Absolutely.
Here’s a Waymo blog with some cool video of simulation, including lidar simulation.
I’m glad we are back to a happy state, again, this is a discussion board.
Those restrictions you list are the same that Tesla has with human supervised taxis.
Geofence, ride around with lidar to “pre-do-something” (maybe not mapping, who knows, but well documented).
In short, Tesla is not operating its Robotaxis in a meaningfully less restrictive scenario than Waymo, although it’s really not even comparable because Tesla is not autonomous.
Waymos do highways, go over 35 mph - you know all this.
Yet today, Waymos are generalizing as mentioned above and at 150 million autonomous miles and accelerating.
Tesla anywhere, geofence, no geofence, only operate with human supervision.
Tesla only drives coast to coast while supervised. Pretty sure Waymo can do this.
As far as solving different problems, I have seen Waymo and Tesla both working on robotaxi for some time and both in the Bay area and Austin.
That looks like about the same problem to me.
You mention POVs. L4 autonomous POVs look far away to me.
Do you think Tesla will go true autonomous with some meaningful L4 mileage accruing for taxi first or POV first?
All indications to me are taxi first, but who knows.
I get that dev on taxi and pov can happen at the same time and inform each other, but there is a lot to be said for solving the simpler problem first: taxi, and then incrementally building on that.
1 Like
I haven’t seen that claim. Seems high. That would 8x from end of 2025.
Do you have a link in which Waymo claims autonomous customer fares in 40 cities by end of 2026?
You’re fond of insisting that Tesla show the autonomous miles. Yet you won’t apply that to Waymo. You should be insisting that Waymo show you the coast to coast driving while supervised trips. They don’t have any such trips. And, since they haven’t pre-mapped the roads necessary to go coast to coast, they can’t yet make such a trip, especially without any interventions.
I know it’s hard for you, but it’s different problems in different orders. Waymo started robotaxis way before Tesla, by several years. It’s astounding you can’t see the difference, or refuse to acknowledge the difference.
Tesla’s approach gave them many more miles of data sooner and faster, and enabled more vehicles being tested sooner. And if taxis are indeed easier than POVs, you should agree that Tesla should have no issue moving from POVs to Taxis, and will catch up and then exceed Waymo in relatively short order.
But, no, you’ll spin your way out of that, too. Because you think spinning and discussing are the same. One more denial of the different orders and we’re done.
1 Like
Seriously, is this necessary?
Really, it’s just a discussion board.
2 Likes
Pretty sure Waymo can go coast to coast supervised.
On to more interesting things.
Is it different problems now, looking at the taxi efforts?
I see for some time now Waymo and Tesla working very hard on robotaxi. Even in the same two metros.
Tesla is mostly making comments on when taxi will go unsupervised. That looks like the focus for L4 autonomy now and in the near term. I don’t think anyone who follows this would dispute that.
And what have those miles gotten them?
Not L4 autonomy.
With vastly fewer miles, even today while Waymo is accruing millions of true autonomous miles each week, Tesla’s data vastly dwarfs Waymo.
Yet, no Tesla L4 autonomy.
The massive data argument for Tesla just isn’t compelling when we look at what the two companies are achieving in autonomy today, after many years of effort on both sides.
Again, same question:
Seems like you’re arguing POV first.
Maybe Tesla was thinking POV first and last few years shifted to robotaxi first?
They are building a dedicated cybercab, afterall.
But if we compare Waymo taxi to Tesla taxi, Tesla geofences, Tesla pre-validates with lidar (well documented), and also applies the ultimate restriction of human supervision (“young robotaxi, no driving without a chaperone”).
Tesla has not solved that “simpler” problem yet Waymo has solved it to some degree despite a vastly smaller data set.
Is your argument that Tesla hasn’t solved the “simpler” taxi problem because they are also trying to solve the “harder” POV problem?
Maybe?
But that just sounds like a bad plan:
“We are working on a really hard problem that no one has solved.
There are easier elements and harder elements.
I know! Let’s work on the hardest parts first!
And to make it extra challenging, let’s use mostly 1 sensor type and try to do it with fewer modalities of information.”
Are most difficult technical problems solved that way? Take the big, hard step first?
Anyway, Tesla seems to be putting a lot of focus on taxi, with all of the talk the last few years and a purpose-built vehicle, cybercab.
I would hope we can agree on that - a lot of focus on robotaxi and autonomy.
In a way, I think Musk is trapped. He can’t go after the simpler problems first, because if he does, TSLA loses it’s 300 PE and $1.3 trillion market cap.
Instead, Musk (always with an eye to maintain his reputation as a super-genius) has to try and change the world in one fell swoop. As we are seeing with FSD, this approach tends not to work.
For example, Musk is touting Optimus as becoming ‘your personal surgeon.’ He’s going to try and give it a surgeon’s brain and human-like hands.
Meanwhile, there are companies like LEM Surgical (as shown at around 7:00 of the video below) that are designing robots to assist spine surgeons. LEM Surgical is going after the simpler problem and is much more likely to be successful. But, if Musk said that TSLA was going to create robots to assist surgeons, there goes TSLA’s 300 PE and $1.3 trillion valuation.
2 Likes
SAE J3016 spec, page 37:
Fallback performance and minimal risk condition achievement at Levels 4 and 5 require that the ADS is still functional after occurrence of a DDT performance-relevant system failure or out-of-ODD condition. If the ADS is not functional, a failure mitigation strategy may apply (see 3.11 and 8.6).
And in Section 8.6 (pages 37-38) :
ADS-equipped vehicles may have an additional failure mitigation strategy designed to bring the vehicle to a controlled stop wherever the vehicle happens to be.
…
Level 4 and Level 5 ADS-equipped vehicles may also have a failure mitigation strategy of stop-in-place under certain rare, catastrophic failure conditions that render the ADS non-functional through, for example, loss of backup power after initial power failure or incapacitation of the ADS’s computing capability, which render it incapable of performing the fallback and achieving a minimal risk condition.
Also, you claimed:
Here’s a 6.5 year old article on Model 3’s steering, which says:
Tesla designed the Model 3’s electric power steering system to have a rapid 10:1 ratio. The power steering is equipped with full redundancy with separate power feeds taken directly from the vehicle’s high-voltage battery. The engineers also mentioned two electronic modules and two inverters providing “hot backup” to the system if one fails.
It would seem that Teslas have enough redundancy, and should there be a failure regardless, SAE allows L4 vehicle to just stop in place. I doubt regulators will have a problem with Tesla’s hardware redundancy.
Not to worry. That’s the ninth time he’s threatened to stop responding, just in this thread alone.
1 Like
I’m glad you care enough to count, and given my take-down of the hardware redundancy argument, that’s what you care about.
What about the pre-trip mapping?
But the point isn’t that they are both working on taxis now. The point is the trajectory. Waymo has zero in their trajectory for general driving around with no pre-mapping. Tesla has tons of that. What happens next is likely to be very different.
No, it can’t. It’s telling that you are willing to say Waymo can do things it hasn’t done while you say other companies can’t do things they haven’t done.
They’ve gotten FSD to autonomous driving, at least as defined by Waymo themselves in 2020, for which they included “a trained vehicle operator.”
Tesla operates on any road in the US and Canada. Waymo operates on a limited subset of 5 cities. That’s a massive differences, and it looks like Waymo will take decades to catch up.
Oh, so it only took me a half dozen or so post (maybe Goofy can tell us exactly) for you to recognize this (even though you’re wrong on the second “first”)? Good show.
Is it? Is it likely to be very different?
There’s lots of conversation about the bear case on Tesla, but it’s actually pretty simple. The bear case is that Tesla’s technology isn’t going to be sufficient to do what has to happen for robotaxis to “work.” I put “work” in quotes, because obviously it’s possible to run some type of robotaxi service today (Waymo and/or Tesla with safety monitors).
But for robotaxis to scale fast - for anyone - they need to get to the point where the system can run with a pretty small number of humans. How small? Small enough that the entire TaaS program can be cheaper than human-driven alternatives like Uber.
The bear case for Tesla is that they can’t do that with their tech. Right now, after eight months in Austin with an even-more-advanced-than-FSD AV driver, they haven’t reached the point where they have pulled all their safety drivers - but let’s assume they can get there. They need to get to the point where they have few enough interventions that they don’t need all that many remote operators and support personnel (which go out to the cars when remote can’t handle it) for this to make sense economically to start scaling fast. Because the number of interventions is what drives the economics of this type of business - even if the cars can be safe enough to run autonomously, if you have to hire too many people to handle the cars when they don’t do what they need to do, the economics won’t work.
That’s the question. Not whether Tesla’s can operate with a human driver in the car anywhere compared to Waymo’s requirement for pre-mapping - it’s whether the tech is/will be good enough in a relevant time frame to operate with few enough humans supporting it to be competitive with human driven TaaS.
The bear case looking at Tesla says the answer is (and will be for a relevant time frame) “no.” Tesla’s tech needs too many interventions - both in comparison to Waymo and just generally. Tesla’s tech can go anywhere, but it goes anywhere with a really high rate of needing a human to do something. Too high. That’s why they haven’t been able to pull the safety monitors in Austin (despite having claimed they would have done so a month ago), that’s why they haven’t even applied for the autonomy permits in California (which would require them to disclose their intervention rates), and that’s why they are walking back expectations on Cybercab. The tech isn’t good enough to do what they need it to do, and might not be good enough for any relevant time frame.
Human-driven TaaS services have enormous advantages over the “robotaxi” model. They have the cost of the driver, but the TaaS service doesn’t have to bear capital costs for owning or storing the cars. Services like Uber soak up the “unused capacity” of vehicles that people would already have purchased and had in their driveways, so there’s little marginal cost of capital or storage. A driverless system avoids the substantial cost of the driver, but the robotaxi system (unlike the Tesla Network proposal which now seems dead) has some non-trivial costs that Uber doesn’t have - so the intervention rate has to be low.
So again, it all comes down to whether Tesla gets their intervention rate low enough so that the business becomes viable. From a competitive standpoint, Waymo is betting that their extra sensor package and pre-mapping will reduce their intervention rate relative to Tesla’s, so that while they have to pay more upfront for their cars and to move into a city their operating costs will be lower. If Tesla can’t get their intervention rate low enough, Waymo will be right.
3 Likes
Then we are comparing Waymo having focused on this exact problem for many years while Tesla was focused on the general problem for many years and only focusing on taxis very recently, Perhaps Tesla was overly confident (hello Elon) about how easy the taxi problem could be solved on top of the general base .. we’ll see … but the context is very different.
1 Like
Waymo is now autonomous in 6 cities.
Waymo began autonomous ride-hailing operations in Miami, Florida today, covering a 60-square-mile service area stretching from the design district to South Miami.
Correct. And this is a foundational reason why Tesla’s hardware is insufficient to be either Level 4 or Level 5 by definition (except possibly in very limited conditions). Tesla’s failure mitigation strategy in the event of an ADS failure is to hand control back to the human driver.
To put it another way, in the event of the ADS failure in a current Tesla, there is no mechanism or provision to for the to safely vehicle to safely come to a stop other than the driver taking control of the vehicle. By definition that is Level 2.
I didn’t see any secrets revealed, but it is wrong. I previously gave a you a list of single points of failure in the driving system. But because you missed it, here is a schematic. Carefully count the number of control motors. (Hint: it is less than two).
Also, a while back Tesla stopped installing the second ECM due to chip shortages. I never heard if they resumed or not.
1 Like
I don’t think that’s correct. In various interviews Dmitri Dolgov has said Waymo can operate in unmapped areas, but they don’t so for safety reasons. He compared it to using another sensor. The driving computers spend less time validating traffic lights and crosswalk, so they are freed up to process the rest of the surroundings. He said essentially the same thing about vision-only. They can operate on vision only, but they don’t.
1 Like
You’re wrong: Tesla’s robotaxi software does not attempt to do that. There is no human driver to hand control back to. From where do you come up with this crap?
No, you’re wrong. I have no idea where your graphic comes from, whereas the article I linked quotes Tesla engineers directly.
The power steering is equipped with full redundancy with separate power feeds taken directly from the vehicle’s high-voltage battery. The engineers also mentioned two electronic modules and two inverters providing “hot backup” to the system if one fails.
That’s might strong evidence that you’re wrong. Again.
Well, driving safely is kind of the whole point. They NEED maps and they NEED more than vision to drive safely, as Dolgov essentially admits.
Huh? The whole time I’ve made it very clear I’m talking about current Teslas.
All Tesla “robotaxis” use human safety operators, so I don’t understand what software you’re talking about. If Tesla robotaxi hardware/software fails, there is a human ready to intervene.
What you talking about where there is no human driver in a Tesla?
Did you read the article? There are no quotes at all regarding steering in the article you linked too.
I refer you to the Model 3 service manual. The Tesla manual could be lying about the design of their steering system, but I don’t think so.
Love to see any evidence. Maybe a direct quote from a Tesla engineer.
Sure, but that wasn’t the point I was responding too. Tamas suggested Tesla has “zero in their trajectory for general driving around with no pre-mapping.” But as per Dolgov, they do. It works, just not as well as they want it to. That’s not zero.
People are throwing this down as some kind of trump card for Tesla. Waymo is providing one million paid AV ride hailing miles per week. Telsa is providing zero.
But Tesla is the winner because they would rather provide zero service than rely on the twin horrors of mapping and Lidar.
I’m just going to throw this out there: Maybe at some point it is better just to roll up your sleeves and get the job done, rather than complain about other people who are getting the job done.
2 Likes