TL;DR: Self-Driving Teslas Rear-End Motorcyclists, Killing at Least 5
Brevity is the spirit of wit, and I am just not that witty. This is a long article, here is the gist of it:
- The NHTSA’s self-driving crash data reveals that Tesla’s self-driving technology is, by far, the most dangerous for motorcyclists, with five fatal crashes that we know of.
- This issue is unique to Tesla. Other self-driving manufacturers have logged zero motorcycle fatalities with the NHTSA in the same time frame.
- The crashes are overwhelmingly Teslas rear-ending motorcyclists.
Read our full analysis as we go case-by-case and connect the heavily redacted government data to news reports and police documents.
Oh, and read our thoughts about what this means for the robotaxi launch that is slated for Austin in less than 60 days.
I imagine bicyclists must be æffected as well if they’re on the road (as we should be, technically). As somebody who has already been literally inches away from being rear-ended, this makes me never want to bike in the US again.
Time to go to Netherlands.
*affected
Thank you for your service.
Affectively, does it realy mater if someone has slite misstakes in there righting?
I think i had a stroke reading that. Take your upvote and get out!
I’m not going to lie, I almost had a stroke writing it…
I upvoted every comment in this sub-thread shitshow and hated all of it.
Yer welcome!
this makes me never want to bike in the US again.
I live close enough to work for it to be a very reasonable biking distance. But there is no safe route. A high-speed “stroad” with a narrow little bike lane. It would only be a matter of time before some asshole with their face in their phone drifts into me.
I am deeply resentful of our automobile-centric infrastructure in the U.S. It’s bad for the environment, bad for our wallets, bad for our waistlines, and bad for physical safety.
human driving cars still target bicyclists on purpose so i don’t know see how teslas could be any worse…
p.s. painting a couple lines on the side of the road does not make a safe bike lane… they need a physical barrier separating the road from them… like how curbs separate the road from sidewalks…
I mean yeah, I just said above that someone almost killed me. They were probably a human driver. But that’s a “might happen, never know.” If self driving cars are rear-ending people, that’s an inherent artifact of it’s programming, even though it’s not intentionally programmed to do that.
So it’s like, things were already bad. I already do not feel safe doing any biking anymore. But as self driving cars become more prevalent, that threat upgrades to a kind of defacto, “Oh, these vast stretches of land are places where only cars and trucks are allowed. Everything else is roadkill waiting to happen.”
Hey guys relax! It’s all part of the learning experience of Tesla FSD.
Some of you may die, but that’s a sacrifice I’m willing to make.Regards
Elon Musk
CEO of TeslaP.S. Volunteers needed for the Mars mission as well.
Is musk going, because I vote to be on whatever planet he isn’t.
If there are only fElon fan boys going I’ll take the hit and go to open the airlock halfway through.
News on the first mission: Meteoroid crashes into full flying SpaceX rocket, killing all aboard.
deleted by creator
+1 for you. However, replace “Regards” with the more appropriate words from the German language. The first with an S, and the second an H. I will not type that shit, fuck Leon and I hope the fucking Nazi owned Tesla factory outside of Berlin closes.
Yes I’m not writing that shit, even in a sarcastic post. Bu I get your drift.
On the other hand, since you are from Germany, VW group is absolutely killing it on EV recently IMO.
They totally dominate top 10 EV here in Denmark, with 7 out of 10 top selling models!!
They are competitively priced, and they are the best combination of quality and range in their price ranges.
This is news? Fortnine talked about it two years ago.
TL;DR Tesla removed LIDAR to save a buck and the cameras see two red dots that the 'puter thinks it’s a far away car at night when indeed it’s a close motorcycle.It can’t even perceive the depth of the lights?
Not with cameras alone, no.
Why not? It’s got multiple cameras so could judge distances the same way humans do.
However there have been both hardware and software updates since most of those, so the critical question is how much of a problem is it still? The article had no info or speculation on that
It’s helpful to remember that not everyone has seen the same stories you have. If we want something to change, like regulators not allowing dangerous products, then raising public awareness is important. Expressing surprise that not everyone knows about something can be counterproductive.
Going beyond that, wouldn’t the new information here be the statistics?
My state allowed motorcycle filtering in 2019 (not the same as California’s lane splitting). They ran a study and found a ton of motorcyclists were being severely injured or killed while getting rear ended sitting at stop lights. Filtering allows them to move to the front of the traffic light while the light is red and traffic is stationary. Many people are super aggravated about it even though most of the world has been doing it basically forever.
like regulators not allowing dangerous products,
I include human drivers in the list of dangerous products I don’t want allowed. The question is self driving safer overall (despite possible regressions like this). I don’t want regulators to pick favorites. I want them to find “the truth”
Sure, we’re in agreement as far as that goes. My point was just the commenter above me was indicating it should be common knowledge that Tesla self driving hits motorcycles more than other self driving cars. And whether their comment was about this or some other subject, I think it’s counterproductive to be like “everyone knows that.”
It could be two motorcycles side by side.
The argument is that humans can drive with just 2 eyes, so cameras are enough. I disagree with this position, given that the limitations of a camera-only system. But that’s what it is.
Different sensors excel at different tasks and different conditions, and cameras are not always it.
I had ignored the video, as I didn’t expect Mark to expose Tesla
For what it’s worth, it really isn’t clear if this is FSD or AP based on the constant mention of self driving even when it’s older collisions when it would definitely been AP, and is even listed as AP if you click on the links to the crash.
So these may all be AP, or one or two might be FSD, it’s unclear.
Every Tesla has AP as well, so the likelihood of that being the case is higher.
In this case, does it matter? Both are supposed to follow a vehicle at a safe distance
I’d be more interested in how it changes over time, as new software is pushed. While it’s important that know it had problems judging distance to a motorcycle, it’s more important to know whether it still does
In this case, does it matter? Both are supposed to follow a vehicle at a safe distance
I think it does matter, while both are supposed to follow at safe distances, the FSD stack is doing it in a completely different way. They haven’t really been making any major updates to AP for many years now, all focus has been on FSD. I think the only real changes it’s had for quite awhile have been around making sure people are paying attention better.
AP is looking at the world frame by frame, each individual camera on it’s own, while FSD is taking the input of all cameras, turning into 3d vector space, and then driving based off that. Doing that on city streets and highways is only a pretty recent development. Updates for doing it this way on highway and streets only went out to all cars with FSD in the past few months. For a long time it was on city streets only.
I’d be more interested in how it changes over time, as new software is pushed.
I think that’s why it’s important to make a real distinction between AP and FSD today (and specifically which FSD versions)
They’re wholly different systems, one that gets older every day, and one that keeps getting better every few months. Making an article like this that groups them together over the span of years muddies the water on what / if any progress has been made.
Fair enough!
At least one of the fatalities is Full-Self Driving (it was cited by name in the police reports). The remainder are Autopilot. So, both systems kill motorcyclists. Tesla requests this data redacted from their NHTSA reporting, which specifically makes it difficult for consumers to measure which system is safer or if incremental safety improvements are actually being made.
You’re placing a lot if faith that the incremental updates are improvements without equivalent regressions. That data is specifically being concealed from you, and I think you should probably ask why. If there was good news behind those redactions, they wouldn’t be redactions.
I didn’t publish the software version data point because I agree with AA5B, it doesn’t matter. I honestly don’t care how it works. I care that it works well enough to safely cohabit the road with my manual transmission cromagnon self.
I’m not a “Tesla reporter,” I’m not trying to cover the incremental changes in their software versions. Plenty of Tesla fans doing that already. It only has my attention at all because it’s killing vulnerable road users, and for that analysis we don’t actually need to know which self-driving system version is killing people, just the make of car it is installed on.
I’d say it’s a pretty important distinction to know if one or both systems have a problem and the level of how bad that problem is.
Also are you referencing the one in Seattle in 2024 for FSD? The CNBC article says FSD, but the driver said AP.
And especially back then, there’s also an important distinction of how they work.
FSD on highways wasn’t released until November 2024, and even then not everyone got it right away. So even if FSD was enabled, the crash may have been under AP.Edit: Also if it was FSD for real
(that 2024 crash would have had to happen on city streets, not a highway)then thats 1 motorcycle fatality in 3.6 billion miles. The other 4 happened over 10 billion miles. Is that not an improvement? (edit again: I should say we can’t tell it’s an improvement yet as we’d have to pass 5 billion, so the jury is still out I guess IF that crash was really on FSD)Edit: I will cede though that as a motorcyclist, you can’t know what the Tesla is using, so you’d have to assume the worst.
Edit: Just correcting myself that I was wrong about FSD in 2024. The change over to neural nets happened in November, but FSD was still FSD on highways when this accident happened. It was even earlier than that when FSD became AP when you transitioned to higways
Police report for 2024 case attached, it is also linked in the original article: https://www.opb.org/article/2025/01/15/tesla-may-face-less-accountability-for-crashes-under-trump/
It was Full Self Driving, according to the police. They know because they downloaded the data off the vehicle’s computer. The motorcyclist was killed on a freeway merge ramp.
All the rest is beyond my brief. Thought you might like the data to chew on, though.
The motorcyclist was killed on a freeway merge ramp.
I’d say that means it’s a very good chance that yes, while FSD was enabled, the crash happened under the older AP mode of driving, as it wasn’t until November 2024 that it was moved over to the new FSD neural net driving code.. I was wrong here, it actually was FSD then, it just wasn’t end to end neural nets then like it is now.Also yikes… the report says the AEB kicked in, and the driver overrode it by pressing on the accelerator!
No shit on that yikes. That blew my fucking mind.
Half the time when your AEB activates, you are unconscious or dazed and you’re just flailing around your cabin like a rag doll, because you’ve crashed. If your foot happens to flail into the accelerator, get ready for a very exciting (if short-lived) application of that impressive 0 to 60 time.
Okay, so I’m going to edit my earlier replies but replying again so you see, as I was wrong.
Version 11/12 in 2023/2024 wasn’t using the AP code, it just wasn’t using the neural nets. So it was legitimately FSD, but it was running different code on the freeways (non neural net) vs on city streets (neural net)
But it was indeed FSD. Version 11.x was the change where it stopped using AP when you left city streets.
That’s not good though, right? “We have the technology to save lives, it works on all of our cars, and we have the ability to push it to every car in the fleet. But these people haven’t paid extra for it, so…”
Well, only 1 or 2 of those were in a time frame where I’d consider FSD superior to AP, it’s a more recent development where that’s likely the case.
But to your point, at some point I expect Tesla to use the FSD software for AP for the exact reasons you mentioned. My guess is they’d just do something like disable making left/right turns , so you wouldn’t be able to use it outside of straight stretches like AP today.
Tesla self driving is never going to work well enough without sensors - cameras are not enough. It’s fundamentally dangerous and should not be driving unsupervised (or maybe at all).
Accurate.
Each fatality I found where a Tesla kills a motorcyclist is a cascade of 3 failures.
- The car’s cameras don’t detect the biker, or it just doesn’t stop for some reason.
- The driver isn’t paying attention to detect the system failure.
- The Tesla’s driver alertness tech fails to detect that the driver isn’t paying attention.
Taking out the driver will make this already-unacceptably-lethal system even more lethal.
- Self-driving turns itself off seconds before a crash, giving the driver an impossibly short timespan to rectify the situation.
… Also accurate.
God, it really is a nut punch. The system detects the crash is imminent.
Rather than automatically try to evade… the self-driving tech turns off. I assume it is to reduce liability or make the stats look better. God.
Yep, that one was purely about hitting a certain KPI of ‘miles driven on autopilot without incident’. If it turns off before the accident, technically the driver was in control and to blame, so it won’t show up in the stats and probably also won’t be investigated by the NTSB.
so it won’t show up in the stats
Hopefully they wised up by now and record these stats properly…?
NHTSA collects data if self-driving tech was active within 30 seconds of the impact.
The companies themselves do all sorts of wildcat shit with their numbers. Tesla’s claimed safety factor right now is 8x human. So to drive with FSD is 8x safer than your average human driver, that’s what they say on their stock earnings calls. Of course, that’s not true, not based on any data I’ve seen, they haven’t published data that makes it externally verifiable (unlike Waymo, who has excellent academic articles and insurance papers written about their 12x safer than human system).
So to drive with FSD is 8x safer than your average human driver.
WITH a supervising human.
Once it reaches a certain quality, it should be safer if a human is properly supervising it, because if the car tries to do something really stupid, the human takes over. The vast vast vast majority of crashes are from inattentive drivers, which is obviously a problem and they need to keep improving the attentiveness monitoring, but it should be safer than a human with human supervision because it can also detect things the human will ultimately miss.
Now, if you take the human entirely out of the equation, I very much doubt that FSD is safer than a human at it’s current state.
Fascinating! I don’t know all this. Thanks
If they ever fixed it, I’m sure Musk fired whomever is keeping score now. He’s going to launch the robotaxi stuff soon and it’s going to kill a bunch of people.
Even when it is just milliseconds before the crash, the computer turns itself off.
Later, Tesla brags that the autopilot was not in use during this ( terribly, overwhelmingly) unfortunate accident.
deleted by creator
There’s at least two steps before those three:
-1. Society has been built around the needs of the auto industry, locking people into car dependency
- A legal system exists in which the people who build, sell and drive cars are not meaningfully liable when the car hurts somebody
- A legal system exists in which the people who build, sell and drive cars are not meaningfully liable when the car hurts somebody
That’s a good thing, because the alternative would be flipping the notion of property rights on its head. Making the owner not responsible for his property would be used to justify stripping him of his right to modify it.
You’re absolutely right about point -1 though.
build, sell and drive
You two don’t seem to strongly disagree. The driver is liable but should then sue the builder/seller for “self driving” fraud.
Maybe, if that two-step determination of liability is really what the parent commenter had in mind.
I’m not so sure he’d agree with my proposed way of resolving the dispute over liability, which would be to legally require that all self-driving systems (and software running on the car in general) be forced to be Free Software and put it squarely and completely within the control of the vehicle owner.
I would assume everyone here would agree with that 😘
I mean, maybe, but previously when I’ve said that it’s typically gone over like a lead balloon. Even in tech forums, a lot of people have drunk the kool-aid that it’s somehow suddenly too dangerous to allow owners to control their property just because software is involved.
Most frustrating thing is, as far as I can tell, Tesla doesn’t even have binocular vision, which makes all the claims about humans being able to drive with vision only even more blatantly stupid. At least humans have depth perception. And supposedly their goal is to outperform humans?
Tesla’s argument of “well human eyes are like cameras therefore we shouldn’t use LiDAR” is so fucking dumb.
Human eyes have good depth perception and absolutely exceptional dynamic range and focusing ability. They also happen to be linked up to a rapid and highly efficient super computer far outclassing anything that humanity has ever devised, certainly more so than any computer added to a car.
And even with all those advantages humans have, we still crash from time to time and make smaller mistakes regularly.
They also happen to be linked up to a rapid and highly efficient super computer far outclassing anything that humanity has ever devised
A neural network that has been in development for 650 million years.
Ok, maybe project managers are good for something.
Anyone who has driven (or walked) into a sunrise/sunset knows that human vision is not very good. I’ve also driven in blizzards, heavy rain, and fog - all times when human vision is terrible. I’ve also not seen green lights (I’m colorblind).
Human vision is very, very, very good. If you think a camera installed to a car is even close to human eyesight, then you are extremely mistaken.
Human eyes are so far beyond it’s hard to even quantify.
And bullshit on you not being able to see the lights. They’re specifically designed so that’s not an issue for colourblind people.
And bullshit on you not being able to see the lights. They’re specifically designed so that’s not an issue for colour blind people
Some lights are, but not all of them are. I often say I go when the light turns blue. However not all lights have that blue tint and so I often cannot tell the difference between a white light and a green light by color. (but white is not used in a stoplight and I can see red/yellow just fine) Where I live all stoplights have green on the bottom so that is always a cheat I use, but that only works if I can see the relative position - in an otherwise dark situation I only see a light in front of me and not the rest of the structure and so I cannot tell. I have driven where stoplights are not green on bottom and I can never remember if green is left/right.
Even when the try though, not all colorblind is the same. There may not be a mitigation that will work from two different people with different aspects of colorblind.
Human vision is very, very, very good. If you think a camera installed to a car is even close to human eyesight, then you are extremely mistaken.
Why are you trying to limit cars to just vision? That is all I have as a human. However robots have radar, lidar, radio, and other options, there is no reasons they can’t use them and get information eyes cannot. Every option has limits.
Please read my comments before you respond to them.
Bro I’m colorblind too and if you’re not sure what color the light is, you have to stop. Don’t put that on the rest of us.
I can see red clearly and so not sure means I can go.
I’ve only noticed issues in a few situations. When I’m driving at night and suddenly the weirdly aimed streetlight turns yellow - until it changed I didn’t even know there was a stoplight there. The second was I was making a left turn at sunset (sun behind me) and the green arrow came on but the red light remained on so I couldn’t see it was time/safe to go until my wife alerted me.
These fatalities are a Tesla business advantage. Every one is a data point they can use to program their self-driving intelligence. No one has killed as many as Tesla, so no one knows more about what kills people than Tesla. We don’t have to turn this into a bad thing just because they’re killing people /s
they originally had lidar, or radar, but musk had them disabled in the older models.
They had radar. Tesla has never had lidar, but they do use lidar on test vehicles to ground truth their camera depth / velocity calculations.
Lidar needs to be a mandated requirement for these systems.
Or at least something other than just cameras. Even just adding ultrasonic senses to the front would be an improvement.
The range on ultrasonics is too short. They only ever get used for parking type situations, not driving on the roadways.
Honestly, emergency braking with LIDAR is mature and cheap enough at this point that is should be mandated for all new cars.
No, emergency braking with radar is mature and cheap. Lidar is very expensive and relatively nascent
How about we disallow it completely, until it’s proven to be SAFER than a human driver. Because, why even allow it if it’s only as safe?
As an engineer, I strongly agree with requirements based on empirical results rather than requiring a specific technology. The latter never ages well. Thank you.
It’s hardly either / or though. What we have here is empirical data showing that cars without lidar perform worse. So it’s based in empirical results to mandate lidar. You can build a clear, robust requirement around a tech spec. You cannot build a clear, robust law around fatality statistics targets.
We frequently build clear, robust laws around mandatory testing. Like that recent YouTube video where the Tesla crashed through a wall, but with crash test dummies.
Those are ways to gather empirical results, though they rely on artificial, staged situations.
I think it’s fine to have both. Seat belts save lives. I see no problem mandating them. That kind of thing can still be well founded in data.
You mean like this Euro NCAP testing, where Tesla does stop and most others don’t including some vehicles with lidar?
This sounds good until you realize how unsafe human drivers are. People won’t accept a self-driving system that’s only 50% safer than humans, because that will still be a self-driving car that kills 20,000 Americans a year. Look at the outrage right here, and we’re nowhere near those numbers. I also don’t see anyone comparing these numbers to human drivers on any per-mile basis. Waymos compared favorably to human drivers in their most recently released data. Does anyone even know where Teslas stand compared to human drivers?
There’s been 54 reported fatalities involving their software over the years in the US.
That’s around 10 billion AP miles (9 billion at end of 2024), and around 3.6 billion on the various version of FSD (beta / supervised). Most of the fatal accidents happened on AP though not FSD.
Lets just double those fatal accidents to 108 to make it for the world, but that probably skews high. Most of the fatal stuff I’ve seen is always in the US.
That equates to 1 fatal accident every 125.9 million miles.
The USA average per 100 million miles is 1.33 deaths, so even doubling the deaths it’s less than the current national average. That’s the equivalent of 1.33 deaths every 167 million miles with Tesla’s software.
Edit: I couldn’t math, fixed it. Also for FSD specifically, very few places have it. Mainly North America, and just recently, China. I wish we had fatalities for FSD specifically.
Stop dehumanizing drivers who killed people.
Feature, wrongly called, Full Self-Driving, shall be supervised at any time.If you’re going to say your car has “full self driving”, it should have that, not “full self driving (but needs monitoring.)” or “full self driving (but it disconnects 2 seconds before impact.)”.
I think it’s important to call out inattentive drivers while also calling out the systems and false advertising that may lead them to become less attentive.
If these systems were marketed as “driver assistance systems” instead of “full self driving”, certainly more people would pay attention. The fact that they’ve been allowed to get away with this blatant false advertising is astonishing.
They’re also obviously not adequately monitoring for driver attentiveness.
Sounds like NHTSA needs a visit from DOGE!
Gotta get rid of the evidence.
Why is self-driving even allowed?
Because muh freedum, EU are a bunch of commies for not allowing this awesome innovation on their roads
(I fucking love living in the EU)
Robots don’t get drunk, or distracted, or text, or speed…
Anecdotally, I think the Waymos are more courteous than human drivers. Though waymo seems to be the best ones out so far, idk about the other services.
Don’t waymos have remote drivers that take control in unexpected situationsml?
They have remote drivers that CAN take control in very corner case situations that the software can’t handle. The vast majority of driving is don’t without humans in the loop.
They don’t even do that, according to Waymo’s claims.
They can suggest what the car should do, but they aren’t actually doing it. The car is in complete control.
Its a nuanced difference, but it is a difference. A Waymo employee never takes control of or operates the vehicle.
Interesting! I did not know that - I assumed the teleoperators took direct control, but that makes much more sense for latency reasons (among others)
I always just assumed it was their way to ensure the vehicle was really autonomous. If you have someone remotely driving it, you could argue it isn’t actually an AV. Your latency idea makes a lot of sense as well though. Imagine taking over and causing an accident due to latency? This way even if the operator gives a bad suggestion, it was the car that ultimately did it.
So they say
Humans are terrible drivers. The open question is are self driving cars overall safer than human driven cars. So far the only people talking either don’t have data, or have reason cherry pick only parts of the data that make self driving look good. This is the one exception where someone seemingly independent has done analysis - the question is are they unbiased, or are they cherry picking data to make self driving look bad (I’m not familiar with the source so I can’t answer that)
Either way more study is needed.
I am absolutely biased. It’s me, I’m the source :)
I’m a motorcyclist, and I don’t want to die. Also just generally, motorcyclists deserve to get where they are going safely.
I agree with you. Self-driving cars will overall greatly improve highway safety.
I disagree with you when you suggest that pointing out flaws in the technology is evidence of bias, or “cherry picking to make self driving look bad.” I think we can improve on the technology by pointing out its systemic defects. If it hits motorcyclists, take it off the road, fix it, and then save lives by putting it back on the road.
That’s the intention of the coverage, at least: I am hoping to apply pressure to improve rather than remove. Read my Waymo coverage, I’m actually a big automation enthusiast, because fewer crashes is a good thing.
I wasn’t trying to suggest that you are biased, only that I have no clue and so it is possible you are somehow unfairly doing something.
Perfectly fair. Sorry, I jumped the gun! Good on you for being incredulous and inspecting the piece for manipulation, that’s smart.
Humans are terrible. The human eyes and brain are good at detecting certain things though that allow a reaction where computer vision, especially only using one method of detection, fails often. There are times when an automated system will prevent a problem before a human could even see it. So far neither is the clear winner, human driving just has a legacy that automation has to beat by a great length and not just be good enough.
On the topic of human drivers, I think most on the road drive reactively and not based on prediction and anticipation. Given the speed and possible detection methods, a well designed automated system should be excelling at this. It costs more and it more complex to design such a thing, so we’re getting the bare bones of the best minimum tech can give us right now, which again is not a replacement for all cases.
Because the only thing worse than self driving is human driving.
Because the march of technological advancement is inevitable?
In light of recent (and let’s face it, long ago cases) Tesla’s “Full Self Driving” needs to be downgraded to level 2 at best.
Level 2: Partial Automation
The vehicle can handle both steering and acceleration/deceleration, but the driver must remain engaged and ready to take control.
Pretty much the same level as other brands self driving feature.
The other brands, such as Audi and VW, work much better than Tesla’s system. Their LIDAR systems aren’t blinded by fog, and rain the way the Tesla is. Someone recently tested an Audi with its system against a Tesla with its system. The Tesla failed either 3/5 or 4/5 tests. The Audi passed 3/5 or 4/5. Neither system is perfect, but the one that doesn’t rely on just cameras is clearly superior.
Edit: it was Mark Rober.
It’s hard to tell, but from about 15 minutes of searching, I was unable to locate any consumer vehicles that include a LIDAR system. Lots of cars include RADAR, for object detection, even multiple RADAR systems for parking. There may be some which includes a TimeOfFlight sensor, which is like LIDAR, but static and lacks the resolution/fidelity. My Mach-E which has level 2 automation uses a combination of computer vision, RADAR and GPS. I was unable to locate a LIDAR sensor for the vehicle.
The LIDAR system in Mark’s video is quite clearly a pre-production device that is not affiliated with the vehicle manufacturer it was being tested on.
Adding, after more searching, it looks like the polestar 3, some trim levels of the Audi A8 and the Volvo EX90 include a LiDAR sensor. Curious to see how the consumer grade tech works out in real world.
Please do not mistake this comment as “AI/computer vision” evangelisim. I currently have a car that uses those technologies for automation, and I would not and do not trust my life or anyone else’s to that system.
The way I understand it, is that Audi, Volvo, and VW have had the hardware in place for a few years. They are collecting real world data about how we drive before they allow the systems to be used at all. There are also legal issues with liability.
Mercedes uses LiDAR. They also operate the sole Level 3 driver automation system in the USA. Two models only, the new S-Class and EQS sedans.
Tesla alleges they’ll be Level 4+ in Austin in 60 days, and just skip Level 3 altogether. We’ll see.
Yeah, keep in mind that Elon couldn’t get level 3 working in a closed, pre-mapped circuit. The robotaxis were just remotely operated.
Bribes to local governments and police, mostly.
Five years ago, you could not have brought this up without Musk simps defending it.
There seems to be people/bots down-voting critical takes up and down this very thread. What chumps.
the cybertruck is sharp enough to cut a deer in half, surely a biker is just as vulnerable.
I wonder if it’s happened yet
There was an article where he sliced a deer in half
On a quick read, I didn’t see the struck motorcycles listed. Last I heard, a few years ago, was that this mainly affected motorcycles with two rear lights that are spaced apart and fairly low to the ground. I believe this is mostly true for Harleys.
The theory I recall was that this rear light configuration made the Tesla assume it was looking (remember, only cameras without depth data) at a car that was further down the road - and acceleration was safe as a result. It miscategorised the motorcycle so badly that it misjudged it’s position entirely.
The ridiculous thing is, it has 3 cameras pointing forward, you only need 2 to get stereoscopic depth perception with cameras…why the fuck are they not using that!?
Edit: I mean, I know why, it’s because it’s cameras with three different lenses used for different things (normal, wide angle, and telescopic) so they’re not suitable for it, but it just seems stupid to not utilise that concept when you insist on a camera only solution.
That seems like a spectacular oversight. How is it supposed to replicate human vision without depth perception?
The video 0x0 linked to in another comment describes the likely method used to infer distance to objects without a stereoscopic setup, and why it (likely) had issues determining distance in the cases where they hit motorcycles.
Little known fact: the Model S (P) actually stands for Polyphemus Edition, not Plaid Edition.
I also saw that theory! That’s in the first link in the article.
The only problem with the theory: Many of the crashes are in broad daylight. No lights on at all.
I didn’t include the motorcycle make and model, but I did find it. Because I do journalism, and sometimes I even do good journalism!
The models I found are: Kawasaki Vulcan (a cruiser bike, just like the Harleys you describe), Yamaha YZF-R6 (a racing-style sport bike with high-mount lights), and a Yamaha V-Star (a “standard” bike, fairly low lights, and generally a low-slung bike). Weirdly, the bike models run the full gamut of the different motorcycles people ride on highways, every type is represented (sadly) in the fatalities.
I think you’re onto something with the faulty depth sensors. Sensing distance is difficult with optical sensors. That’s why Tesla would be alone in the motorcycle fatality bracket, and that’s why it would always be rear-end crashes by the Tesla.
At least in EU, you can’t turn off motorcycle lights. They’re always on. In eu since 2003, and in US, according to the internet, since the 70s.
I assume older motorcycles built before 2003 are still legal in the EU today, and that the drivers’ are responsible for turning on the lights when riding those.
Point taken: Feel free to amend my comment from “No lights at all” to “No lights visible at all.”
Because I do journalism, and sometimes I even do good journalism!
In that case, you wouldn’t happen to know whether or not Teslas are unusually dangerous to bicycles too, would you?
Surprisingly, there is a data bucket for accidents with bicyclists, but hardly any bicycle crashes are reported.
That either means that they are not occurring (woohoo!), or that means they are being lumped in as one of the multiple pedestrian buckets (not woohoo!), or they are in the absolutely fucking vast collection of “severity: unknown” accidents where we have no details and Tesla requested redaction to make finding the details very difficult.
Thanks!
Any time :)
Whatever it is, it’s unacceptable and they should really ban Tesla’s implementation until they fix some fundamental issues.
Still probably a good idea to keep an eye on that Tesla behind you. Or just let them past.
This video proposes that theory.
Ah, thanks for jogging my memory
Are you saying Harley drivers are fair game?
Trucks in general have gotten so big they are pedestrian deathtraps
as daily rider, i must add having a tesla behind to the list of road hazards to look out
I’m on mine far more often than I’m in a car. I think Tesla found out that I point and laugh at any cyber trucks I see at red lights while I’m out and is trying to kill me.
I feel like that as a driver. Tesla’s do not move at a consistent speed, which drives me mad
You’re not wrong, but good luck watching out for a vehicle approaching you at a 30 mph differential (which is what I recall from fortnine covering the topic years ago) from behind.
