I find it amusing that people who have nothing to do with Tesla are obsessed with and follow Elon closely. They debate it endlessly what Elon said when.
Long term shareholders are smiling.
2 Likes
There is over on the paid TSLA board, lol.
Maybe I’m a fanboi but I still think that WAS possible had Elon not yanked Model 2 and not burdened CyberTruck with all sorts of new technology that delayed the release and increased the price. And the, of course, had Elon not gone into DOGE.
That’s like judging a chess match solely on piece count while ignoring positional advantages. If Elon’s correct, which he may or may not be, then Tesla’s approach will enable faster scaling to more cities/vehicles than Waymo within a couple years. And in terms of applying that technology to personally owned vehicles, Tesla is probably far ahead for that market opportunity than Waymo/Toyota.
1 Like
Again, only if Elon is correct. If Waymo is correct, and full autonomy cannot be achieved with only vision and instead will require LIDAR and some degree of mapping for many years to come, then they will be very far ahead in applying the tech to personally owned vehicles.
3 Likes
Hmmm… Haven’t we been regaled with criticisms of Tesla cause Tesla is collecting LIDAR data in its “test cars”?
That suggests, at least to me, that IF LIDAR is required, then Tesla has experience, software, neutral net, Dojo, Colossus, xAI, … ie a path to success.
ralph
That’s the bull case. Let me make the non-bull case. Tesla’s approach will be successful if and only if Tesla can get to Level 5 autonomy with the current hardware and sensor package.
Also, how fast to we expect Tesla to scale into more cities? As far as I’m aware, Tesla has not applied for AV ride hailing permits in any jurisdiction. Because the permitting process is time consuming, it does not appear Tesla is in a rush to scale ride hailing. Which is likely the prudent approach, but not fast.
And if Waymo is first by a few years, just as they were with riding hailing, then they get all the scale advantages.
Yes, but the claim and the selling point is that the tech will be applied to currently existing cars with just a software update. You can do some quick math to see what that means. I imagine most owners would want full autonomy. So that means millions of FSD subscriptions at $100/month, with millions more each year. They’ll be making more money on the software than they will on the cars.
But if they have to go back and incorporate LIDAR, then they are just another AV company.
Haven’t we also been told that Tesla is using Lidar in a limited role, and hasn’t integrated it deeply into any of their software stacks (and obviously it isn’t part of the hardware of their vehicles)?
Tesla has chosen a different approach to autonomy than Waymo (and most other automakers in that area). If they’re not correct - if autonomy can’t be “solved” with just vision (or can’t be solved economically with just vision) - then they’re way behind Waymo and others in developing a LIDAR-oriented AI driver and/or the mapping necessary to support it.
To add more specifics on the word limited, for data, this means that Tesla has a very large training data set that is camera-only, and a much smaller contemporary data set that is camera with lidar.
I wouldn’t criticize Tesla for combining sensor data (eg, camera plus lidar) for some level of validation. Seems to make sense to me, but I really wouldn’t know enough to opine on the best sensor configuration for any given use case. We’ll find out whenever we get some level of apples-to-apples safety benchmarking.
Which one could win by being the bigger chicken?
On a recent CNBC interview, Waymo’s co-CEO allowed for an eventual them not using LiDAR:
Start about 19:45 in.
Her answer is “Not yet” to the question of “a world in which you don’t need LiDAR.”
Level 4 is certainly good enough - depending on the areas incorporated into that. And so the race is on to see how fast the L4 geo-fenced areas expand.
This will be an interesting area to watch. Waymo started driverless rides in Dec 2020, so it’s been 4.5 years to get to what I believe is only 5 cities: Phoenix, SF, LA, Austin and recently Atlanta. Their plans are for 3 more cities: Toyko, Wash DC, and Miami in 2026. That’s pretty slow.
If Tesla is correct that their approach needs between none and just light pre-mapping, and with Tesla’s ability to manufacture autonomous cars (just plain Model Ys that Tesla makes at the rate of a hundred thousand a month) they may indeed scale faster.
Waymo has no apparent scale advantages. They just bought Magna Stenyer’s assembly line for the iPace and are bringing from Europe to the US, with the eventual goal of making 2,000 vehicles in 20 months. Even at full capacity, its only tens of thousands of vehicles a month - assuming Waymo still wants iPaces in 2028.
To the question of whether she can even see a world in which LIDAR isn’t necessary.
Her point was that when you go far enough down the road of technology, you might not need what’s necessary today - but you have to go through today’s step in order to reach that later world. You couldn’t skip over the punchcard/magnetic card stage of computers, even though we didn’t end up using that tech forever.
It’s very possible - even likely - that even if vision-only AI drivers are impossible today that they will one day possible. But there can be a very, very long gap between today and then.
She talked about Waymo’s history of using LiDAR and how she wouldn’t have wanted to have started without it. That doesn’t mean Tesla was/is unable to start without it.
1 Like
I’m trying to figure out how you would use LIDAR data in the training if your goal is to not use it for actual use. How do you construct a dataset that has anything to do with it if you know that any data it provides cannot be used in the dataset?
I must be missing something.
When it comes to my investments, I ponder “what if my thesis is wrong?”.
I still have high confidence in Tesla/Musk’s ability to overcome any technological obstacles.
Those with low confidence in Tesla/Musk really shouldn’t be invested in TSLA.
So far, WRT TSLA, “key man risk” caused me to prune my position.
ralph
1 Like
Did Waymo run a demonstration of autonomous driving for ride hail service using stock cars (off a regular production line) 4 years ago???
Not only that, but lane keeping is also rapidly becoming a standard feature. That’s why Tesla calling it “autopilot” is absurd, it’s not autopilot in any sense of the word, it’s just fancy cruise control that keeps speed, distance, and lane, like most other modern cars out there can do.
You would think! But I’m still waiting for other cars to allow me to walk up to the car, open the door, get in and sit down, drive away, park, get out of the car, shut the door, and walk away. As far as I am aware, there are no cars like that, and that’s one of the easiest features to duplicate. The other Tesla feature that I adore, it that many people can share a car with ease, no searching for keys or anything like that. Just last week, one of our Teslas finished charging in my driveway at about 1am, and one of my kids that lives nearby needed it early the next morning, so at about 1:30am, I drove it over to their place, parked it in their driveway, and walked home. Next morning, they simply walked up to the car, go in, and drove to work. No muss, no fuss.
I bet the Chinese EVs have all the relevant features.
3 Likes
As slow as it is, that’s still much faster than Tesla, right? Robotaxis were included as part of the Master Plan Part Deux back in 2016 and in 2019 Musk said there would be robotaxis on the road by 2020. I mention that only because Musk views it as core part of the future business and has for a long time. This isn’t a new project they are just getting around to recently.
AFAIK, Telsa has not announced plans or even applied for AV ride hailing permits in any jurisdiction. We know the permitting process can be lengthy (except in Texas), so so it is unlikely there will be any expansion this year, and maybe not even next year.
So with Elon banging on the robotaxi drum all these years, why hasn’t Tesla started the permitting process? The only logical reason I can think of is they aren’t ready to start it.
The qualifier was if Waymo has a couple year lead. Which they seem to right now. The scale advantage is time. Waymo is increasing the number of paid rides exponentially, sometimes by double digits per month.
2 Likes
As of May 2025, Waymo’s fleet consists of over 1,500 vehicles across San Francisco, Los Angeles, Phoenix, and Austin, with plans to expand to 3,500 vehicles by 2026.
Tesla makes 35k cars a week.
Are they hiring 35K safety operators a week too?
2 Likes
The Tesla owners pay $ to Tesla for FSD.
I can only think of one reason (maybe there are others?), which is to correlate camera data to lidar data to get the aspects of camera data that provide the equivalent information as the lidar data (or, similarly, identify aspects of lidar data that cannot be found in camera data).
It would be like calibrating your measuring device (cameras) with a standard (lidar), somewhat like one would calibrate a scale with weights of known masses.
The guess is that cameras have all of the information required to make driving decisions. But, with current data processing techniques, including machine learning, we recognize that some sensor types are better at measuring certain features of the environment than other sensor types.
Use lidar to measure features of the environment for which cameras are not as good as lidar (3-d positional mapping? - again, based on what we know today in terms of how to use the sensor data), measure the same environment with cameras, correlate the details of the camera data with the lidar data, and then extract the camera data pieces that most highly correlate with the lidar pieces that you most care about (3-d positions?).
A key question is, will the much smaller camera plus lidar training set generalize enough?
Call me when you figure this all out across enough environment scenarios to meet safety benchmarks for real world driving.