There’s no need for solid state, though. It’s true that there could be a short period of time (<100 ms) where both circuits are engaged. However, this is no problem–household breakers are designed to handle at least a 3x overcurrent for seconds at a time with no problem, and often in the 5-10x range (depends on whether they follow a B, C, or D curve). Motors require some margin when first spinning up.
The relay I want is this one, a 50A DPDT electromechanical model. But it also seems to be out of stock everywhere.
I haven’t had a chance to read it in detail, but it generally seems to be within expectations. They managed to produce >2k/wk of Model 3s for three weeks in a row, which at shows that they can sustain that rate at least.
It’s also nice to see the Model 3 compared against the A4, C-class, 3-series, and Lexus IS. It’s outselling all of them except for the C-class, and even there it’s very close. If they can reach even 3k/wk they’ll be solidly outselling all of them. At 5k/wk, they’ll probably be outselling all the rest combined.
The letter also puts lie to the clickbait garbage that came out after some misinterpreted comments by Musk. In particular (bolding mine):
In other words, the automation mistake was just in misjudging the optimal rate of expanding the automation, and has nothing to do with their long-term goals.
Heh. On the conference call, Elon’s giving examples of overautomation–apparently there was a robot to place a fluffy fiberglass mat on the battery pack. It is aptly named… Flufferbot.
eh, I think the use of “eventually” here pretty much allows them to do whatever and still never be “wrong”. 1 year from now? “Eventually still isn’t here yet!” 4 years from now, when, coincidentally, the best practices of all the major automakers happen to be using the same or more amount of automation? “We were right!”
Sure–they don’t get a free pass forever. But one also has to be fair about it; depending on the area we’re talking about, some increased automation should be doable in the relative short term (pack assembly); some requires relatively minor redesign (the solution for the Flufferbot problem was to eliminate the fluff); and other parts require a non-trivial redesign (like the reduced wiring harness for the Model Y). So it may well be that some stuff will never be solved for the Model 3, at least for this version of the platform. Doesn’t mean they have given up on it.
That conference call was pretty funny. Elon completely shot down some analyst with a boring question about reservation numbers, and then let some guy from a YouTube channel and a list of crowdsourced questions go on for like ten questions. And then basically ended the call with “if you can’t handle volatility, don’t buy our stock.”
Probably not a ton in the grand scheme of things, but Flufferbot was described as having a complicated computer vision system, so probably not some cheap off the shelf thing.
Of course, computer vision is currently undergoing very rapid progress, so what’s true now may not be true a year from now, and almost certainly not true five years from now.
All that said, automated handling of soft items like textile and “fluff” is notoriously difficult, so it may be a while before that kind of thing is really handled properly.
I think it’s foolish to ignore what other car companies have done right for efficiencies of production, and invest what is very likely a ton of money in immature technologies, especially when the company is very unprofitable. Then having those investments fail to the point that 400 assembly workers need to be hired each week for… who knows long… I mean, wow.
You have to admit: if this were any other CEO than Elon, that CEO would be out of a job.
There are a handful of others that could get away with it, but sure. But Elon has earned an enormous amount of credibility from the success of both Tesla and SpaceX–both of which were founded on the promise of then-immature technologies. Lithium-ion automotive batteries and vertically-integrated aerospace manufacturing were huge bets that paid off.
Tesla is probably farther from the edge now than at any previous point in their history. They may still be losing money but they can get away with some experiments. They can’t survive complacency, though.
Not exactly. The basic point is that Tesla’s data shows a dip in Autopilot use whenever an accident is reported. This causes a decrease in overall safety, because on average Autopilot is safer than human drivers. Journalists are thus indirectly killing people.
This isn’t exclusive to Tesla, of course. Every time there’s some report on a new cancer treatment or nutrition finding or whatever, the results are misunderstood (either due to poor journalism or poor comprehension) and people end up dying because of it. And then of course there’s blatantly false journalism like linking vaccines to autism and the like.
So Elon should probably settle down a bit, since clickbait is just the modern form of yellow journalism and is nothing new, and while it does harm people, any purported cure is worse than the disease. Still, it’s understandable that he’d take the attacks personally.
From my perspective, the concern is not just theoretical either. Immediately after mentioned that I got the Autopilot feature, my mother emailed me saying “Don’t use it! Some guy died from it!” I calmly explained that the accident was a rare incident caused by complacency and that it nevertheless saves more people than not. But still, this kind of astatistical thinking is rampant among the general population.
What would the alternative explanation be? The feature does nothing unless engaged, so if there was a statistically significant drop in crash rates after installing the feature, there aren’t many other explanations. It’s entirely fair for the NHTSA to hedge their claim, but that doesn’t mean that a genuine reduction in crash rates isn’t best explained by Autopilot doing its job.
I don’t know enough about the subject to know whether there are significant variations between models of cars, or model years of cars. If there’s variations between types of cars that aren’t equipped with Autopilot, then we should consider whether such variations are attributable to other factors.
But it’s perfectly clear to me that you can’t measure Autopilot safety if you aren’t measuring if Autopilot is on. Maybe people own cars that are equipped with the latest Autopilot version, but not using it at all, and are driving more carefully.
Tesla reports that Autopilot drivers spend roughly half to 2/3s of their highway miles with Autopilot on. You can choose not to believe them, of course.
The nature of statistics is that you can never be fully confident in any one explanation. There are always alternatives. But generally, we can look at the alternatives and see which is the best explanation. There’s not much reason to believe that people drive more safely simply from having installed a feature. And arguing about feature variations between between vehicles is not very convincing since they looked at airbag deployment data, not fatality/injury rates or the like.
The NHTSA didn’t want to commit to a strong claim about the merits of Autopilot, which is fine. But even if you think the alternative explanations have reasonable merit, your claim that there is “no evidence” of a safety improvement is false. The evidence may be weaker than you’d like, but that is not the same as no evidence.
I just looked up some fatality data by model and model year.
From 2006-2008, the Subraru Legacy had 83 fatalities per million miles driven, which was among the highest of midsize cars. From 2009-2011, zero fatalities, the lowest of midsize cars. The Insurance Institute for Highway Safety crash tested each model year of those cars, and rated them “good” in all aspects tested.
How does a model of car, rated to be just as safe in crashes for five years running, drop it’s fatalty rate by 100%? I have no damned idea.
But if the number of fatal crashes for one model of car can vary so greatly, I’m strongly inclined not to assume that Autopilot is responsible for a drop in fatalities in another model of car.
It could be Autopilot. It could have literally nothing to do with Autopilot and be just pure chance. But at this moment, I’m more confident that hat we’re just hearing more of the Musk Reality Distortion Field in effect here.
In more mundane news, the number of Model 3s at work increases unabated. There’s now at least one of every color. I count 10 just in the part of the lot where I usually park. That’s maybe 1/4 of the total spots, so there’s probably 40 or so total. I think there may be more Model 3s than any one other car model.
We’re a Silicon Valley company not far from Fremont and with some connections to Tesla, so this isn’t a huge shock. But still pretty cool to see so many around, especially this quickly.
