Tesla cut a steering component from some cars, didn’t tell customers
cnbc.com
cnbc.com
" So the cars in the fleet essentially becoming self-driving via software update, I think, might end up being the biggest increase in asset value of any asset class in history"
This sentence is packed so tightly with irony that I can't believe it was uttered by a self-aware human. Yes, it's a great idea that two ton vehicles capable of 100mph just "update" themselves to be fully self-driving. Because what could go wrong? Forget about terminators, just the strategic risk in the event of a cyberwar. Astonishing lack of awareness. And he complains that AI is "summoning the demon"?
And yes, they are just "an asset class" whose value needs to be increased because STONKS. I can't even.
The whole thing is a stock pump that went too far
If you have FSD Beta 10.10 installed in your car (which 60K+ people have in the US), you can input a destination and it will take you there. That includes stop/go at green lights, changing lanes, going around parked cars, making un-protective left turns etc.
It's not perfect, but it's beta, and the iterative improvements have been very noticeable over the past few months.
edit: FSD is clearly still alpha; ive seen it steer towards on coming traffic, and almost cut off large vehicles but for human intervention.
People will be elated if they can buy a car that can chauffeur them around as well as a teenager who's had their learner's permit for two days. At least with the car you're already in the driver's seat and don't need to articulate to it what it's doing wrong. Sure you'll have to keep an eye on it to make sure it's not setting itself up for a situation it can't handle but you have to keep an eye on the teenager too.
Crappy half-baked automation is prevalent the world over in all sorts of situations. The normal, unwashed, much derided masses are perfectly capable of learning its quirks so they can keep it out of trouble. The idea that a car "can't see" bollards or bike lane marking or partial lane obstructions or whatever is not a hard one for people to grasp.
Considering how strongly HN advocates for strict liability for car drivers I find your reply baffling. Why wouldn't the driver be liable?
There's a whole field of high-assurance software, with verification, formally-proven compilers with translation validation. Only there is there hope that one can write software and prove it, with a machine-checked proof, correct. Tesla is not that far. Not even close.
But no matter, let's just deploy it anyway.
Tesla can’t even implement auto breaking on a straightaway that doesn’t kill model pedestrians in controlled testing. Only the truly credulous think they can get to FSD in any realistic timeframe let alone Musk’s fantasies.
So far it's ridiculously underwhelming, even I didn't think it was going to flop that bad and I was very skeptical to begin with.
Tesla users are experiencing some sort of Stockholm's syndrome, it's scary
I've heard that way too often.
Tesla's FSD will never come to fruition as a stand-alone system.
Sure, it might be more precise than what is possible with vision alone, but it's also not needed.
It is the absolute standard in any safety-critical industry to have multiple sensors all feeding data and making up for each other's weaknesses. Cars are not excluded from that.
And I am talking about LIDAR not RADAR, even though RADAR also could be used as yet another sensor in the stack. Using only regular vision is a typical dumb Musk thing; LIDAR was quite a bit more expensive when these self-driving things begun to take hold and Musk didn't want that cost so he went all-in with his own dumb solution of "software will fix it."
By now the launch escape system has been scrapped wholesale, the new approach; Just have to fly enough rockets, without exploding, until safety becomes a side-effect of economies of scale and improved tech.
It's a combination of "tech your way out of the problem" with "grow your way out of the problem". Which might be a viable approach for creating non-integral software products, but seems like a really bad fit when peoples lives are at stake.
My Tesla once tried to steer me into a guard rail (multiple times) because the setting sun caused a lens flare. You can't tell me that this is well tested technology that will enable level 3. Only after doing that twice, autopilot noticed it and disables autosteer.
The cameras also have condensation and freezing issues, causing them to not work on cold, wet days.
This was a car that was, according to Elon, fully equipped for self driving.
I really enjoyed autosteer, but there's major flaws in the current implementation and I seriously doubt they'll deliver on current hardware. If they don't, I'm really curious about the outcome of the lawsuits.
The concept that automobiles will travel without human assistance to override IO mistakes from sensors is ridiculous.
Compared to actual safety critical automation systems, designed to operate with humans? It hasn’t even started.
Edit, as I pressed enter by mistake.
What’s happening now isn’t testing, as a system as whole isn’t ready for that. It starts with proper design and long validation, before you can start talking about proper testing. Tesla is rewriting all parts of it constantly. What’s happening right now is mostly PR exercise, plus collection of low quality training data.
Moving to automatic driving will probably be a thing long term, but for the short term, it's a human problem. If only people were held more accountable for driving badly.
In https://www.pbs.org/wgbh/nova/planecrash/risky.html I did find some older numbers. For commercial carries, it was 1.9 deaths per vehicle mile instead of 1.3.
https://www.pbs.org/wgbh/nova/planecrash/risk-01.html (99 through 03)
However, private planes are more risky and don't seem to be included in those numbers.
Cars crash in pretty much every situation. Fatal incidents are more clumped but no where near as much as planes.
The problem is that no one wants to be regulated. Just look at crypto, people love the lack of regulation. However, once there is a crash or enough people get scammed, there will be demand for regulations to prevent that from happening again.
https://video.hardlimit.com/videos/watch/ede3d957-2a1f-4ec4-...
The preceding bit, just as bad IMO.
> "There's still a mechanical component of course. But in today's vehicles, when you 'turn the wheel' you are providing an electronic signal telling your car to go left or right."
Tesla’s aren’t drive by wire.
Electric power steering allows for configurability of steering difficulty, which tends to engender the feeling of precision at a certain point along the effort curve. (Variable for different people, naturally).
Plus, with more powerful electric systems (better alternators, basically), moving hydraulic steering pump from belt-driven frees up the engine a bit, yielding less drag and better fuel efficiency.
But fundamentally, I believe that in modern vehicles, due to their weight, direct steering linkages are a bit rare.
Also this situation is no different than normal power steering, which assists you with a different mechanism.
This powersteering unit now has a lot of field data about failures - they have 3M+ power steering units on the road.
So they may now have the confidence to put the powersteering into that category of "single redundant, but sufficiently reliable to be acceptable". Whereas before they had that data, it wasn't.
'Internally, Tesla employees said that adding “level 3” functionality, which would allow a driver to use their Tesla hands-free without steering in normal driving scenarios, would need the dual electronic control unit system and therefore require a retrofit at a service visit.'
There was V2, v2.5, V3. With both compute and camera improvements. It's pretty much expected that you will need some form of hw upgrade if your car is >2 years old when the FSD becomes a thing. And it's included in the pricing.
Sounds like the worst that can come of this would be Tesla has to retrofit the part into a bunch of cars for FSD.
Would you expect Tesla to also put out an announcement if they started sourcing their upholstery from a different source, or used different stitching on the steering wheel? None of this thing impact the usability of the car in any meaningful way, and none of them are listed as part of the detailed specs.
Teslas also add new things to their cars all the time, such as matrix headlights that they added to the Model 3 last year, and heat pump based heating system.
> Internally, Tesla employees said that adding “level 3” functionality, which would allow a driver to use their Tesla hands-free without steering in normal driving scenarios, would need the dual electronic control unit system and therefore require a retrofit at a service visit.
How many of these customers are expecting to simply update their car to FSD over the air? I agree, other manufacturers would not have the responsibility. But one of the main selling points of Tesla has been that, one day, your car will drive itself if you choose to buy $X upgrade with the heavy implication of over the air.
Edit: Consider this, in the future, when people go to sell their cars, potential buyers will ask "does this car have the dual ECU?" for the sole reason of inquiring if they need to go to get the car serviced for FSD. If you don't think this makes Tesla bound to inform their customers that their car will obviously be impacted on resale and allow them to back out of the purchase, I don't think we will ever see eye to eye.
I understand your point but it would be just as simple as that, when you purchase FSD, you are notified that you have to go to the nearest Tesla service to get it, at no extra cost (well, beside the FSD cost obviously). It's an annoyance, sure, but not a big deal.
"... except the parts that you'll need to go book a service appointment to have installed at a later date."
Probably not many since the market is China where FSD is not currently offered.
> Tesla did not disclose the exclusion, which has already affected tens of thousands of vehicles being shipped to customers in China, Australia, the U.K., Germany and other parts of Europe.
Do you honestly think that cars produced in a country stay in that country?
I agree that Tesla are lying about the idea their cars have all the hardware needed for FSD. But that was true (at least in EU cars) before this change.
My reaction was the same as to this news. I don’t care if Tesla made a small change, and I don’t care what Adams eats. I care about transparency and honesty.
I'm not paying Adams many thousands of dollars to be Vegan.
People are paying Tesla many tens of thousands of dollars to do stuff they claim they will do, haven't yet done, and by these actions are even more unlikely to be able to do.
Now, I agree they are not the same severity. But the line you drew doesn't make sense.
What we're seeing here is a rather pesky bias called "Isolated Demand for Rigor". You shouldn't feel bad: it's quite common, very hard to figure out if you don't know about it, and almost impossible if it mixes with confirmation bias. For a fun post about it: https://slatestarcodex.com/2014/08/14/beware-isolated-demand...
Well. For certain values of fun. I always thought Scott was so pissed at this bias he was in a very "fuck you" mood when he wrote it.
If it is impactful to the customer then, sure, a complaint is merited. But if not, not an issue.
But steering by wire? That's what the article implies, though another comment already debunks it. I hope there are exactly no cars on the road today that are steering-by-wire. I.e. they all degrade to non-power, mechanical steering if the electronic or hydraulic gimcrackery fails.
It is also used by stability control to counter-steer small amounts of oversteer. But it can only do so for a relatively small amount of wheel angle change, not turn the wheel 90 degrees or something like that.
If the powered steering fails, then the computer cannot steer. Therefore, there was redundancy in the powered steering system. The redundancy was removed because of a chip shortage, and this was considered acceptable because the self driving feature still requires the driver to keep hands on the steering wheel anyway.
But aren't we pretending that this year the autopilot /FSD will actually live to their names? Or will Tesla put the chip back later this year ?
Electric power steering and steer by wire are two different things. A steer by wire system does not have a "mechanical link" backing up the function of the electronics.
If your Antilock Braking System fails, then your brakes will continue to work as they always did, save the fact that they’ll lock the wheels if you press too hard. If there weren’t a light on the dashboard, it is unlikely one would notice ABS failure (unless, of course, one is in a situation where ABS would come in handy, like an ice-covered road).
Is this even remotely possible?
And fearmongering like this - "We have to keep trying, otherwise it'll be regulated out of existence and we'll have to wait generations" is absolutely uncalled for.
It's being regulated because people like Musk value their cars, sorry, "asset class" more than they do about the risks that they are taking.
This Twitter post documents how many times Musk has promised FSD is coming “next year.” https://twitter.com/BS__Exposed/status/1482465346061848583?s...
Especially since L3 FSD - when and if it comes - requires the second motor.
So, what happens if that electric steering subsystem breaks without a redundant backup? "Customer took the risk, we don't care?"
You even didn't notify the customer about it in the first place.
"Tesla vehicles can still use the current “level 2” versions of its driver assistance systems, Autopilot and Full Self-Driving (or FSD), without the dual-control steering system.
But employees told CNBC if Tesla launches a more sophisticated FSD update, owners with the affected cars who use that premium system will need to get a steering rack retrofit from a Tesla service center."
(Emphasis mine)
FSD is trash, all these Teslas on the road braking to a stop for no reason, driving slow in the fast lane >:( Cut them off the computer wont care.
> There's really nothing in the EPAS system that can wear out, maybe the motor itself...
After the whole logging and wearing down flash memory saga, I'm not sure about that. In theory you're right, but in practice "It's a Tesla".
What level of system redundancy do they offer on brakes?
No customer was ever aware that there was a redundant sensor. To me the 'they didn't tell customers' part of this is really odd, since it's not like it's part of the sales pitch. It's not like when they just stopped adding USB-C ports.
I had to use at least one of these modes twice during the ownership of my car, too. My ignition coil's one output went dead, twice. Once overnight, once during driving at highway speeds. In both cases I went to where I need to go safely, because the car knew what to do, and the facilities to do so.
These features are written in the user manual as fine prints, but they're not sales points. Nobody advertised these to me.
Tesla is selling an autonomous, connected car with enough electronics to fly a small plane, and they rightfully add redundancies inside it. That's great, that's the way it should be. Then they're removing a redundancy feature which might save lives even if it's rarely used.
It's akin to MCAS system. "It doesn't change the behavior, so the pilots don't need to know". We know how well it went.
In one case, my parent's car's ECU (we drive the same make/model) blown one of its coil signalization lines. In that case the car has no backup for example. You need to tow it, because it can't understand that it's output is gone, and can't get into proper limp mode. You need to tow the car.
Cylinder disable on overheat and other modes are proper "limp modes" which keeps car operational when other systems get damaged, but not the ECU.
The automatic transmission has its own limp mode too, which is activated by selecting the 3rd gear and making some other internal pressure adjustments.
"2001 Ford Focus MK.I"
Looks like later switched to Bosh https://www.ecufile.net/ecu-file-description/BOSCH-EDC16C34-... https://www.ecufile.net/ecu-file-description/BOSCH-ME9.0-026...
no second ecu hidden inside
A complete non-story, as long as it still passes safety checks. The automotive industry is always looking to save a few cents here and there - the only requirement is that it does not compromise safety.
> "So the cars in the fleet essentially becoming self-driving via software update, I think, might end up being the biggest increase in asset value of any asset class in history. We shall see."
Elon Musk, the billion dollar man with a billion promises. I'll make a few bets on this:
1. It won't happen this year. He has been incorrectly guessing it at when it will be ready for many years now.
2. If it does happen this year, it will be blocked legally until proper regulation is ready. If a self-driving car crashes, who is responsible? This needs to be answered, because if it turns out to be Tesla, things will get expensive real quick.
3. If it's not blocked legally, it will be disabled for everybody except those on a subscription service:
> Drivers can also buy a more advanced version, called Full Self-Driving, or FSD, for $12,000 or $199 a month (in the U.S.).
It won't end up being quite the value-add it's claimed to be. Whatever happened to the Tesla car that would pay for itself? So far it seems to just require even more money, for features already suggested to be free when the idea of the car was first sold.
> "If something like a chip or an ECU is not providing additional functionality, if it is truly redundant, you may be able to turn it off or leave it out. With chips and software, there's a little bit of wiggle room. I can reassign stuff here and there," he said.
Not entirely true. Whilst the infotainment system may have enough processing power to run all the safety critical tasks, you don't want it to. Crashing your infotainment is inconvenient, but a 5 second reboot is not the end of the world. Crashing your ABS, throttle control, steering control, etc, is not an option.
A lot of safety critical stuff is usually also proven with tonnes of testing (rather than formal validation) - but that testing is highly coupled to the hardware its running on. You know for example that system X will reliably poll some pin at 1k times a second, because it is the only thing running on the chip. It cannot be interrupted or delayed. The more tasks you add, the great the risk of not hitting timing targets, the larger the testing surface, the greater number of possible states it can be in, etc.
FWIW: It's a backup for the power steering system, which is in fact not a safety device even in modern vehicles. Tesla's steering is mechanical, like almost all cars (obviously a true drive-by-wire vehicle would need a ton more redundancy, of course).