I clicked on the link thinking about this issue and found it at the bottom of the article.
Currently it's showing one critical disengagement every 206 city miles.
Maybe they could call it Tesla Robocrash?
Some differences compared to Waymo:
- Waymo has / can use more on-board compute, from [0] "It has also been revealed that Waymo is using around four NVIDIA H100 GPUSs at a unit price of 10,000 dollars per vehicle to cover the necessary computing requirements."
- Waymo uses remote operators. This includes humans but can also have remote compute.
- Waymo's neural network model can be trained / overfit on specific route or area. FSD uses the same model everywhere.
- Waymo's on-board hardware can use more energy, because it's possible to charge the battery between trips.
- Robotaxi services charge customers per mile, so it makes sense to run longer routes which are also easier to drive, i.e. the routing algorithm can be tuned to avoid challenging routes. This would be possible to implement on FSD too, but it seems that FSD drives fastest route.
[0] https://thelastdriverlicenseholder.com/2024/10/27/waymos-5-6...
We have an interstate highway system that's fairly well-maintained and understood, and is a finite space to map. Hypertrain on that, and you can offer an experience of 10 minutes hands-on-wheel at the start and end of the journey, and 3 hours of doomscrolling in the driver's seat. The highway miles are the most boring, both from a surprise-hazard standpoint and from a driver's-attention standpoint (there's nothing cool or interesting to see except the trunk lid of the car in front of you)
It offers a nationwide level of service that Waymo's city-by-city rollout lacks, and the chance for route-specific hueristics that Tesla's cameras-and-local-compute might miss.
I don't think that cars will have SOTA level LLMs running locally for a long time, and it seems that they actually need that kind of intelligence for full autonomy. However, it might also be totally fine if the passenger makes the difficult high-level decisions through a voice interface.
The vehicle can ask human remote operators for recommendations or clarification, but the vehicle itself decides whether to use them. Most of the time, the vehicle doesn't end up needing it.
The system though provides a way to let humans create training data for edge cases.
Tesla has yet to get good enough to achieve level 4 so they can't actually run a robotaxi yet. Tesla's bet is that if they can reach level 4 with their approach, they'll be able to roll out robotaxis much more widely than Waymo can.
So far, the bet has not paid off and Tesla needs it to pay off before Waymo's slower rollout gets too far ahead.
https://www.youtube.com/watch?v=AQir90MktRc
Just because it's supervised doesn't mean its not self driving
https://maps.app.goo.gl/Q3VPJvJ6WwXe3gdZ7
Four lanes, left to right: straight+left, right turn only, concrete divider, right turn only, right turn only. Note that there are only two lanes when you turn right, so you can turn into the rightmost lane from lane 2, the leftmost lane from lane 3, rightmost from lane 4.
Traffic lights (four signals) on the far end of the intersection work thus:
1. Left two lane lights turn green (Right two lanes are red). You can have traffic going straight, left or right. Traffic in lane 2 can turn right, but lanes 3 and 4 cannot. 2. Right two lane lights turn green, left red. Lane 2 cannot turn right but lanes 3 and 4 can.
All the lights are circles, no arrows. The only indication of weirdness is that there's a "No turn on red".
I do not see FSD behaving appropriately.
And they all do it better than Tesla. 0 fatalities. 0 injuries. 0 crashes where the self-driving was at fault (but a few where the car behind them crashed into them).
Glorified cruise control does not equal self driving. I know everyone has to hate on Tesla, but FSD in its current form is a decade ahead of these. It doesn't take long looking at videos of actual people using these systems to understand the massive capability gap between Teslas FSD and everyone elses driver assistance system.
FSD is still only level 2.
Honda and Mercedes are the only two companies to sell level 3 capable cars, but these are only level 3 under certain limited conditions.
You may be correct that the level 2 performance of FSD is ahead of the level 2 performance of any other car, but I don’t think we can call Tesla king until they also match the level 3 performance of these other cars under those conditions.
Of course, with GM bringing Cruise in-house and abandoning the taxi service, there's no telling how much of their technology will be used.
FSD in its current form is a decade ahead of these Yes, in terms of accidents and fatalities, FSR is way ahead of the entire rest of the industry. In terms of actual driving quality? Super Cruise and BMW have it beat. Yes, they're geographically limited. That's because GM and BMW are acting responsibly and making sure it works before they open it up everywhere. Move fast and break things doesn't work when the things being broken are people.
I got a video of my own: https://www.youtube.com/watch?v=mPUGh0qAqWA Goes into detail about why FSD crashes so frequently and why the problem cannot be fixed as long as FSD remains reliant on cameras. Indeed, if you type in "FSD accident" or any variation of "accident" there are thousands of videos, many of them taken by the Tesla owner themself.
Whatever you want to believe. Keep fighting the good fight.
He owned a company that merged with Confinity, which had already built a prototype of PayPal, registered trademarks, etc. at the time of merger.
He was made CEO. For four months. Which he spent trying to throw out the prototype written in Java because he only knew ASP.
Then the board fired him in absentia, the morning he left for his honeymoon. Not asked him to resign, not "focus on his family", but the moment he's gone, fired his ass.
Musk's contribution to the non-vaporware PayPal is "cashing the dividend checks".
Thiel and Levchin fired Musk, but they made up. Thiel bet on SpaceX later when it needed cash to reach its first successful launch.
How did a Schwarzenegger movie from 1990 do a better job of naming a robot taxi than Tesla?
Trying to figure out who invented first could be hard because your priority date was not necessarily when you actually thought of the invention. It was the latest date where you started working diligently to reduce your invention to practice and continued so working until you succeeded.
So if you came up with the idea and started right away working diligently on it and keep doing so until you succeeded then your priority date would be when you came up with the idea.
But if you took breaks you might lose that priority date, and your new priority date would be when you resumed work.
So then we have to decide when a break will reset your priority date. Is it just the length? Does the reason for the break matter?
And what counts as working diligently? Does it need to be full time or is it OK if you are working on your invention every evening after your job?
It was quite messy.
And then there’s the case of special configurations, e.g., car seats for those with young children, wheelchair access, etc. Even once FSD gets solved (if it ever does), these use cases also need to be accounted for as well.
But these are such easy issues to solve. Look at a society where public transport use is much much higher:
> All major train stations in Japan, and most minor stations, have coin lockers available to use. Coin lockers are a part of daily life in Japan, and they can be found also on bus terminals, shopping centers or department stores, some convenience stores and tourist attractions, and even on city streets.
For example I don't see how lockers solve the issue of multi-store shopping trips that was mentioned in the comment above. The locker would give you some place to accumulate the purchases from each store, but unless you could carry them all simultaneously you'd still need to make multiple trips between the locker and home to get everything home.
I suppose you could use public transit when going store to store accumulating things in the locker, then use your car to move everything from the locker to home in one trip. That's probably optimal. While going store to store you don't have to deal with finding parking at each store because you are taking public transit so might save time and money there, and you save time and money by getting your stuff home in one car trip instead of multiple transit trips.
Honest question: Why? They seem to be dragging up the rear in everything but unfulfilled promises.
No, not clear at all actually. Every day I see Teslas swerving dangerously while their "drivers" are using their phones instead of driving the car. When I used to commute past the SpaceX HQ in Hawthorne, I'd see at least one Tesla accident a day.
1. It compares "all drivers in all vehicles on all road conditions in all weather conditions" versus "the subset of miles driven in FSD conditions, where neither weather nor road conditions turn FSD off because it can't be sufficiently functional".
2. If your airbags don't deploy, Tesla doesn't consider it an accident for the purposes of that report (modern safety systems don't blindly deploy airbags, they evaluate g-forces, speeds, angles of impact, etc., so you can hit something at 25mph and the vehicle decides your seatbelts are sufficient. Tesla decides "that's not a reportable collision"). Know when else your airbags might not deploy? Very serious accidents, when hardware or controllers are damaged.
3. Speaking of which, fatalities are not included in that report. "It was a collision where someone died, but doesn't merit inclusion in a safety report" is a weird position to take.
We were working on e2e models for manipulation and motion for a long time at Google X and I was at Wayve working on e2e models for driving. From a science perspective, Tesla's FSD is very impressive accomplishment.
Edit: Sorry I think I now understand your comment. I mean FSD, not their factory robots. I don't know anything about Tesla's factories.