DMV ‘revisiting’ its approach to regulating Tesla’s public self-driving test
latimes.com
latimes.com
This data, https://www.tesladeaths.com/, doesn't tell us much if accurate, other than Tesla crashes don't seem to be any kind of real problem in an absolute sense.
There is quite the difference getting into an accident with a 20 year old Honda Accord and a 3 year old Model 3.
My point is that anti Tesla self driving sentiment is almost exclusively driven by emotion, not hard data.
Show me the data that justifies why I need to be afraid of public FSD.
Burden of proof is on Tesla to show that FSD is safe, not on the DMV to prove that it isn't.
You are welcome to argue that this principle is in need of reform. People may or may not agree with you. The discussion might be interesting.
But like it or not, the underlying principle for an exception to an established safety principle is that the burden of proof is on Tesla.
On a relative basis, it's even scarier: scaled up to Toyota or GM numbers (in terms of vehicles sold), hundreds of people would be dying each year from self-driving crashes.
That's enough reason to be very wary of Teslas on the road. There are others, but even one is enough.
Tesla doesn't release total yearly mileage data anymore so there isn't much to work with. There is some IIHS data and from that Tesla models do worse than comparable vehicles without a doubt (https://www.tesladeaths.com/index.html#FAQ).
...yeah, that's exactly what I asked for in the first sentence.
I'm not defending Tesla here, just stating the obvious. Not a fan of their cars for quite a few reasons
Not really. Most of the data is based on comparing all sorts of cars against all sorts of other cars.
Even ignoring how many years other brands have been making cars compared to Tesla, car models have very specific demographics they cater to.
Take the Corvette for example, one of the highest deaths per mile, and it's not because the car is inherently unsafe... it just attracts a very specific type of buyer with a very specific purpose in driving.
-
Adding in Autopilot makes the comparison even more hopeless since it works best in the absolute easiest situation (basic highways) doesn't work in the hardest ("stroads") and limits speeding.
You could try comparing only cars that have AEB/LKA/etc. but no one seems to have done that (Tesla certainly won't since it doesn't sell their narrative)
FSD would be slightly more interesting, but again, will work best on highways (so people will enable it there more) and also Tesla hasn't shared FSD numbers separately
Just like drug companies have to prove that their medicines work before FDA approves them to be provided to patients, similarly Tesla has to provide data and prove that it works before it should have been unleashed on public roads since both, drugs and FSD/autopilot, have the capacity to kill if they don't work as designed/promised.
Hopefully this evaluation will force Tesla to share the data as the other companies, hopefully the army of fanboys have nothing against transparency.
You can compare all cars vs Tesla and you still have bad stats since Tesla are new and expensive cars, you can't compare with old cheap cars.
You can compare autopilot vs human drivers and is still bad stats because most of autopilot miles are on good conditions on big hihgways so you would need to compare on the exact same roads.
TL:DR you need stats for the similar driver population, similar car prices and age and drive assist, similar roads. The TFA asks for more transparency so we would get more data and maybe we get better stats or more bad statistics, that we need to debug.
1 they are not alpha/beta testing on public roads
2 not all cars track you, my brother Ford won't call to the homebase with crash info when he hit a dog.
Why would the model/price/class of the car affect the crash rate of a self driving car?
I'm not sure its the AI drivers of the future that worry me.
[0]Just about. https://www.gmc.com/electric/hummer-ev
Once self driving is a thing the government should make the cars do a billion mile driving test in a simulation they design, with all possible crazy things happening
I'm pro-EV but Tesla is a reckless shit show.
Have you been able to watch Twitter videos of all driving situations FSD is exposed to, or was your research biased to "videos that get a lot of views," itself a subset of "FSD driving that was caught on camera"?
If there were multiple viral videos of a single person veering into oncoming traffic, I wouldn't care how many hours of video there was of them NOT veering into oncoming traffic there was tbh, I would still like to see some intervention.
That would likely convince the DMV, and opponents. Just think of the win in terms of PR.
Thought experiment: view all cars as a single imperfect product that kills 1.3 million people and injures tens of millions every year - vast majority due to poor driving. What's you solution? And make it fast - another 1.3 million will die this year.
[edit] I'm pro EPA too btw, its nice that rivers don't catch fire.
What I'm seeing is that Tesla's appraoch, fleet learning, is likely to accelerate the advent of L4. If Tesla can solve this using the existing cheap camera sensors, they can scale this out incredibly fast. This year they are likely to produce upwards of 2 million vehicles, and by 2025 it's likely close to 6-7 million. Even if other self-driving solutions get there first, they have no way of reaching Tesla's scale as they need specialized sensors and partnerships with OEMs etc.
Consider the possibility that I'm right. By stoking fears and getting FSD banned or delayed, many lives would be lost in the name of safety, when there has been no fatality recorded on FSD.
I think this is simply a cost-saving measure. They have not developed a sufficiently reliable system, they are marketing it as though it is, and putting it on the road. I strongly suspect that they do not in fact require more data to perfect this system - I suspect they need more sensors. If they do need more data, they're free to obtain it otherwise.
Autopilot has tried to murder me a few times lol.
I don't want to see that experimental learning happening on public roads which are shared by people who are not willing participants in the experiment.
Every time a Tesla crashes into a barrier killing the driver that is sad, but at least they chose that fate so ok.
When Teslas cause damage to other cars or humans who didn't want to be part of the alpha program, that's criminal.
If I say I just ate a lot of rice, do I define how much rice I ate by the percentage of global rice consumption I accounted for?
-
Please, 2,000 cars is mass testing.
It's not like there's nothing between 0 safety features and FSD. We could just have passive safety features like AEB and LKA and get a ton of benefit over nothing...
Millions of people are able to drive safely, and do, every minute of every day.
73 per cent of American drivers consider themselves above average [0].
[0] https://www.insurancejournal.com/magazines/mag-editorsnote/2...
I don't doubt that there are many safe drivers out there, but probably fewer than we think. With modern cars, you can risk a lot before having an accident.
Basically saying that the quoted statistic (90% or whatever) is a way to NOT look at how road design contributes to the error.
Earlier on this threat, someone commented that a road marked as 50 mph often has cars driving 150 mph on it, because it is a shortcut which is build like a racetrack. Obviously 150mph in a 50mph zone is human error, but a few simple changes to road design could take away the 150mph attractiveness of this road.
Or see some of the articles on stroads, or badly placed left turns, etc.
FSD might be a step to better road safety, but a focus on designing roads for safety might also be a good step.
While we're at it, let's argue that FDA shouldn't worry about drug/vaccine trials? Perhaps as soon as a company makes something, it should be given to the people, since hackers. I'm sure pharma companies will love that, human lives be damned.
To me it's pretty clear that Tesla's audatiousness in pushing FSD could move up the advent of L4 by years, possibly saving hundreds of thousands of lives, for very little downside. In fact, I believe human+FSD is already safer on average, just due to the fact that FSD requires the driver to watch the road.
Are you saying that 'Full Self Driving' is safer because it is not full self driving?
I have no idea why that's clear to you. Seems to me from the FSD Beta videos that not only Tesla isn't in the same ballpark yet as the companies that are going for offering geofenced L4, but it doesn't seem to be improving much. Then there's the fact they've handcuffed themselves by widely promising/releasing the product with the hardware customers already have.
I want my culture back.
If that is what the community is saying.
May I suggest an alternate "if", based not just on your statement, but on the general tone of this whole article?
I just feel like if the Hacker community is saying "react emotionally based on if we like Tesla or not and just yell at each other without trying to communicate", then maybe I don't want to be a Hacker anymore.