Sure, there are some narrow situations that won’t be covered. Team driving doesn’t seem too common–there are 3.47M class 8 trucks in the US and 3.5M drivers. Although I wouldn’t expect a perfect pairing, this doesn’t seem to leave room for huge numbers of team drivers.
Beefing up the grid will certainly need to happen. Tesla is working on their “megacharger” network to solve this problem. A couple of decades is a lot of time, in any case. I’m not predicting that they will switch over wholesale as soon as the first semi rolls out of the factory.
Another possibility is trailer swapping instead of battery swapping. Stop at a recharge lot with robotic trucks and have a freshly charged semi pick up the trailer.
When you’re arguing with car guys, you have to put on a conservative face :).
Basic learning algorithms for robot control is certainly low-hanging fruit. The training can be entirely simulated to keep costs down and I expect this to be part of the standard suite that the robot makers provide.
I don’t necessarily expect full hands-off training (solely defining the success metric and letting the computer to the rest) to be popular in manufacturing because they do demand consistency and predictability. But there are some really obvious optimization steps that can be applied even with current hardware.
Umm…that’s a metric. Once you have enough trials for a given strategy and the outcomes, you can compute the variance. A low variance and high expected value is in fact exactly the “consistency and predictability” of which you speak.
But yes, the metric problem is a complex one. I expect that there will just be best of the industry commercial metrics libraries that engineers will use.
Whether or not this whole process is completely unsupervised just depends on how good the algorithm is. From my own brief foray into this, when I was debugging my early implementation of a recent machine learning algorithm, at first I was examining the algorithm’s choices carefully. Once I got it right, I stopped watching the Atari game it was playing and just looked at average score and variance.
I could see factory managers of the future just checking to see if the line is flowing and a particular inspection station is passing most of the produced products instead of looking into the details of how the robotic system solved the problem.
Well, they’ve assigned my VIN. Been roughly 4 weeks since the reservation. Probably still some weeks to go, but it’s getting closer. Guess it’s time to call up my insurance company. No solid delivery date estimate, though.
I don’t really feel like the board “resident experts” on car manufacturing are doing any kind of service to elucidate anything. I’m going to drop a really cool youtube playlist of car factories around the world:
I’m sure that those on here that have actually worked in large production auto facilities for a major auto maker do, in fact, know what they’re talking about, especially in light of these videos of assembly lines of super low production super cars like the Chiron. Matters of scale times eleventy.
Anyway, every time I hear that Tesla had decided to deliver the 3 to employees first always makes me scratch my head. Why would you do that? If I went to a store, or a car dealer, wanting to buy a relatively scarce item, I’d be kinda pissed at the company for telling me that they were out and that there was a wait because they sold them all to their employees first. Shouldn’t it be the other way around in just about every business?
That’s really good point about employee purchase. I think it’s been suggested that they got it first because it was less than perfect. Wasn’t there a level of secrecy asked of employees?
Sure they do, just their efforts to showcase their knowledge has been lacking at this board. Anyway, that playlist i linked to has like 150 videos, the majority of them of high-volume cars. There’s also videos that give insight to production of components not on the main assembly line, such as engines, upholstery, interior trims, woodwork, etc. neat stuff, imo.
what, exactly, do you want to know which you think hasn’t been adequately explained? if you want someone to tell you something, it’s better to ask than to take snarky swipes at them.
No idea about the secrecy. Maybe Elon Musk just wanted to treat his employees like guinea pigs for the first run while they refine the model 3 and the assembly phase going forward.
I think its fairly obvious Tesla’s intent was to use the employee cars as beta test vehicles, so that initial problems/bugs could be identified on cars that are both close to the factory, and with people that have an disincentive to complain publicly about them.
This was so that they could claim they were “in production” without actually being in production, at least per auto industry norms.
*A Tesla spokesperson said that if the bolts fail, the driver should still be able to control the vehicle. The issue can make the car more difficult to steer at low speeds and to parallel park, and it should “not materially affect control at high speed.”*This is, as far as I’m aware, the first broad recall Tesla has done, and while it isn’t a major safety risk like an exploding battery, it is a serious functional issue and a substantial cost for the company at a time that it is already missing its delivery projections by a significant margin.
Not their first broad recall–they recalled every Model S ever built in 2015 for a seatbelt problem. This one seems similar in magnitude–not an immediate safety issue, but a potential concern that’s best to take care of before there’s a rash of incidents.
It’s not good news by any means, but I do appreciate that Tesla generally seems more proactive about these things than other companies. I’m not aware of any problem on their part that was only fixed after a run of deaths.
Well,it seems that Autopilot’s kill count is up to 2.
Unfortunately I have to say their detractors are right. Both these deaths would have been prevented with properly functioning lidar. In both cases, these are just crashes with a clearly visible obstruction (that the machine vision missed). Also, the recent Uber crash is the same story - a clearly visible victim that lidar would have seen.
They did give us an updated death rate : once per 320 million miles. So Autopilot, even with it’s edge case flaws, is still better than humans. Probably because even though it will sometimes fail to see a stationary obstacle and kill you, the rest of the time it never gets sleepy or bored or miscalculates stopping distance.
Really, all 3 autonomy deaths are from an easy case that a single functioning sensor would have prevented. It goes to show just how bad human drivers are if the death rate is still lower despite this massive flaw. Would be like a new generation of airliner with an altimeter that fails every time it flies over water having a lower death rate anyways - it shows that the previous planes are hugely unsafe.
A quick digression : the reason why you can’t see stationary obstacles with radar is that the beam is extremely low resolution. You get lots of reflections from the environment. While they do scan horizontally, it’s easier to imagine if you think of it as a fixed sensor aimed straight ahead. You get a huge peak from the sum of all the things in the environment that have a doppler shift equal to the car’s speed. So the car either ignores the entire peak or it may use it but only as a hinting sensor feed to the machine vision.
You then get a series of smaller peaks. The biggest peak is going to be the car traveling with you, right in front. It will hopefully have a small doppler shift (or you are about to crash). Whether the shift is positive or negative tells you whether you should speed up or slow down, which is the core functionality of auto-cruise control.
There will be other peaks. One neat one is you get a peak from the car in front of the car in front - Tesla detects large rapid shifts in this peak to alert the driver if the car in front crashed.
Once per 320 million miles of driving sounds great but what’s the number of times it’s missed something obvious? A person crossing in front of a vehicle is where it should shine.
On a different note, Tesla has voluntary recall of 123,000 Model S cars for power steering issues. What happens to an automated car when that happens? Does the computer understand it has to apply more power to the steering if it binds up?
IMO, and this is a get off my lawn moment, computers cannot be programmed to think better than a human. Faster maybe, but not better. They should be used to augment human control and not replace it.
It’s now become distinctly possible, though not outright likely yet, that Tesla folds. It will take a few more significant, not necessarily even massive, problems with production rate (or cost of getting it up to speed) or car flaws. Then the cash burn rate (it’s got to burn cash transitioning from boutique to major car maker, and it has, big time) will overtake the company’s diminished ability to raise more cash in the markets.
I wonder if its fans think the company should be bailed out by the US public if that happens. It could be argued it’s a ‘national champion’ in ‘the future of cars’. I’m sure it will be said. I wonder if it would happen, even lets say hypothetically under a Democratic admin in a few years.
It might not take all that long. While I wouldn’t take the opinion of one market analyst as gospel, it is pretty obvious that Tesla’s market valuation is based on speculation of future value and not on positive cashflow in the present or near future. And when you can’t pay your suppliers it makes it tough to ship product.
Yes, a more neutral take not dependent on any particular aggressive ‘call’ by one analyst would be to look at the credit rating, now B3 by Moody’s. In 1983-2008 B3 credits defaulted within a year 7.3% of the time.
Another one that would give an answer in the same large ballpark would be the yield on Tesla’s benchmark 2025 bond, 7.7%. US treasuries of that maturity yield around 2.7% so a 5% per year over 7 yrs ‘risk neutral’ probability of default (ignoring other differences like liquidity and tax treatment of govt v Tesla bonds, speaking very roughly).
So Tesla going bust has become distinctly more possible in the market’s view than not long ago, though not outright likely yet judging by rating and bond price.
But like the analyst says, and it’s conventional wisdom, if Tesla does go down it’s relatively more likely to happen sooner than later. The risk premium if 5% per year is probably somewhat concentrated in earlier years IOW. That is when it might well burn up too much cash while sowing too much doubt (among investors who’d put in any more cash) transitioning from niche to fairly big car maker. Assuming it can reach profitability as a quasi-mass market maker in the next year or two, then it would have more leeway to fall short of being the super-est car company ever, which its fans see it as, but without going bust either. There have been plenty of mediocre car companies which lasted for decades once they got to a certain scale.
a. They run totally out of cash, can’t borrow any more money, investors have forgotten all about the Falcon heavy car commercial and all their successes, and are forced to sell the brand and IP to GM or Ford for pennies on the dollar.
Likely? How likely are the same investors who drove their stock price up so much going to ditch them now?
b. They do manage to beg and borrow enough to get into full rate production of their products. They create and release additional products as well, and have a smoother time getting those built. But what happens is, competing car companies - you know, those companies that are a century old - clone their every trick, creating products that consumers consider just as good.
Maybe this will happen.
c. They take over the auto industry and buy the assets of Toyota or another major with their immense profits.
This seems unlikely.
Another factor is that mass vehicle autonomy potentially throws a wrench in the whole automobile industry. *If *it happens.
The prediction is that as most cars on the road become part of large rental fleets, the rental fleet owners will have a lot more market power with automakers. They will demand much more reliable and durable cars - and get them.
So initially, autonomous cars are predicted to lead to a big expansion in the auto industry, as about 1/4 of the world’s vehicle fleet need to be replaced with vehicles that cost about twice as much. But then, as the autonomy fleets have expanded to meet demand, and replacement rates slow down, it would cause a contraction. Maybe. Too many other outcomes to be certain.