That is pretty weird. And it may well be that the best explanation (for both Subaru and Tesla) is a statistical anomaly–without a close look at the numbers, it may not be possible to know.
(I’m assuming of course that Subaru actually manufactured more than 10 cars for the 2009-11 model years :))
And this is even more crazy: from 2009-2011, the Crown Victoria had a driver death rate of 4 per million miles driven. Meanwhile, the Grand Marquis had a death rate of 57 per million miles.
Seeing as how they are the same car, should I presume that the Ford badge on the Crown Vic makes it 14 times safer? Why shouldn’t I? It seems like the only obvious different between otherwise identical cars.
Crown Vic was only avalable to fleet (police/taxi/livery) customers after 2007. Grand Marquis continued retail sales until 2011.
In the spirit of this thread, I’m just going to deny that and claim it is indeed the Ford badges that are responsible.
Setting the snark aside, I found this paper by RAND which has several interesting insights on the difficulty of testing autonomous vehicles.
One takeaway I have is that measuring the safety of autonomous technologies with the number of fatalities is a bad metric, since it can take decades to centuries to demonstrate the safety if we’re mainly interested in counting dead people. Instead, maybe we should be counting crashes that don’t result in fatalities or injuries, which is certainly not the issue being debated between NHTSA and Tesla.
The NHTSA data was airbag deployments, not fatalities, so I’m not sure what your point is.
And accidents that deploy airbags is only a fraction of crashes, just as fatalities are a fraction of airbag deployments. Note that Tesla is basing its claims of safety on the fatality rate: cite.
I’ve only skimmed the RAND paper so far, but I will note this: Tesla *does *have hundreds of millions of miles (and perhaps billions at this point) of Autopilot use. By the paper’s own metrics, that is enough to establish a 95% level for a 20% reduction in injury rate (77 per 100 megamiles). The paper appears mostly targeted at Waymo and the like, which run only small fleets that can’t possibly rack up so many miles.
Tesla’s claims about the fatality rates are indeed statistically untenable by themselves. However, if injury rates are a good proxy for fatalities, then the claims are reasonable. They would need additional supporting data to show that it is a good proxy–but that could be done independently and wouldn’t necessarily have to be limited to solely Autopilot data.
If you are uncomfortable making this leap, then so be it: we can still conclude (potentially) that Autopilot reduces injuries, even if we aren’t certain about fatalities. That’s still a win. Heck, even without an injury, avoiding a crash is still a huge win.
Here is an article quoting two professors who say that Tesla’s arguments supporting its claims of safety are absurd.
I already acknowledged that Tesla’s claims about fatality rates cannot be supported based solely on the small number of incidents. That same reasoning does not apply to injury or overall crash rates, since the numbers for the latter are significantly higher (70x to 350x).
The CS Monitor article focused solely on fatality rates, and while entirely reasonable, does not contradict anything I’ve said.
Ok, this is pretty cool:
Tesla supports a full web-based API for their cars. You can query all sorts of things like temperature and location, as well as changing the state of the car–including starting and stopping charging.
You know how I was complaining about not having a charging stop time? Well, it’s really more complicated than that–I have a time-of-day rate plan that is different between weekdays and weekends, and ideally the charging would follow that schedule without me having to fiddle with things.
And in fact I can implement exactly that. I can also geofence it to my condo so that it doesn’t do anything weird when I’m on the road. I have to be a little careful about not waking the car up too often–it goes into a deep sleep by default and if you poke it too often, it’ll burn through 20 miles of charge a day instead of the 1-2 that I see. But that’s all solvable.
Obviously this stuff isn’t a benefit to average users but for a geek like me, it’s great. It’s all pretty straightforward OAuth token authentication with JSON data. Really easy to whip up a Perl script that does what I want.
To anyone that’s interested, the API docs are available here. I can provide sample code if requested, but really it’s not too difficult for any language that can do GET/POST requests with custom headers.
Those commands are the same that is available through the app. I don’t see anything that relates to charge times starting and stopping.
That is cool. I like the idea of being able to do complicated and arbitrary logic, such as, stop charging at 5am, unless below 50%, then keep charging; and if the TRIP flag is set, then charge to 100% regardless of anything else.
Right, but it doesn’t need that --the script handles it. The script just runs continuously as a little background process on my computer, and does stuff like “if the car is by my condo and it’s 3 pm on a weekend, then stop charging”. Or arbitrarily complicated scenarios like what echoreply mentioned.
Got my VIN (23xxx) assigned today, so it looks like I only have a short time left to wait.
Was all set to get parts to run a 50A line in my garage to support a NEMA 14-50 plug, when I looked at my existing breaker panel and noticed (feel stupid for not realizing this earlier), that my house has one extra breaker labeled as “A/C unit” than I physically have A/C units. My house used to be the model home for our development, and they used to have an office set up inside the garage. And, this being florida, they needed air conditioning. They ripped out the A/C unit before selling the house (and patched up the hole where it looks like the refrigerant lines used to run through the garage wall), but left the 10Ga Romex wiring in place to a 2-gangbox on the garage wall, right where I currently park my car.
So turns out all I have to do is install a 14-30 outlet (I bought one of the Hubbell industrial-use ones for good measure), and I should be good to go. Won’t charge as quick as a 14-50, but since the Tesla mobile connector maxes out at 32A when plugged into a 14-50 anyway, it’s not much difference in time, and still enough (22 mi/hr) to pretty much get a full charge overnight. May eventually decide to run a new 6Ga line anyway for the 14-50 (30 mi/hr), but at least I’ll have something set up as soon as the car gets here, that is plenty sufficient for my commute (~75 mi total a day).
True, I wish those things were native to the app. Seems like pretty easy QOL things that can be added actually, and they have promised app upgrades in the future for time of use charging, etc.
Lame question, but I’ll ask it once the normal way, and then once in the style this thread has turned to.
Dr. S, how do turn signals display themselves? I assume there’s arrows that pop up near the speedometer on the center display. And is there a traditional ticking noise, too?
How do milliamp Hertz turn signals pound-feet display themselves OLED pixel 4k? I assume there’s arc radius quantum arrows that pop up 9/m/s/s vector near the speedometer furlongs/fortnight 16 amp ROMEX on the center display IMAX 70mm Panaflex. And is there a traditional DC 1.6v ticking 65 dB noise too jargon jargon SQL megabyte?
The turn signal stalk itself has a soft and hard push in each direction. A soft push just blinks the lights 3 times in that direction. A hard push does the normal thing where it stays on until you’ve completed a turn. You can cancel the signal with a soft push in either direction.
Doing a hard push also initiates an auto lane-change when in autopilot mode. One has to be a little careful here–although it won’t change lanes directly into a car, it doesn’t yet detect if someone is coming up at high speed. So, use the same discretion as manual driving.
Nice! I have heard of some people sitting on VINs for weeks, so I don’t think it’s guaranteed that you’ll get yours soon, but it’s a pretty good bet. Hope you’re one of the lucky ones!
That’s great news about your outlet. You’ll need it, though; 75 mi is a hell of a commute.
Agreed. A lot of this stuff could be rolled into the app pretty easily. But in the meantime, the API tickles my need for fine-tuning this stuff.
Supposedly there’s an OTA update rolling out soon. Big feature is apparently auto-dimming headlights (something that Musk said would be coming eventually, so this can go in the category of “promise fulfilled”). They’re really pushing the updates out at a decent rate–I hope they keep it up.
My biggest UI complaint at the moment is with USB audio. As mentioned, FLAC over USB is the best way to get music to the car and sounds genuinely fantastic. However, the file navigation leaves a lot to be desired. For instance, an album with multiple artists will show up as separate albums for each track. Also, it doesn’t really resume where I left off most of the time.
So there’s still plenty of stuff to fix, even if it works well overall. It’s nice to know that there’s at least the possibility of it getting fixed.