I don’t quite understand why, for instance, Mercedes wouldn’t have a sufficient amount of data for autonomous driving development, or at least significantly less than Tesla. As I understand it, Mercedes has been developing autonomous driving for years, as have many other automakers. There are at least as many Mercedes, VW, BMW, and other brand vehicles on the road as there are Teslas, so it is at least possible to collect data.
Mercedes doesn’t have thousands of cars in city traffic driving themselves, filming traffic, and registering every driver intervention—and the intervention situation can be improved afterwards through simulation. Additionally, Mercedes (and other German manufacturers) certainly follow German privacy laws, meaning driving videos are the driver’s property, and all identifying information in the driving videos (license plates, people) would need to be anonymized before processing. For the Chinese, this is certainly easier, but their autonomous fleet isn’t very large either.
If the Mercedes CLA with LiDAR and NVIDIA software sells well in the US for a few years, it is certainly possible that they will have over a thousand cars recording autonomy data then. Tesla currently has hundreds of thousands of self-driving cars recording edge cases.
1 Like
All the major players have data for development—or rather, for struggling with this issue. You have to be involved in the “impossible” when everyone else is, just to maintain credibility in the eyes of the average joe.
But even the combined data of those players isn’t nearly enough. Nor is the computing power or anything else. A human’s neural network and its experiences are not easily replicable by a machine. Life experience beyond what is just collected from the road is required.
The grandmother who will drive a steering-wheel-less Tesla in Finland hasn’t even been born yet. It requires heavy regulation, infrastructure building, etc., and even then, the car won’t work on our cabin roads. In the city, when every pole has xxG, there might be some benefit. Cars will recognize the environment, it’s maintained, and vehicles communicate with each other. But by then, drone transporters will have already taken over the skies.
3 Likes
It would be nice if we didn’t rely on guesswork.
Or what is the basis for that claim that Mercedes might have over 1,000 cars?
And as I understand it, the correct way to talk about Tesla is that they are self-driving cars that a human monitors at all times.
So not self-driving, but an advanced cruise control.
8 Likes
I’ll comment on some of the topics under discussion here. I haven’t bothered to participate more actively in the debates because, in my opinion, there is quite strong tunnel vision on both sides. But here are some of these thoughts:
At least in aviation, autopilot is autopilot, even if it can’t perform all human tricks. Additionally, in fault situations, it can disengage and tell the human to fly the plane. In driving, L-levels are fine for defining the level of self-driving or lack thereof, but my own view is that if a car drives itself, then it drives itself (even if the boundary conditions are strict and a human supervises, etc., etc.). If it can’t drive itself, then a human drives it all the time. Even if a shitty autopilot only works in certain conditions and situations, it is still an autopilot nonetheless (both in aviation and in driving). Note, I am not taking a stand on the meaning of the acronym FSD, but on the fact that the car simply drives itself, even if only on the highway in summer weather.
Steering-wheel-less cars are unlikely to be in traffic in all weather conditions anytime soon, at least in Europe, at least not without a remote connection. In this case, too, I would compare the development to the development of aviation. After all, things like the Global Hawk fly in the sky among air traffic with the autopilot on, but it still has to be possible to control it remotely.
Then a comment on Tesla’s phantom braking. Last winter, I experienced phantom braking frequently in my American crude product (raakavalmiste), especially on narrow roads in winter conditions when a truck was approaching. This winter there hasn’t been any, even though conditions have been very challenging in places. So, at least for my part, phantom braking is no longer a valid fault but was already fixed by an update that arrived sometime during the snow-free season.
ps. So far, the worst and most dangerous driving assistant I have ever used in a car has been in a 2024 model year Skoda Superb. On a road with tar patching strips parallel to the lane, it identified them as lane lines and repeatedly tried to steer the car into the oncoming lane or the ditch. The first time, a head-on collision was narrowly avoided; the situation was that surprising. I wonder why, for example, Tekniikan maailma doesn’t mention anything about these, even though it’s a significantly more dangerous feature than, say, a car slowing down in its own lane (aka phantom braking) 
15 Likes
Well. That’s how it is. On the other hand, at least in my experience from being on a plane, there have been significantly fewer pedestrians, intersections, dog walkers, etc..
So in principle we’re talking about the same thing, and in principle a completely different thing.
But now we are waiting for when the safety driver is removed. And what the number of cars moving without a safety driver will be. Whether it’s 10/100/1,000/10,000/1,000,000, none of us know.
2 Likes
In aviation, the purpose of autopilot is not to replace human pilots; rather, it is one pilot assistance system among others. In road traffic, if autopilot remains just an assistance system, its financial potential is orders of magnitude smaller than if the human driver doesn’t have to monitor the driving at all. If I could buy a car in which I can focus on working while the car drives itself, its value to me would be roughly my hourly wage*commute time (+ the comfort value of leisure driving). If, on the other hand, I have to supervise the driving, I treat it like cruise control — a nice feature that I can pay something for, but definitely not hundreds of euros a year.
3 Likes
It’s true that in both aviation and maritime traffic, the autopilot plays more of an assistant role, as these systems essentially follow a pre-defined route quite blindly, don’t they? I mean that, aside from some potential minimum depths/altitudes etc., they steer the vehicle somewhat blindly to a pre-determined location, allowing the pilot/captain to even take a nap? In short, autopilots in aviation and maritime contexts execute defined commands, whereas the target level of autopilot in cars would also involve making decisions independently.
The difference between automotive FSD versus aviation and maritime autopilots is quite substantial, given that there isn’t much other traffic out there—and when there is, they tend to actively announce their presence. I understand that solo ocean sailors, for instance, use autopilot for a significant portion of the total time to allow for sleeping, eating, and so on. In these cases, while the captain is sleeping, the boat’s autopilot maintains the course based on GPS and heading, regardless of whether a tanker or some other obstacle appears in its path.
4 Likes
Apparently, the safety driver is staying after all… 
Waymo continues to operate without drivers, but supposedly that’s the “wrong way” to self-drive.
29 Likes
Traditionally, an autopilot refers to automatic control that aims to maintain a specific heading (and speed) by adjusting steering in response to prevailing conditions. There can also be a pre-defined route where automatic turns are made at waypoints along the way. The latter, however, isn’t quite so simple in sailing, for example. Indeed, a traditional autopilot doesn’t typically care about obstacles; they are instead picked up by radar, which can be set to trigger an alarm. In this sense, FSD is a quite different system, as it is given a multi-dimensional task to handle in its “own” way.
1 Like
Many writers here have also noted that the challenges only really begin when we start talking about those nines after the decimal point. “Reality has a super long tail of complexity”. It’s great that Elon finally realized it too and didn’t start force-pushing unsafe cars onto the streets without supervision! Driving coast-to-coast along wide and quiet highways is a much easier problem than driving in urban traffic without supervision without killing people.
4 Likes
It’s a shame that Lepikko has a posting penalty until March.
Can @Kasleew come and tell us what this means? And how far is that “can” being kicked now?
13 Likes
This is encouraging news. Now time is just starting to run out, at least regarding Tesla’s European sales; these updates definitely need to be brought over to this side of the pond as well.
Traditionally, autopilot is known from aircraft, and the first Atlantic crossing using only autopilot occurred in 1947. This included automated takeoff and landing in addition to cruising.
Somehow FSD, especially if it refers to a steering-wheel-less car driving alone anywhere, doesn’t seem like a thing for the very near future.
Passenger planes obviously have more people on board than a passenger car, but the first functional autopilot was introduced in 1912, and here we are 113 years later and pilotless planes haven’t taken over the market. Following this thread, I’ve gotten the impression that FSD isn’t “full self-driving” but rather an advanced cruise control, an aid.
\[https://en.wikipedia.org/wiki/Autopilot\](Autopilot - Wikipedia)
This FSD thing is already starting to resemble promises of fusion power plants; has it been 30 years away for 50 years now? Or even longer.
Musk’s habit, though, is to promise things more like 30 days or three months out.
Yeah, I’ll leave the thread alone in the future; I just can’t understand on what basis Tesla is or would be a good/fast-growing company or a good business. There’s plenty of promises, hype, plans, and rocketing share prices, and it could certainly be a much better investment than any of the stocks in my portfolio.
I think it was Buffett who said that you shouldn’t invest in a company whose business you don’t understand. I try to follow this at least to the point where I only invest in things I at least think I understand.
16 Likes
I’ll try, although it requires a bit of a step up from the notion that self-driving FSD in city traffic is just an advanced cruise control.
Fred Lambert is oversimplifying things with his own anti-Tesla bias when he claims Unsupervised FSD is the same thing as a Robotaxi without a safety person and counts miles toward Musk’s comment. Musk’s comment reflects his view on the “long tail” of autonomous driving, where simulation models alone, such as those announced by NVIDIA at CES, wouldn’t suffice for safe autonomy without massive fleet data that includes human intervention and experience.
A Robotaxi is not “Unsupervised.” It is teleoperated, at Tesla just like with others. Waymo also confirmed that a teleoperator was driving when an empty Waymo was once cruising on a golf course. Therefore, the safety driver can be removed before any specific mileage target is met. Another point is that Unsupervised FSD allows the driver to surf on their phone or eat yogurt without any teleoperation. More details will be shared either before or during the Q4 earnings call.
2 Likes
“Tesla investors need to ask themselves: If xAI owns the intelligence behind the robot, what exactly is Tesla owning?”
25 Likes
Yle reminds that China is going its own way in autonomous driving:
Last year, Musk gave investors advance information that Cybercab production would start in just a few months. Is the CEO once again telling a “modified truth”? 
15 Likes
Regarding this, we’ve seen video/images of the Cybercab production line testing, so in that respect, the timeline sounds realistic. It is a different matter when these will actually be allowed on the road and under what conditions. And that quote doesn’t really take a stance on that. They can certainly churn these out of the factory into inventory for a while if the software and/or regulatory approvals are still pending.
2 Likes
China is going its own way in everything else, too. Cities are already under surveillance.
3 Likes
This is an interesting question. Tesla has now introduced new Standard models, whose range relative to price is very competitive. In Europe and the US, there aren’t really other cars with over 500km WLTP range whose price is close to €42k (cf. corresponding US price). EVs in the same price range fall 100km short, and that’s why competition in the US is mainly with internal combustion engine (ICE) vehicles. In China, competition is fierce, which is why FSD approval would be important for Tesla. Chinese consumers value autonomy when purchasing a car, as the McKinsey study showed (link earlier).
By the way, lithium battery prices are still on a downward trend. The drop in prices depends a lot on what contracts and price mechanisms have been established. At the end of July, Tesla last bought $4.3 billion worth of battery cells from LG under a long-term contract, at a time when at least the spot price for lithium derivatives was low. The price war is far from over; Hyundai has cut EV prices in the US several times at the end of last year and early this year, by a total of nearly 20%, with leasing prices under $200/month.
Considering the expansions of Giga Berlin and other factories, I personally predict 1.7 million in sales for this year. As a numerical comparison, EVs account for only 11% of VW’s deliveries (and probably half of that in terms of profit — ICEs drive the numbers), while Tesla’s energy business accounts for 23% of its profit. In the future, the share of car sales in profit will decrease even further.
2 Likes