE%20...
Which makes me think: if FSD requires constant hands on steering wheel and concentration what is the point? May as well drive yourself.
The highest level of ADAS system I use regularly has facial attentiveness tracking. If you spend too much time drinking coffee or even looking out the sides of the car it will alert you and eventually turn off. So you're not spending a ton of time drinking coffee or reading emails.
It's really nice having the car just want to stay in the center of the lane and keep the following distance all on its own. It's less fatiguing on your hands and arms having the car feel like it's in a groove following all the curves for you instead of resisting your input all the time for hours and hours. It's incredibly nice not having to switch between the brake and the gas over and over in stop and go traffic. Instead, the only thing I need to focus on are the drivers around me and be ready to brake.
I've driven between Houston, Dallas, and Austin dozens of times with ADAS systems and another dozen or so times with only basic cruise control. It's way nicer when the only time I have to touch the gas and brake are getting on and off the highways. I'm considerably more relaxed and less exhausted getting to my destination.
Let's assume all these options are either the same price or an immaterial difference to the price of your next car. If you had an option for a car with basic cruise control or no cruise control, which one would you take? If the option was basic cruise or adaptive cruise which kept pace with traffic and operated in stop and go conditions, which would you choose?
And then there's the skill atrophy. How do you learn to perform in stressful situations? By building up confidence and experience with constant repetition in more mundane ones, which this robs you of.
This isn't correct. Level 4 doesn't require driver intervention[0]. Hence, why I'm arguing "Full Self Driving" starts here, level 4.
So now if you're explicitly not required to "be present" then the system should be liable or at least the "driver" isn't to blame directly.
It's actually level 5 is the same as level 4 but add heavy rain, snow, ice, name-your-adverse-condition.
[0] https://www.sae.org/blog/sae-j3016-update
Level 2 requires the driver to choose whether to intervene at all times. This is an unreasonable task for humans.
Level 3 puts the car in charge of when intervention is needed, and even once it wants intervention it still has to maintain safe control for several seconds as part of the system spec.
Level 4 puts the car in charge of when intervention is wanted, but you can refuse to intervene and it has to be able to park itself.
So I will double down on my claim. Until the car requests intervention AND the timer runs out, level 3 and 4 are the same. They require the same abilities out of the car. And that section of time, between wanting intervention and getting intervention, is the hardest part of level 3 driving by far. If you can solve that, you're 90% of the way to level 4.
A level 3 car has to be able to handle emergencies several seconds long, and turning it into level 4 is mostly adding the ability to park on the shoulder after you get out of the initial emergency.
The gap between 4 and 5 is a bunch bigger. A level 4 car can refuse to drive based on weather, or location, or type of road, or presence of construction, or basically anything it finds mildly confusing. 5 can't.
I edited a bit for clarity, but also I'll append a thought experiment as an extra edit:
A level 3 car with an hours-long driver intervention timer is basically identical to a level 4 car.
If you have a 0 second intervention timer, you're barely better than a level 2 car.
How long does the timer have to be before developing your level 3 system is almost as difficult as a level 4 system?
I don't think it's very long.
I still stand by level 3 != level 4 in terms of real world liability.
Level 3 allows too much wiggle room and sloppiness to be able to legally shift liability away from the driver. At that point you're playing that "intervention period" length. Manufacturers claiming Level 3 will want to lower it as much as possible, regulators raise it. To me, Level 3 simply shouldn't exist.
Only at Level 4 is the expectation, without a doubt, the machine is in control. A person in the driver seat is optional because the steering wheel and pedals are as well. When people bought "Full Self Driving" they seriously believe "when can I go to sleep?" ability is where it belongs, which always put the expectation at Level 4.
It looked like the Mercedes system is 10 seconds which seems like plenty to me.
And while it would be nice to sleep I'll be pretty happy just looking away from the road.
I've clocked nearly a half million miles on the road (I'll be there sometime in the next 9 months), and the range of technical ability you need to drive in just the US, no, scratch that, any given state or even county varies so much and potentially so often that FSD is just a lie to sell cars. I'm willing to upload a full hour drive touring a few parishes around here in my quite heavy Lexus, front and rear cameras, just to prove my point. I'd do it in the subaru but the dashcam isn't very good and also it's lineage is rally so it exaggerates how poor the roads are. My YouTube has dashcam footage of drives that I'm willing to bet no automated system could handle, even if it claimed to be "level 5". Driving after a storm or hurricane is another issue. I know the hazards in general and specifically for the areas I'd need to travel during or after an emergency. I cannot fathom the amount of storage and processing that would take, to have that for every location with roads. On board, in the car? Maybe in 20 years.
Doing some napkin math, with 4 million miles of road in the US, if you wanted to store 1KB of data per meter of road, hundreds of data points, you'd only need 7TB for the entire database.
And the processing to make it shouldn't be anything special, should it? Collection would be hard.
Currently that would probably cost ~$500 per car to implement based on retail pricing of 8TB SSDs. It would need to be updated constantly, too, with road closures, potholes, missing signage, construction. With an external GPS unit like a tomtom, they had radio receivers in the power cord that tuned to traffic frequencies, if available, and could route you around closures, construction, and the like, so you need a nationwide network to handle this. Cellphone won't cut it. Starlink might, but regardless, you need to add that radio and accoutrements to the BOM for each car.
and i'm not talking about the processing of the dataset that gets put onto the 8TB SSD in the car; i am talking about the processing of the data on the 8TB SSD on the car while at speed.
furthermore, i am fairly certain that it would take, on average, more than 1.6MB per mile to describe the road, road condition, hazards, etc. a shapefile of all roads in the US - that which gets one closer to knowing where the lanes are, how wide the shoulders are, etc is 616MB. and it's incomplete - i put in two roads near me with fairly unique names and neither are in the dataset. So your self driving car using these GIS datasets won't know those roads.
I had an idea to put an atomicpi in my car, with two cameras. it has a bosch 9-dof sensor on the board, coupled with the cameras you can map road surface perturbations, hazards, and the like, which i believe will be much more than 1KB per meter, especially as you need "base" conditions and updates and current conditions (reported by the cars in front of you, ideally). the csv GIS dataset looks like this:
>OBJECTID,ID,DIR,LENGTH,LINKID,COUNTRY,JURISCODE,JURISNAME,ROADNUM,ROADNAME,ADMIN,SURFACE,LANES,SPEEDLIM,CLASS,NHS,BORDER,Shape__Length
> 568143,964990,0,0.07,02_36250355,2,02_39,Ohio,S161,DUBLIN GRANVILLE RD,State,Paved,4,88,3,7,0,0.000759951397761616
and i ran, for example `awk -F, '/PACIFIC COAST/ {sum += $4} END {print sum}' NTAD*.csv` and it spat out 79.04, which i think is a bit shorter than reality. Looks like the dataset i pulled is only "major roads" as well - but that doesn't explain 79.04 as the sum of lengths of all rows with "PACIFIC COAST" in them. It does show the total length of interstate 10 is 3986.55, which is roughly double what the actual length is (2460mi), so perhaps i'm just not understanding this dataset.
Anyhow 600+ MB for just that sort of information (plus shapes) for only a really quite small subset of roads in the US.
anyhow my thoughts are scattered, this input box is too small, and i'm not really arguing. Maybe it is possible, but it will raise the price thousands of dollars per auto, you need infrastructure (starlink will work) to update the cars, and so on. I'm prepared to admit i am wrong, but your comment didn't move the needle for me.
also, just to be fun, which self driving car could manage this entire drive? https://www.youtube.com/watch?v=sNqFN7KeOYE
> i am talking about the processing of the data on the 8TB SSD on the car while at speed.
I'm not worried about that. The actual driving takes such powerful computers that even if there was a petabyte of total data, the amount the car would have to process as it moves would be a trickle in comparison to what it's already doing. Max 50KB per 10 milliseconds. And obviously the data would by sorted by location so there's very little extra processing required.
But you tell me, how many data points do you think you need per meter of road?
I really don't think you need millimeter-level surface perturbations all the way across. Mapping the precise edges of the road and lanes should only need dozens of data points, 4 bytes each. And then you can throw a few more dozen at points inside the lanes to flesh it out. You can throw a hundred data points at each pothole without breaking a sweat. Measuring the surface texture in various ways and how it responds to weather is only going to take a handful of bytes per square meter, in a way that repeats a lot and is easy to compress.
> 568143,964990,0,0.07,02_36250355,2,02_39,Ohio,S161,DUBLIN GRANVILLE RD,State,Paved,4,88,3,7,0,0.000759951397761616
That's an extremely inefficient format. Unnecessary object ids, repeating metadata over and over, way too many decimal places, and all stored as text.
But even then, your database is so tiny compared to the size I suggested that I don't think we can extrapolate anything useful. Even if we 4x it or whatever to compensate for a lack of rural roads.
none of this has to do with visual or proprioception. It's knowing "every inch" of road. It's knowing how far i can leave the center of the lane if someone else crowds me or goes over the center divider, because the shoulder is soft through here because logging trucks have been exiting the forest onto the highway. It's knowing what part of I-605 floods - not the whole thing, some lanes, some places, and "flood" means 2+ inches of water on the road surface, hitting it at speed makes a tidal wave flying into other lanes. If someone hits that in front of you, you're blind for a couple of seconds minimum. If we want to have semi trucks be "FSD" it needs to know, for the traffic and other conditions, how fast to go and what gear to be in to climb each hill, and then the hazards that are over the hill - that a trucker would know. Where's the gravel bed on more mountainous passes? Or more simply, what time of day neighborhoods are more likely to have people approaching or going through / out of intersections, blind or otherwise. How many "bytes" is that information, times every neighborhood? If many cars brake at the same place, there's probably a reason, and that needs to be either in the dataset or updated somehow if conditions change. You ever used Waze and had a report of something on the road or a cop parked somewhere, and it's nowhere to be seen? And that's updated much more frequently than the radio-info on the GPS systems i referred to earlier. Some roads become impassable in the rain, some roads ice more readily.
If this was easy/simple/solved, waymo et al would be bragging about it, the tech in their cars. Waymo (or the other one) specifically, because they cover less than 0.1% of road surfaces in the US, in some of the most maintained and heavily traveled corridors in the world. So, if anyone from a robotaxi company happens by and knows roughly how much storage is needed for <0.1% of the road surfaces in the US, then we could actually start to have this dialog in a meaningful way. Also i am unsure how much coverage robotaxis actually have in their service area. A "grid system" of roads makes mapping and aggregate data "simple", for sure.
This reminded me a bit of the idea that somewhere in the US there's a database of every sms sent to or from US cellular phones. "it's just text; it'll compress well" - belies how much text there is, there.
for reference, the map in my lexus is ~8GB, for the US. And that's just "shapes" and POI and knowing how the addressing works on each road. It doesn't know what lane i'm in, it doesn't track curves in the road effectively (the icon leaves the road while i'm driving quite often), and overpasses and the like confuse all GPS systems i've ever used - like in Dallas, TX where it's 4 layers high and parallel roads stacked. furthermore, just the road data on google maps for the nearest metro area to my house is 20MB. i have a recollection it goes real quick into hundreds of MB if you need to download maps for the swaths of areas where there is no cellphone reception, like areas in west...
The part of the computer that knows how to drive is completely separate from the 7TB database of the exact shape and location of every lane and edge and defect.
> knowing how far i can leave the center of the lane if someone else crowds me or goes over the center divider
Experience, not in the database.
> knowing what part of I-605 floods
> Where's the gravel bed on more mountainous passes?
That goes in the database but it's less than one byte per meter.
> How many "bytes" is that information, times every neighborhood?
I don't know why you would want that data, you should be wary of blind traffic at all times, but that's easy math. There's less than a million neighborhoods and time based activity levels for each neighborhood would be about a hundred bytes. So: Less than 1 byte per meter and less than 100MB total.
> If this was easy/simple/solved, waymo et al would be bragging about it
This doesn't happen for two reasons. One they are collecting orders of magnitude more data than road info, two like I keep saying the collection is extremely difficult and I'm only defending the storage and use as being feasible.
> This reminded me a bit of the idea that somewhere in the US there's a database of every sms sent to or from US cellular phones. "it's just text; it'll compress well" - belies how much text there is, there.
Well we know how many meters of road there are. So it's basic multiplication.
I can tell you how many hard drives you need to store a trillion texts. It's five hard drives.
Google thinks the human race sends almost ten trillion text messages per year. So I guess you could store them all very easily? Why do you think it's not doable?
> Your claim is that a mere 1000x more data is enough to autonomously self-drive anywhere in the US, at any time of day or year, etc...
My claim is that 1000x is enough for utterly exhaustive road maps. Figuring out how to drive is another thing entirely.
an SMS isn't just "140 characters/bytes" or whatever (i honestly don't care what your definition of "SMS" is). Of course you could fit 140 characters * 1e12 onto 5 hard drives. Where are you going to put the 1PB (for 1e12, but your own cite says it's 1e13, so 10PB) of metadata, minimum? the most barebones amount of metadata you need to actually have actionable "intelligence" is 1KB per message (technically i was able to finagle it to ~1016 bytes.) And that's for every message, even an SMS that is the single character "K".
you need the metadata to derive any information from the SMS. "Lunch?" "yeah" "where?" "the place with the wheel" "okay see you in 25, bring Joel" This is what you propose to save. (quick math shows you went off something like ~32TB of sms data per 1e12 messages)
in the same way that you propose that the shapes of a road and it's direction and distance "plus 1KB of metadata per meter" is enough to derive the ability to drive upon those roads.
It's pretty obvious that just using sensors is not going to get FSD. Maybe in the next 20 years we will develop sensor technology (and swarm networking and whatever else) that will allow us to dispense with the "7TB" of metadata. My argument is that: we need much more "metadata" than 1KB per meter to "know the road baseline, current conditions, hazards", much in the same way a text message is more than 140 bytes. Driving with "only sensors" and rough GPS has killed people. It does not matter if human drivers have more death per million miles or whatever, because i am strictly talking about FSD, what other people are calling level 5 (i'd even concede level 4; although i wouldn't be able to use a level 4 car where i live for roughly 1/4th the year - and other areas would have more than 1/4th the year.)
enjoy your night!
note: the metadata for a meter of road could be:
How did you reach that number?
I figure the most important metadata is source and destination phone numbers and a timestamp, and I guess what cell tower each phone was on. A phone number needs 8 bytes, and timestamp and cell tower can be 4 bytes, so that's 28 bytes of important metadata.
> (quick math shows you went off something like ~32TB of sms data per 1e12 messages)
I was going for a full 140TB of data. 20-30TB hard drives are available.
I did consider metadata, but I figured you could probably put that in the savings from non-full-length messages.
> Where are you going to put the 1PB (for 1e12, but your own cite says it's 1e13, so 10PB) of metadata, minimum?
Well for just the US it would be closer to 1PB. But, uh, I'd store it in a single server rack? (ideally with backups somewhere) As of backblaze's last storage pod post, almost three years ago, it cost them $20k per petabyte. That's absolutely trivial on the scale of telecomms or governments or whatever.
> My argument is that: we need much more "metadata" than 1KB per meter to "know the road baseline, current conditions, hazards", much in the same way a text message is more than 140 bytes.
I mean, I agree with you about needing extra information.
But that's why the number I gave is 10000x larger than your CSV. My number is supposed to be big enough to include those things!
> note: the metadata for a meter of road could be:
I really appreciate the effort you put into this. I have two main things to say.
A) That's less than a kilobyte of information. Most of the bytes in the JSON are key names, and even without a schema for good compression, you can replace key names with 2-byte identifier numbers. And things like "critical" and "Active roadwork zone with lane closure" should also be 1-byte or 2-byte indexes into a table. And all the numbers in there could be stored as 4 byte values. Apply all that and it goes down below 300 bytes. If you had a special schema for this, it would be even lower by a significant amount.
B) Most of those values would not need to be repeated per meter. Add one byte to each hazard to say how long it lasts, 0-255 meters, instant 99% savings on storing hazard data.
> each "trick" you do to reduce the size of the metadata necessarily implies more CPU needed to parse and process it.
CPUs are measured in billions of cycles per second. They can handle some lookup tables and basic level compression easily. Hell, these keys are just going to feed into a lookup table anyway, using integers makes it faster. And not repeating unchanged sections makes it a lot faster.
a phone number is not a 64 bit integer, like, just off the top of my head, a phone number can start with "0"
and again - if you use clever tricks to reduce this, you increase the overhead to actually use the data.get a celltower snooper on your phone and watch the data it shows - that's the metadata for your phone. SMS dragnet would need that for both phones, plus the message itself.
It's not an integer. But you can store it inside 64 bits. You can split it into country code and then number, or you can use 60 bits to store 18 digits and then use the top 4 bits to say how many leading 0s to keep/remove. Or other things. A 64 bit integer is enough bits to store variable length numbers up to 19 digits while remembering how many leading zeros they have.
If you want really simple and extremely fast to decode you can use BCD to store up to 16 digits and pad it with F nibbles.
> JSON
Most of this is unimportant. Routing path, really? And we don't need to store the location of a cell tower ten million times, we can have a central listing of cell towers.
I don't think we really need both phone number and IMEI but fine let's add it. Two IMEI means another 16 bytes. And two timestamps sure.
Phone number, IMEI, timestamp, cell tower ID, all times two. That's still well under 100 bytes if we put even the slightest effort into using a binary format.
> and again - if you use clever tricks to reduce this, you increase the overhead to actually use the data.
No no no. Most of the things I would do are faster than JSON.
even if that 80% was 99%, that last 1% will be the cause of some mishaps.
my subaru is within a few percent of 80% FSD if everything is turned on. I still technically have to hold the wheel, but the steering only shuts off about 20% of the time with that being met.
I’ve heard rumors of that happening.
That amount of time is also nowhere near enough for a human to switch tasks like that. I would say it should be at least a minute or two, and even that is pushing it with how people are bound to use the system (spacing out while watching a TV show, etc).
Let's say you were parked, taking a nap in the drivers set - when you woke up, would you immediately start driving or would you wait a minute to get your bearings? How about at a highway rest stop? It feels like trying to push back on that by asserting "the driver is supposed to always be alert and ready to be attentive!" is another bout of fanciful fiction like L2. Being outright asleep would seem to be derelict, but I can imagine many fuzzy mental middle grounds, especially in a droning car.
(Somewhat related, if I haven't driven in a month I would say it takes me tens of minutes (maybe 20?) to get back up to the usual groove. Obviously the only way to do it is to do it, but I drive much differently until then)
It would take me a minute to wake up. Napping is one of the few things you can't do with level 3.
> How about at a highway rest stop?
Wait, doesn't this go against your argument?
Someone at a rest stop can go from "not driving at all" to "full speed down the highway and merging" in a handful of seconds. And it works fine.
> It feels like trying to push back on that by asserting "the driver is supposed to always be alert and ready to be attentive!" is another bout of fanciful fiction like L2.
Depends on what "alert" means. I can be in a general readiness state, with no particular requirements on my focus, for hours on end.
> Being outright asleep would seem to be derelict, but I can imagine many fuzzy mental middle grounds, especially in a droning car.
"The boring droning car made me zone out" is something that can happen while you're driving. A TV show could actually reduce the risk of falling half asleep.
> if I haven't driven in a month I would say it takes me tens of minutes (maybe 20?) to get back up to the usual groove. Obviously the only way to do it is to do it, but I drive much differently until then
How much differently? Also an autonomous car asking for takeover is probably driving super cautiously too.
It's really not inhuman levels of constant vigilance, not anymore than actually driving the car regularly. I just don't have to actively be keeping the pedals just right to maintain the following distance myself and I don't have to constantly fight the wheel.
If we're going to allow companies to write code in which human safety is in danger if that code misbehaves, that code should be auditable by a 3rd party, and those audits should regularly happen.
Code which affects the safety of humans should be reviewed with AT LEAST as much rigor as code for slot machines.
^ reading anecdotes about accident scenes, someone if not everyone is always lying about what happened.
In one case there was a claim the driver was in the backseat. This got widely published in all media outlets. And turned out to be complete nonsense, it wasn't even in autopilot.
But of course it could be true but I would wait for the data.
FSDs failures are either far more boring (imagining a stop sign) or put's the user in danger (driving onto train tracks).
If that’s the case they should show it.
With how popular Musk is these days I can 100% where Tesla is coming from here.
Like Tesla?
If your airbags don't deploy, Tesla doesn't consider it an accident for the purposes of reporting (modern safety systems don't blindly deploy airbags, they evaluate g-forces, speeds, angles of impact, etc., so you can hit something at 25mph and the vehicle decides your seatbelts are sufficient. Tesla decides "that's not a reportable collision"). Know when else your airbags might not deploy? Very serious accidents, when hardware or controllers are damaged.
Speaking of which, fatalities are not included in that report. "It was a collision where someone died, but doesn't merit inclusion in a safety report" is a weird position to take.
The actual article is about how Tesla claims that providing this data would be a competitive disadvantage that rivals could use.
Would we accept Pfizer releasing a new pill without evidence?
“It’s better at preventing heart attacks than anything else. But we can’t show you data, that would hurt our competitive advantage.”
https://www.reuters.com/legal/government/wait-what-fda-wants...
> The records must be reviewed to redact “confidential business and trade secret information of Pfizer or BioNTech and personal privacy information of patients who participated in clinical trials,” wrote DOJ lawyers in a joint status report, filed Monday.
...
> But we can’t show you data
Tesla shows the data to the NHTSA whose experts look at it and can force recalls so your analogy and argument make no sense.
Literally felt like the difference between flying a helicopter (actively trying to kill u lol) and an airplane.
I honestly did not get the hype until this specific HW4/v12 combination which didn't exist until last summer or so. It's the first time FSD felt like a safety feature for just $99 a month.
That's exactly the problem. It's great right until it isn't, at which point it's likely to make a decision that will kill you or someone else if you aren't lucky.
(most) Humans are REALLY good at paying attention to something that will actively kill them at any moment - you don't see a lot of people running a chainsaw while sending a text to their friend about drinks later in the day.
Humans are REALLY bad at stopping something they trust (IMO foolishly), with less than a half a second of notice, from killing them or someone else. It is completely natural to get lulled into a sense of security when something mostly works exactly as you'd expect.
Meanwhile Tesla wants to act as if it's the driver's fault anytime there's a crash without acknowledging they are actively perpetuating the myth of: "this thing drives itself". It's literally called "Full Self Driving" and Telsa expects the average person to look at that name and think: you need to be vigilant anytime you turn this on because it is a beta feature that may drive into oncoming traffic at any moment.
So for example, if I look at the screen, my phone, or start day-dreaming for even a few seconds, it'll beep and quickly strike me out from using FSD. "FSD (supervised)" is how it shows up in the UI too at least giving some expectation of it not being autonomous.
So in practice, I'm picturing the right driving inputs and watching what it's doing.
I haven't tested it but I assume the same is true if you put tape over the camera
An unsubstantiated claim given that there are many, many safe human drivers who have neither LIDAR sensors nor hyper-accurate pre-mapping at their disposal.
Entertaining a No True Scotsmen is a bit of a silly exercise anyway, but this semantic game is extra silly.
The gap between humans and computers is enormous, not some weird gotcha tactic.
No True Scotsman was obviously in reference to GC, not you.
> No True Scotsman was obviously in reference to GC, not you.
I'm unsure what they said that would qualify. Was it adding "true unsupervised"? I think that's a fair qualification, because most of the point of self driving is lost if I can't look away from the road.
Note Waymo announced a partnership with Toyota, pretty hand wavy, but at least it seems there’s hope the technology may come to regular car owners at some point.
I get that the vast majority aren't car enthusiasts and that's ok, but there is actual pleasure in driving.
And even me - when I want a rest from it - lane assist and cruise control are MORE than enough. I can even add these two to old classics without much bother.
Going all in on autonomy doesn't interest me at all.
Polarised sunglasses. Works on my Subaru. Works on my buddy’s Tesla.
Tesla can't help but know that the supposed non-driver getting constantly nagged to be vigilant as if actively driving destroys most of the value proposition of FSD. Vision Attention Monitoring has quite a bit of potential to be very useful … precisely in situations in which vehicles are not driving themselves.
This should be weighed against the fallibility of human drivers, surely? Our point of comparison is not "perfect", it's "human." Inasmuch, with millions of miles driven, FSD appears to be many times safer than humans: https://www.notebookcheck.net/Tesla-Autopilot-and-FSD-are-no...
Not perfect, and there will be crashes, but much better, and I think that's the yard stick we should be using, because no system will ever be perfect.
Also to note, FSD disengagements are probably common enough to still be on the left hand side of the "Valley of Degraded Supervision"[2], where mistakes are common enough that users stay viligant. As mi/de increases to 5,000 or 50,000, the quality of the supervision could degrade to the point that the supervised system is less safe than an unaided driver.
[1] https://teslafsdtracker.com/Main
[2] https://users.ece.cmu.edu/~koopman/pubs/koopman19_TestingSaf...
That said, something being excessively baked does not mean it is good.
The point is - it didnt deliver, and still doesnt. Its a securities fraud out in the open, but clearly from a guy who is above the threshold of applicable law
Is this statement actually true? From what I heard, it was designed by highly overworked stressed engineers working in pretty bad workplace conditions. They work there, because there are not many other places to work at and doing similar work.
And their primary goal was to produce as fast as possible.
> It is very thoroughly tested.
Is it?
I won't trust Tesla FSD for that reason alone. You've got a way to have far more reliable distance sensing and you choose not to use it? That's crazy.
As soon as you buy into LIDAR (or any other non-human sensing mechanism) you introduce the possibility that your solution could have strict upper bounds on scale.
Which is indeed what we have seen so far with Waymo – it may have fewer issues per mile than Tesla FSD, but every one of those miles is expensively pre-mapped and requires $30k of extra hardware on every vehicle.
And for the use-case of "driverless taxi service in a well-established metro area", the Waymo solution does currently seem to be the better one. But that is also clearly not the ambition Tesla is pursuing.
The $30k figure you're giving for Waymo's LIDAR cost is quite old. As often happens, a new technology got cheaper over time.
"But LIDARS that are $5,000 today are forecast to be under $200 when bought by the millions, meaning the bill-of-materials for the extra hardware should drop below $2,000, and even $1,000, in time."
https://www.forbes.com/sites/bradtempleton/2024/08/20/waymos...
In perfect conditions LIDAR is helpful, but what about the million situations where conditions are not perfect? e.g. what about rain/snow/fog? What about telling the difference between a cardboard box and a metal one? If you want a vehicle that can operate in all conditions and situations on the road today, your vision component needs to be incredibly robust to the point of (possibly) obviating the need for LIDAR.
That $30k figure is actually the lower bound on the current Waymo fleet as it exists today. The $5k figure from the article is the estimated cost per LIDAR sensor (there are 4 on a Waymo vehicle) if you bought them today, which is still $20k total. Additionally the vehicle cited in the article isn't on the road yet, and therefore seems a bit premature to cite as a reference for real-world cost. Imagined future costs are just that, imagined. Those are uninteresting in the same way that Tesla's promises of vision-only capability are uninteresting. What matters is what is actually achieved.
If Waymo achieves the range of driving conditions you can currently operate Tesla FSD, I will be impressed. Likewise if Tesla achieves the safety and consistency per mile that Waymo has, I will be impressed. The question is: which hurdle is higher?
Why would you handicap yourself by leaving available data on the floor?
So yes, if you have infinite resources you should obviously use LIDAR. But we don't so its not nearly as clear.
1. Tesla needs to figure out how to achieve performance of their system without LIDAR. However since humans do so with two eyes, there is strong precedent that this is possible.
2. Waymo needs to figure out how to self-drive in situations when LIDAR cannot be relied upon. If they cannot do this, their system will never achieve Level 5 (but that's OK! Level 4 taxi service seems like a good business). If they can do this, then it is not clear they ever needed LIDAR in the first place.
Why are they hiding the results?
All this effort and brilliance is demonstrably not enough for the software to do what it’s supposed to do.
Are you hearing yourself.
How can a "safety feature" be a subscription? Next they'll charge you a microtransaction every time you fasten your seatbelt?
There is a bunch of data that is "missing", both from Tesla and other manufacturers.
Tesla declines/redacts data far more than other manufacturers, to the point where in many cases, the majority of data is redacted for a given incident.
Data about a hitting a pedestrian or having an accident isn't proprietary tech. They're not asking for source code, but for data that should arguably be made available for people to see in the interest of transparency and this information is sought consistently from other car makers.
Tesla is of course sticking out like a sore thumb, because they have put the most investment into EV's and "autopilot" features the data might show that they stick out.
https://www.ntsb.gov/investigations/Pages/HWY18FH011.aspx
https://www.ntsb.gov/investigations/Pages/HWY18FH004.aspx
https://www.ntsb.gov/investigations/Pages/HWY16FH018.aspx
Battery fire
https://www.ntsb.gov/investigations/Pages/HWY21FH011.aspx
Trump may not care about reelection but congressmen do.
You can't just make up motivations for people. He has said repeatedly that the government should get rid of subsidies. When the government then gets rid of subsidies, people say he is mad about that. A citation is very much needed.
Countries take time to decide how to implement the UN regulations so in countries such as Australia, there is (from a quick check) still no regulation requiring light passenger road vehicles to record any telemetry. The US already had a form of regulation requiring limited telemetry about a vehicle for -20 to +5 seconds around a crash event to be recorded.[2] This US regulation also did not require recording of fields relevant to ADS/ADAS.[2]
What this article describes is access to telemetry data that manufacturers such as Tesla are voluntarily recording within vehicles that may include some idea of ADS/ADAS operation during a crash event. For example, Tesla may be recording the human throttle input separate from recording of the ADS/ADAS throttle input, showing whether it was the driver or vehicle who caused the car to accelerate dangerously before a crash. But the UN regulation and older US regulation didn't expect Tesla to record more than just a single throttle position field, ignoring whether ADS/ADAS or the driver directed the throttle position.
[1] UN Regulation No. 160 - Event Data Recorder (EDR) - https://unece.org/sites/default/files/2023-10/R160E.pdf
[2] CFR Title 49 Subtitle B Chapter V Part 563 - https://www.ecfr.gov/current/title-49/subtitle-B/chapter-V/p...
[3] https://unece.org/fileadmin/DAM/trans/doc/2019/wp29grva/GRVA...
VW, BYD or Toyota or judged on how many cars the sell. What is Tesla judged on?
But then I remember that investors just want to dump their shares on the next sucker they don't really care about the underlying business case.
Like who cares if software version numbers are released on crash reports or not?
Me giving them capital helps them keep more people from ending up writing boilerplate code for fintech startups and other fairly fruitless endeavors.
So in a way it’s great that they’ve been convinced to buy zero-emissions vehicles by giving them a reactionary edgelord option that’s just like every other EV. (Except for the suicide FSD mode which is more like a Darwin awards filter.)
The only part I don't know why people would trust is the FSD/Autopilot of which I wouldn't recommend people to buy. But as an EV its an excellent car.
However, if you look at the latest press release by JDPower (https://www.jdpower.com/business/press-releases/2025-us-vehi...) you'll see that Tesla now ranks right near the average. Significantly better than in previous years and ahead of other common car makers.
The dependability study could do with a segment purely on EV's, given that EV's as a whole are improving by roughly 33 PP100 per year.
You are citing a study that specifically excludes Tesla. And even then you’re bragging about Tesla being … average?
If they really wanted to exclude them they wouldn’t have included them all.
I cited the study, because it was the same study commented above where the most recent study wasn’t mentioned.
I’m not bragging at all, but they’re just not as bad of a car as some people are giving them a wrap for. Heck they had the same number of problem reported as Ford, does that mean Fords are bad cars? No.
If you think about issues this way, I think it's fully unproductive. No one was bragging. The fact you see this as either unthinking bragging or unthinking criticism makes it very hard to talk about what matters: the facts of the matter.
> Tesla's are excellent cars and are still ranked at the top compared to the rest of the market.
The actual facts show otherwise, of course, so now he’s trying the squid ink approach of emitting a lot of verbiage trying to say that average and excellent mean the same thing, but that’s just a distraction from the fact that their first statement was based on brand loyalty, not data.
If I said "BMWs are excellent cars" would I be bragging?
A study has found that perhaps I was wrong in the eyes of a dependability study, but I still think they're an excellent car. I still think that if a Ford owner said that they think their Ford Ranger is an excellent car people wouldn't disagree because a dependability study put them as average.
> you'll see that Tesla now ranks right near the average. Significantly better than in previous years and ahead of other common car makers.
That was my mistake
Which is an absolutely asinine axis of comparison seeing as we live in a reality where different brands take different product lines seriously.
Interestingly, MotorEasy found that gas and hybrids were the most reliable. Diesel were the least reliable.
I'll never buy another one as long as Elon Musk is associated with the company, but I'd be crazy to sell it now because it's paid for and it's a great car.
Some of the comments I hear almost universally from prior Model 3 owners when they switch to an Ioniq 5 is how much nicer the ride quality is and how nice it is to have buttons on the dash again.
There have been comments around the ride harshness of the model 3. In the latest model it has been specifically fixed.
I think any self driving tech at the moment is just terrible and shouldn’t be used, for the record
I think Elon found a glitch it tech-bro reasoning: you promise a product, people believe you and buy it, then they realise that it is not what was advertised, but it is OK, because new version has it now, so you can't complain, it is your fault in the eyes of your peers, you should have made a better research or whatever.
Your FSD has almost got you killed? Don't worry, this is your fault, and anyway the bug was already fixed in the update. Probably. This time for sure.
The changes in the new 2024 version, specifically changed the suspension setup and you can mechanically verify the difference between the models because there were mechanical changes.
You need to drop the FSD rhetoric, there are an enormous amount of Tesla drivers who couldn't care about FSD, but simply enjoying the car as an EV.
From the consumer POV who doesn't own shares in the brand, it makes no sense at all.
> enormous amount of Tesla drivers who couldn't care about FSD but simply enjoying the car as an EV
Are there an enormous amount of disappointed Tesla drivers as well who stuck with it because of a sunk cost fallacy?
Should, for example, CyberTruck owners who spend 150K on a bullet-proof car be happy that in the new 2026 version the panels finally will stop falling off?
Should the few million people who purchased Tesla Model 3 feel better that the new version finally drives like a car?
Either way, in prior comments you couldn't come to terms with them possibly fixing something like that in the first place.
Tesla is a low-trust company. Too much grifting
Is this the new "how would you feel if you didn't eat breakfast yesterday?"?
Successive version always do slight changes like that over time, that how car companies work. Old owners don't need to be happy about it, or think about it at all.
Combine that with FSD rubbish bait and switch, nutjob CEO, and the cratering of any brand goodwill ... I won't touch those cars if you paid me (seriously).
And honestly, I recognise the innate progression of the technology ... but others have caught and surpassed to the point that I don't have to worry about buying the lesser car anymore when I go elsewhere.
On the whole seems a little... over enthusiastic...
https://news.ycombinator.com/newsguidelines.html
Please just make an effort to use HN as it's intended. It's only a place where people want to participate because so many people put effort into keeping it that way.
It’s easy to pick the worst model of the lot and use it to disregard the entire brand, but I can’t expect any more critical thought from some online
Thank you for contributing to lower the overall amount of critical thinking on the internet.
whilst surely top ranked, they apparently share the top with others makers (https://www.euroncap.com/en/ratings-rewards/latest-safety-ra...) (Large Family Cars category, last two years, order by occupant protection).
https://www.carscoops.com/2024/11/tesla-model-3-comes-bottom...
The Cybertruck is a complete disaster of a vehicle with so many issues (eg [1]) that the only reason people buy them is to make a political statement from a group that 3+ years ago wouldn't have been caught buying an EV.
Teslas are drivable iPads. Many people (myself included) not only hate this (because it's hard to use without looking) but it's also lazy design. By this I mean, it allows manufacturers to say "we'll fix it with a software update" (and then probably never get around to it) whereas haptic controls require more thought and effort to be put into the UI/UX during manufacturing.
For other Teslas, there have been a host of other issues, some small, some not. For example, the seats were unreliable if adjusted too often so Tesla made an OTA update to limit how much you can adjust the seats to avoid failure [2].
The only thing propping up Tesla sales now are trade restrictions on BYD.
[1]: https://apnews.com/article/cybertruck-recall-tesla-elon-musk...
[2]: https://driveteslacanada.ca/news/tesla-now-monitors-how-ofte...
But also it’s not hyperbole, though I agree it sounds like it.
Association comes in different strengths and forms. But currently owned and operated by a guy actively destroying my country is much more direct and urgent than most of these associations.
But the larger point is that it seems correct that the only people willing to buy a Tesla anymore have conservative politics.
So as long as you think China's production methods are morally fine, and don't mind them being in control of your car's software, pick BYD
I keep looking for the next EV to buy but every one that I test drive seems lacking, mostly in the software area. Tesla has the software nailed and nobody else can seem to figure it out. I recently drove a Rivian R1S and really liked it, but the software was not great (for a $100K car) and it really disappointing me.
I really feel like Tesla does great work but their leader needs to go. He's smeared the name so badly that I fear it might never recover.
I don’t see how Tesla could even dissociate with him, he owns the board and would have to divest also. I think the only path back is if people forget, conservatives suddenly like EVs, or they strike gold with the robots or something.
I’m not sure why you’re so easy to write off 1000s of deaths, but 300,000 so far is one model, but even if it’s very wrong it’s going to be hundreds of thousands sometime this year in all likelihood. https://www.washingtonpost.com/politics/2025/05/28/rubio-aid...
Teslas don’t even have a Speedo, let alone a HUD. It’s just an iPad with proprietary software which can’t sync with your phone. Terrible
Bidirectional charging would be great, everything else is already far up there.
For me this is a clear win for Tesla. Anything under 5s is crazy fast to me anyway, so the other things I mention are worth a lot more.
Subjectively, the boot of the Ioniq 5 on the normal models is huge, and I say that as someone for whom boot size was the primary requirement. Moving the seats forward creates epic amounts of space at the cost of reduced leg room for rear passengers.
And for when I want to carry weirder stuff I fold the backseats. I've carried entire Ikea wardrobes like this.
And the point remains, Tesla learns harder into the deception and lying.
Companies used to play these tricks in the 90s. I am NOT happy that Tesla is trying to bring them back.
Other times, I have something that should fit, but can't readily fit through the oddly shaped rear hatch and it becomes a problem.
I'd agree with you with regard to luggage. That's a scenario that really hits the strength of the Y setup, especially with the relatively large frunk. I often end up with a few backpack sized items up there on longer trips.
That’s great but you’re missing the fact that the Ioniq has a HUD. What does the Tesla have? Everything you’ve mentioned is irrelevant. The Ioniq has CarPlay so you don’t touch the software.
Do you mean I could get something significantly better if I choose not to get the things which are important to me? Because that would be worse for me, not better. I have never cared about the badge.
Out of interest, if I reduced the trunk space a little, which similarly priced EVs could you recommend with more range and better performance than the Model Y AWD Long Range?
My Model 3 Performance is rated for I think 300 miles. In the real world, it can be as low as 230 miles if it's 30F outside and I'm going 75 mph. If it's warm, I'm still looking at ~270 miles at 75 mph. The colder it gets, the lower your range because cold air is denser (increasing drag) and higher usage of the heater drains the battery.
And the chinese models are just much better.