I guess you missed the news today about no human safety operators in some Austin robotaxis.
But, even before that, the robotaxi software was never reported as turning control over to the safety monitor in the passenger seat.
Oh, so now you’re going to get pedantic about quotes? As I posted, the article clearly calls out “full redundancy” on power steering, adding what “the engineers also mentioned two…” If you’re not going to believe the article’s authors were telling the truth when they say they talked to Tesla engineers (plural), then we’re done with that part of the discussion.
Ah, you’re confused by a single housing! Here’s a video of an actual tear-down the Model 3 steering rack:
“What’s interesting about this is that there’s two sets of control and power supply…it’s actually for a redundant…you’ve got a back-up for power AND control.”
Oh, we’re back to this again? That it’s OK for Waymo to claim things they haven’t done, like drive without premapping or drive without LiDAR and radar, but when it comes to Tesla, you have a different standard.
..Alphabet’s “Other Bets” (dominated by Waymo) had generated only a few billion in revenue but incurred tens of billions in operating losses over time..
Waymo remains unprofitable at the unit level; it loses money on nearly every ride.
Robotaxi and autonomous service revenues are growing, but costs (vehicles, sensors, safety, mapping, scaling) significantly outweigh revenues today.
“I feel very confident predicting that there will be autonomous robotaxis from Tesla next year — not in all jurisdictions because we won’t have regulatory approval everywhere.”
“From our standpoint, if you fast forward a year, maybe a year and three months, but next year for sure, we’ll have over a million robotaxis on the road.”
Posts like that have been there for years, but without any focused development on the taxi-specific technical issues. I read that as not immediately serious.
I said they use safety operators. I didn’t say they were in the vehicles. The safety operator is in the chase cars. I’m astounded you didn’t know what. We had a couple threads about it.
And yes, the human operator has to take control of the vehicle. That’s why Tesla’s are L2.
This is one of the most unintentionally funny posts I can remember in a long time. Let’s set the table. This is what I said:
You’ve been pushing back on this comment for weeks. So let’s watch your video looking for each of the components and see if my statements are correct.
First at about 6:18, he points out the ball screw. Count the number of ball screws you see in this photo (I added red arrows pointing out all the ball screws to make it easier).
Keep in mind Tesla issued a massive recall for problems related to the ball screw.
Next is the drive belt. Carefully count the number of belts you see. This will be a little trickier because I didn’t identify all the drive belts in this photo, so take your time.
Now we get to the point where the steering assembly has been almost completely taken apart and we’re left with the motor. Count the number of motors you see in this picture.
Hint: You can tell by the number of motors by the number of drive shafts (red arrow).
Second hint: He tells you the number of motors several times in the video.
So according to your video if I’m counting correctly we have:
One ball screw
One drive belt
One steering motor
It is almost like I’ve been saying this the whole time.
But it is even funnier than that. A while back, I saw that old Smorg was confusing the steering on the Cybertruck (which is fully redundant) with the steering systems on other Teslas, which have some redundant components, but lack a second steering motor, a second belt, and a second ball‑screw.
I thought what Smorg needs to understand the issue is a good tear down video so he can see with his own two eyes that how Smorg was describing the Tesla steering system had no relationship to how steering actually works on most Teslas.
So I posted this exact same video in this exact same thread a couple weeks ago. Apparently you chose not to believe your own lyin’ eyes.
This doesn’t sound plausible to me. In 2016 Elon Musk wrote in Master Plan Part Deux:
So, in short, Master Plan, Part Deux is:
Create stunning solar roofs with seamlessly integrated battery storage Expand the electric vehicle product line to address all major segments Develop a self-driving capability that is 10X safer than manual via massive fleet learning Enable your car to make money for you when you aren’t using it
Two of the four come down to robotaxi. So I don’t buy they are just now getting around to working on a core part of their business strategy.
A much more plausible explanation is that they had to go vision-only as a practical consideration for consumer sales, and now are finding it difficult to make that work for L4 autonomy.
As you so nicely pointed out a while back, just as the suspension is not part of the ADS, neither is the steering. What is part of the ADS is the control aspect, for which Tesla is fully redundant, as shown in the tear-down.
You’ve said the Cybertruck’s steering meets redundancy standards, yet there is still just one rack, according to Sandy Munro. Why one rack is OK but not on another is enough to dismiss your arguments here.
Your belief that regulators will not approve Tesla vehicles for L4 because of a lack of redundancy is incorrect, and investing based on that is flawed.
I think it is okay to fail forward. I don’t have a problem with setting far reaching goals and coming up short. It is okay to swing for the fences and miss. At least you took a swing.
My only point is that the narrative has shifted from Tesla is the undisputed leader in autonomy to Tesla hasn’t really started working on autonomy until recently.
No. Autonomy has been Tesla’s core focus the entire time. This has been confirmed over and over again in public statements. Right now Tesla is behind in autonomy. It is possible their vision-only approach will catch up.
But Tesla clearly has been focused on autonomy the entire time.
Wait? What??? The ability to steer isn’t included in the ADS? If the steering system fails how does the car navigate? Telekinesis?
I retract my statement. I was under the false assumption that the Cybertruck steering system was fully redundant. But I will accept your passionate assertion that it isn’t.
By the way, I don’t understand what you mean by “rack.” According to your video, the steering wheel is not mechanically connected to the front axle. The front “rack” is an integrated unit with two electric assist motors mounted on it. If one motor fails, the other can still steer the truck.
At least, that’s my interpretation of the video. Clearly you think my understanding is incorrect. So I will concede your point that the Cybertruck cannot L4 capable except in very limited circumstances.
Oh, and not to put too fine a point on it. You’ve been pushing back my statement for weeks:
So based on the video you posted, which of those statements are incorrect?
In your video, can you give us the timestamps where you saw:
A second steering motor
A second belt, or
a second screw ball
I realize, of course, seeing for yourself a physical breakdown of the components doesn’t hold as much weight as “direct quotes from Tesla engineers” (chortle).
OK, now, which component in any steering system is considered a “screw ball?”
There is no such thing, except as a type of Marx Brothers movie, not a vehicle.
And let’s also be clear that what you previously labeled as the ball screw is actually just the steering rack. It’s not a complete ball screw. You need the ball-bearing sleeve as well to have a complete ball screw, just as you need Groucho to have a complete screw ball.
Waymo also brags about “independent controllers and separate power supplies,” NOT independent nor separate ball screws, or belts, etc. There is still just one steering rack, with its associated singular tie rods to the wheels. Not two. Not redundant.
The section of the video I linked to confirms Tesla incorporated both independent controllers and separate power supplies, as well.
Neither the modified Waymo vehicles nor the Tesla Cybertruck have two steering racks for front steering. They also do not have two sets of tie rods to connect the non-existent two racks to the wheels.
–
Bottom line here, though, is that you are literally saying that BOTH of these are true at the same time:
• Tesla engineering and management know what is needed, either for good engineering practice and/or for regulations, when it comes to steer by wire for CyberTruck.
• Tesla engineering and mangement DON’T know what is needed, either for good engineering practice or regulations, when it comes to steering for L4 vehicles.
That is a nonsensical argument. They know better than you or I.
If you wish, you should seek out the regulators at the CPUC, and the equivalent in Austin, and write to them before they make a terrible mistake.
Tesla will not have to retro fit existing vehicles with two ball screws in order for them to be Level 4 approved by regulators.
Unless you include the single motor and the single drive belt as part of the control system, which are parts of the ADS that actually turn the wheels.
No, I’ve literally never said any of those things. Not even remotely close. You’re simply making a somewhat bizarre straw man argument that defies logic, reason, and common sense.
Let’s do a little more table setting. I gave you some clues in my post I quoted earlier which you missed, so let’s try again (emphasis added):
Keep in mind the phrase: “it has to, it is drive by wire.” Steering is classified as a primary control system. To be street legal in the US and EU, drive by wire systems must be fail operational. In other words, if some part of the system fails, the overall system itself still needs to work.
To that end, cars with drive by wire systems (Cybertruck, Infiniti Q50, etc) must have duplicate systems to be fail operational.
So in my understanding, the Cybertruck steering system would be L4 compliant. You have pushed back on this, and this isn’t a hill I’m willing to die on. If you say the Cybertruck isn’t L4, that’s fine with me.
Other Telsas don’t have fully redundant ADS steering components because they don’t need them. There is a mechanical linkage between the driver and the wheels, just like in most cars.
So the way a person with a functioning brain would view your straw man arguments is that:
The Cybertruck has fully redundant steering components because it can’t legally be sold in the EU or US without them. Tesla engineers might have had an eye towards L4, but without the redundant steering components the number of Cybertrucks sold in the US or EU to date would be zero.
Other Teslas were never designed to be L4. We know this because besides the obvious steering issues, lots of other components are provably not L4 compatible either.
As an aside, in the last few days on Twitter we’ve seen reports of Tesla robotaxis with camera washers. This makes sense because if the cameras are obstructed the vehicle cannot safely operate autonomously. So Teslas need camera washers to be autonomous.
But regular Teslas don’t come from the factory with camera washers. Which means they were never designed to be L4.
I realize you are talking your book. You need Tesla hardware to be L4 capable otherwise the investment case falls apart. Maybe the Cybertruck is. You claim it isn’t, so I’ll believe you. But the rest of them clearly aren’t for a large number easily probable reasons.
And since we’re talking about the investment case, Elon Musk has stated several times that all Teslas have the hardware needed for autonomous driving. That’s provably not true, so you should take that into account when calculating future cash flows.
Yes, you said those things and more. For instance:
After putting to bed “screw balls” (whatever they are), and multiple steering racks, we get a new strawmen from you now:
There’s nothing in SAE J3016 about camera washers. They’re a good idea, of course, especially for robotaxis, but less important for POVs that are driven in L4 mode and can be manually cleaned by their owners. And now you think adding camera washers will hurt Tesla’s cash flow?
I made no such claim. You have reading comprehension problems.
And we circle, yet again, back to your claim:
Please stop confusing L4 autonomy for owner use with dedicated robotaxis or with using a POV as a robotaxi. These have different requirements.
I actually agree with you that Tesla vehicles sold up through last year will not be approved to operate in an L4 mode without modifications, at least by end of next year; I have a side bet with a friend on that.
I also agree that Teslas up through last year are not suitable as produced for use as an L4 robotaxi without some hardware modifications. But, Tesla has already performed those hardware modifications on the Model Y vehicles currently used in Austin without drivers or safety monitors on board, and in the SF Bay Area with safety drivers, so Tesla knows what it takes and can obviously make those modifications in a cost-effective enough manner. It’s not a reason to doubt that Tesla will get regulatory approval where they seek it out, and not a reason to think Tesla will have cash flow issues.
And certainly CyberCabs will be built for robotaxi use from the get-go, and at a lower than Model Y cost and lower than any Waymo modified vehicle cost.
Furthermore, those Model Y modifications for robotaxi deployment do not include adding a second steering rack or second complete ball screw, and probably not even a second steering motor. Even more, they do not include anything that’s particularly hard or costly. And it’s an unknown question whether those mods will even be available to previously sold vehicles for owner’s use, much less than they’re required for all L4 regulatory approvals.
Tesla will continue to innovate and improve their products. We’ve seen reports of new FSD circuitry that may have 3 SOCs on the board, for instance. That doesn’t mean the previous vehicles are unsuitable, it just means the new ones are even better.
Master Plan is about where they are headed, not about what specific projects are currently underway. Empirically, they weren’t working specifically on taxi functions as opposed to general driving functions.
As I’ve explained to you previously, the SAE standard is almost entirely silent about how L4 is accomplished, it simply defines what L4 is. This seems to be a hang up for you, but assure you, it says this in plain language.
But Tesla’s FSD (supervised) absolutely can disengage if the cameras are obscured. So if the current FSD can’t work if the cameras are obscured, how does a software update fix this?
That’s you saying a vehicle designed to meet J3016’s definition of L4 needs to have camera washers. Now, you’re trying to spin out of that,
Who said anything about a “software update fix” for an obscured camera? Like you said, FSD can turn control over to the supervising driver while Robotaxi can perform a DDT fallback if actually needed.
And I say “if actually needed” because Tesla has multiple cameras and so the system could decide that one dirty camera has coverage from other cameras enabling it to continue in a perhaps degraded fashion that can still enable safe completion of the trip.
I’m just applying basic logic and common sense. If the vehicle is dependent upon cameras (as opposed to some other sensor system) to operate then the sensors need to work.
Right now, if Tesla cameras are obscured FSD will deactivate. Tesla is installing camera washers in robotaxis, so apparently they think this is important.
If it can’t “see” it can’t safely perform a DDT fallback.
Unfortunately, Tesla cameras are not fully overlapping. Failure of a single camera can cause a blind spot. If there is a blind spot, the ADS cannot safely operate.
One of the things I am struck by in this discussion is that there is a contention that if X is not working and not supported by a fully redundant alternative, that there is NO possibility of L4 certification. And yet, nothing is failure proof. Redundant system can and do fail. And, some non-redundant systems have extremely low failure rates, potentially even better than some redundant systems where the components are more vulnerable.
I.e., is the criterion the failure-averse design or the actual failure rate in practice?
Oh, and by the way, radar and LIDAR are only fractionally redundant of cameras, so the camera system has to achieve its low failure rate on its own, regardless of whether radar and LIDAR are present.
