Tesla’s Autopilot involved in far more crashes than previously known
seattletimes.com
seattletimes.com
Somebody bothered to make website to track Tesla deaths and Tesla Autopilot related deaths.
Current count is 33 for Autopilot related deaths.
Last time I cross referenced incidents it was accurate, but that was a while ago.
Case in point, Case 312.1 is a pickup truck that ran a red light at an intersection and hit the Tesla, killing its driver, according to the archived new report. The stats on that page shows it as death in "TSLA+cycl / peds" column, even though there were no vulnerable road users involved.
Any reasonable individual would see that this has zero fault on the Tesla, but since it involves a Tesla and a death, it's included.
Involved in an accident does not mean the Tesla is at fault.
> Tesla Deaths is a record of Tesla accidents that involved a driver, occupant, cyclist, motorcyclist, or pedestrian death, whether or not the Tesla or its driver were at fault
The FAQ contains the reasoning behind that, which is very simple:
> Q: Why do you include deaths where the Tesla was not at fault?
> NHTSA death statistics cited by Elon Musk do not differentiate between at-fault and not-at-fault. All manufacturers are treated the same way. For the purposes of calculating Tesla fatalities in accordance with NHTSA methodology, we also do not include the many deaths in other vehicles, despite Teslas being at fault in the majority of those cases.
Note that since this is only based on press reports, this is a lower bound on the number of deaths. Probably by a signinficant margin.
Also, back in the day there was a little barrier that would flip out hinged on the passenger side that would attempt to prevent kids from walking out in front of the bus, at least so close that the bus driver couldn't reasonably see. I don't know if those are still common or not.
Image of a typical North American school bus: https://en.m.wikipedia.org/wiki/School_bus_traffic_stop_laws...
The fact that the car, despite having 8 cameras, all around the vehicle, still failed to notice the motorcyclist is a damning indictment of the poor quality of Tesla's vision software.
My Subaru has absolutely no problem detecting cars, cyclists, or pedestrians around or behind my vehicle, and Subaru doesn't even have a best-in-class vision system.
I'm not trying to bash Cruise. My point is that shit happens, we should look at the complete statistics, not one offs.
[1] https://www.kron4.com/news/bay-area/cruise-car-appears-to-re...
Okay. Who out there has better vision software that can do as much as Tesla?
Tesla is also far less cautious in its marketing. Most reasonable complaints about Tesla's self-driving kit aren't about the kit per se, but the false expectations that encourage drivers to confidently send it where it shouldn't go.
But I wonder how often the equivalent fails for other feature equivalent systems since they are becoming very common now, surely some of them are failing. My system warns me each time not to rely on it to avoid crashes, for example. I get that Tesla gets most of the attention, especially since they also offer FSD as well, but surely there are failures from other car manufacturers that are simply not getting the same microscopic attention.
Wow, just wow. People, don't do stupid stuff like this with security features.
I can 100% understand that people would want a workaround. It almost makes it not worth it. Some sort of capacitive detection of hands would make it a much better experience.
>The car – allegedly in Autopilot mode – never slowed down.
>It struck Mitchell at 45 mph. The teenager was thrown into the windshield, flew into the air and landed face down in the road, according to his great-aunt, Dorothy Lynch. Mitchell’s father heard the crash and rushed from his porch to find his son lying in the middle of the road.
>“If it had been a smaller child,” Lynch said, “the child would be dead.”
Elon should really stop the yearly Level 5 promises. Also, that driver should be charged with negligence.
AutoPilot has little to do with FSD though. I’m assuming these are accidents related to autopilot rather than FSD, or the article would have been titled differently?
It's surprising this isn't the first (and maybe only) salient point. Autopilot is a slightly more capable cruise control. If a driver sets a car to drive 45 mph and then stops paying attention, who's at fault when the car hits something? The car was doing exactly what it was supposed to do.
(As pointed out from another commenter, the article is about Autopilot, not FSD.)
I have to wonder what else the driver was doing. Perhaps sleeping or playing a game on their phone. I am glad they threw the book at this person and I am a Tesla fan.
I don't own a Tesla yet but I will shortly. I now have a Subaru with EyeSight so I am accustomed to some extensive driver assistance. Especially on the highway but you still need to pay attention.
I do a lot of business travel and have rented multiple Tecla's, a Chevy Bolt (no assist), and a Polestar (limited assist). The Tesla driver assist is outstanding but is still only assistance. It is not fully driving the car for you and you need to identify and pay close attention in more complex situations.
It turns out that one of the things that radar is good for is emergency braking regardless of lighting conditions.
It also helps that other carmakers don't oversell the functionality of their advanced driving functionality; if anything they deliberately understate how well the systems work to avoid giving customers the false sense of safety that Tesla does.
Like… yeah, that’s called an unsafe system my dude.
Newer medical treatments have higher risk, just like experimental transportation methods carry greater risk.
Everyone gets sick/injured and everyone needs to get around, so some suffering in the name of advancing these endeavors seems both inevitable and tolerable. The question is to what degree?
Controlling access to and fallout from these automated driving systems is a temporary priority since the roads they're testing on are far more 'public' than an individual's body undergoing new medical treatment, but the long-term priority must be on getting the systems as safe or safer than other human drivers on aggregate. That will happen, sooner or later, and I'd rather see it sooner as long as the cost of the race isn't catastrophe. The only way to advance is to let it learn...
Humans are able to drive with vision alone in part because we have a high-level conceptual model of the world that we can rely on to fill in the gaps when a straight literal interpretation of vision is wrong or inadequate. We know, for example, that there are not likely to be walls under highway overpasses and that stop signs on billboards are not signs. This is a "strong AI" level problem that a car isn't going to be able to to solve, so instead the best answer is to give the car super-human senses to it doesn't need such a model as much.
That and if we have self-driving cars we want them to be safer than human drivers. That could only reasonably be achieved with super-human senses.
Where's the data that proves this? Tesla has ~2% of the auto market, but ~0.4% of deaths.
I am not aware of any truly fair and impartial comparison of modern Teslas with other modern self-driving (whatever level) car technologies that have shown that Teslas are truly prone to more accidents. Fair and impartial would involve comparing rates, rather than absolute values (although that's just a starting point), correctly determining what model and what technologies were enabled, along with the driver's actions. Most newspapers are looking for clicks and "tesla autopilot crashes" get clicks, and the papers don't have time to do high quality investigative journalism.
(i have no opinions on the matter; I don't own a tesla, don't know much about their technology beyond what I read in the press, and have limited real-life experience with "self-driving", basically just the standard lane-keeping and radar cruise control that's included with my car, which I don't use often, and have found is prone to all sorts of problems)
They weren't offering half the functionality AP claims and they realized the need to track driver attentiveness more robustly than looking for steering wheel torque.
And even after incidents like this Tesla resisted for years before actually using their cameras for eye tracking
As the default assumption when dealing with safety-critical systems is that they are unsafe and the explicit burden of proof is on manufacturer to prove safety, we must assume that it is unsafe. In addition, as no untainted safety analysis can occur at this time, any comparison against alternatives is impossible until they stop blocking audits of their usage data. The only thing we can state for certain is that it is unsafe for any usage by customers since no untainted proof of safety has been produced.
[1] https://thelastdriverlicenseholder.com/2023/02/17/2022-disen...
[2] https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...
But the 17 involving a Telsa is where we need to pay attention. LOL