It's all semantics to me. It won't bring Tesla closer to a L3 system regardless. The fact that they have very little business incentive take on liability is just another data point that they won't be delivering such a system to existing cars.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
Knightshade
Well-Known Member
I think the fact another OEM has decided there IS business incentive to take on liability is significant though- currently it's an extremely limited ODD, but as it expands (assuming it does- MB has certainly said that it will, and unlike Audi and others who'd been promising L3 for years now they've actually delivered it) it'll get increasingly difficult for Tesla to justify staying where they are.
As to existing cars, I continue to believe Teslas current HW could deliver at least L3 on highways (and at higher speeds than MB does today)- I think it could even deliver L4 on highways but possibly with an ODD narrow enough to be problematic.
Yes I agree there is some external market pressure, but not significant yet. I was referring to the agreements made by Tesla for existing cars. They've already gotten payment for the product, so there is very little upside for Tesla to go down the L3 route there imho.
I disagree with this assessment. I don't think it's possible in the coming years based on my understanding of where computer vision is at right now.
What is your motive/agenda in joining this forum? Every single one of your 300 posts is specifically on the negative trolling of FSD and Tesla. Clearly you have never owned one nor made assessments based on actual personal use yet you feel compelled to be the technology expert. Are they paying you well or just bored.
I love this. Thanks for asking.
I am interested in robotics and computer vision in general just not autonomous vehicles.
I don't have to work office hours anymore, and need a few hobbies. So why not enlighten fellow Tesla owners in something I know a little something about?
Oh, and I think Elon is full of crap regarding FSD and that there is a massive echo chamber of useful idiots in the Tesla sphere, that doesn't do any research and just repeat whatever pseudo-scientific things Elon says. Like James Douma and Dr Know It All to name a few prominent persons.
Why don't you ask them why they are pumping? I am not invested in TSLA neither long or short.
I own a Model X (Raven) LR w/ FSD since 2019. Still waiting for delivery of the software in the EU. I have been interested in autonomous vehicles much longer, circa 2016. I have ridden and used several FSDb version in the US and in a Waymo. As in investor I have had a chance to ride in a few other self driving car projects both in academia and industry efforts.
I am interested in robotics and computer vision in general just not autonomous vehicles.
I don't have to work office hours anymore, and need a few hobbies. So why not enlighten fellow Tesla owners in something I know a little something about?
Oh, and I think Elon is full of crap regarding FSD and that there is a massive echo chamber of useful idiots in the Tesla sphere, that doesn't do any research and just repeat whatever pseudo-scientific things Elon says. Like James Douma and Dr Know It All to name a few prominent persons.
Why don't you ask them why they are pumping? I am not invested in TSLA neither long or short.
Last edited:
Certainly nobody can argue that Waymo are experts at not deploying at scale.
Level 3 or 4 (!) without radar, LiDAR… just vision only… ain’t going to happen… tesla would get their pants sued off …
Yup. No radar, no LiDAR, 1.2MP webcams until recently , no night vision capability … but Level 4 is just around the corner. Meanwhile as of now they can’t even offer hands free L2….
I got some time to finally listen to Jesse Levinson (co-founder/CTO Zoox) interview on the Robot Brains. He makes a few good points on why highway isn't necessarily easier than the city.
Bookmarked this discussion:
TLDW;
1. Sensor requirements (range and reliability) grows with speed because physics -> speed and weight -> breaking distance -> time to react.
2. False positives are challenging.
3. When things go wrong people die.
4. It's not acceptable to bring the car to a stop on a highway/freeway if the systems aren't nominal or if there is an issue. This is easyier and less dangerous in the city context. (not applicable to L3 with fallback obv)
Last edited:
Knightshade
Well-Known Member
Tesla still owes L4 to everyone who bought FSD before ~march 2019 though.
Can you expand on that? What, specific, capabilities do you think vision only can't handle as regards L3 highway driving?
I think the fact virtually every car company in the western world offers at least L2 on highways today, but 0 of them other than Tesla offer it in cities, suggests his points aren't all that good.
The "things go wrong people die" one is an especially bad take--- there's (typically) 0 pedestrians on highways.
diplomat33
Average guy who loves autonomous vehicles
Over 10 years, Waymo has tested in 25+ cities across the US. And from public info we have, Waymo is doing autonomous driving in SF, LA, Phoenix, Orlando and Austin at once. Now, if you are asking why they don't test in even more locations at once, I think the answer is that they don't need to. It would not give them any significant benefit. Waymo has lots of data already from the places that they are testing and when Waymo needs specific data, they simply go and get the data they need. For example, when they needed dense urban data, they went to NYC, when they needed rain data, they went to Bellevue, WA State and in Orlando, FL, when they needed snow data, they went to Michigan.
But I am not making up that Waymo is generalized. Waymo co-CEO said so:
Source:
So, the proof that Waymo is generalizing well is the fact that they were able to scale in PH and SF simultaneously and the fact that they launched the same Waymo Driver in LA and he says it performed incredibly well from the get-go. This implies that they did not need to do any overfitting to get it to work in LA. So I think this tweet directly refutes your claim that they need to overfit city by city.
The benefit in assigning 50 to a single city is that they can launch a ride-hailing service in that city. Waymo's whole business model is doing safe, reliable driverless robotaxis. Deploying 1 car in 50 cities would not speed up their research or speed up their scaling of a robotaxi service and it would take resources away from actually deploying a robotaxi service. It would basically just be a stunt. It would not help them in their overall mission.
This is not proof of poor generalization because Waymo was not planning to launch robotaxis in NYC. They were only in NYC to collect data, not to launch a new service. In fact, NY does not allow driverless so Waymo cannot start a robotaxi service in NYC even if they wanted to.
Just because Waymo tests in a city does not mean they plan to launch a ride-hailing service. Many times, Waymo tests in a city, just for data and validation. Unless Waymo actually announces a ride-hailing service like they did with LA, we should not assume that they planning to launch a ride-hailing service in that city.
Yeah, but that's peanuts. They're not going to deliver that, so I expect people will sue and Tesla will settle.
We're talking existing cars here. No radar, vision only (yes, some X/S have the new radar, and perhaps its good enough - I don't think so - and they may retrofit all 3/y).
1. Computer vision doesn't have the reliability levels needed at present. ML isn't safe enough to measure distance and drivable space. Physical measurement is needed at this point in time. It's both faster, doesn't rely on ML and has 100% recall. When there are few reference objects on the road such as at night CV performs worse than when there are more objects. It's just not safe enough right now afaik.
This is the main reason I don't think it will happen anytime soon. Here are some others:
2. I think Tesla's forward cameras are easily blinded because they are close together. In low sun conditions by oncoming traffic, or when entering/exiting tunnels for example.
3. The system seems to have too high latency for highway speeds. A Lidar based system would likely be able to break or react faster.
4. I think that sensor costs (hd-radar, lidar) will drop faster than CV will have break throughs, so it will be impossible at a point not to use more modalities. One would be criminally stupid not to, as it would up reliability by a few orders of magnitude.
5. There are quite a few single points of failures in the design at the moment.
6. UNECE R157 and probably L3 on the highway in general requires the use of external directional microphones for emergency vehicle detection and emergency corridor handling.
There are other road users though, and speed matters.
Last edited:
Seriously? You believe everything that comes out of Elon's mouth and excuse everything that Tesla FSD doesn't do and literally claim they can obviously do it if they wanted or that its not important. lol come on!
To you:
Tesla with human driver on highways = working
Waymo with human driver on highways = not working.
So when Waymo is testing in the 20+ cities you don't count those, but when its Tesla FSD with a safety driver you do.
Why single out western countries? What about the multiple car companies with door to door on city streets in China like Huawei? Tesla fans always love ignoring China when it doesn't favor them. Maybe because he's a CEO of a SDC and he has data and research that higher speeds means less room for error and recovery. For example one of the things he pointed out was sensor range. If you are doing 25mph and your sensor range is 100m, that would be sufficient for city. But if you are on the highway doing 80mph with a 100m sensor range, you are fked.
Another thing is general object detection and classification. Detecting pedestrians and classifying them in good weather is pretty much solved today. But what about a white bowling ball rolling on the highway or a flat tire while you're going 80 mph. The difference in confidence, and the precision you need in your detection, classification and your reaction is high compared to the city where you can slow down, process what's going on and then slowly react. In a high speed scenario, its completely different. You slow down inappropriately? You get a 20 car pileup.
Heck Tesla is another proof example. FSD Beta was only available on city streets for almost two years. Why didn't they start on Highway with FSD Beta and then go to city streets?
Last edited:
willow_hiller
Well-Known Member
Nope. I believe what I can test and experience on my own Tesla.
Believe it or not, the closest Waymo service to me is about 2,200 miles away.
Knightshade
Well-Known Member
Because I (and 99.9% of those on this board) don't live in China and can't buy a car that does that.
They CAN buy a Tesla- which is the only company selling in the far more litigious west that offers L2 on city streets.
If Huawei sold a good EV here with L2 city I'd absolutely consider them- but they don't.
But again, even in china far more companies offer L2 on highways than in cities, so even in THAT example it seems car companies find highways "easier" than cities.
Tesla fans are the only group of people who discuss technological superiority but exclude every other country because the argument doesn't favor them. Tesla is the best in the world but then exclude the world in their comparison.
Other countries have way more legal protections for consumers (EU) or more restriction on ADAS & AV (EU & China) than US.
Huawei is banned in the US.
Every car maker having a lane keeping system doesn't reduce the requirements and challenges an autonomous system on the highway has to meet.
Autonomous Driving on Highway is NOT lane keeping..
diplomat33
Average guy who loves autonomous vehicles
There is tons of info out there on Tesla and Waymo approach. You can watch Tesla's AI Day to get a look at the Tesla approach. You can watch Waymo academic talks to get a good look at their approach. No, we might not know every single detail since companies do keep some industry secrets but anyone who cares to dive deep can absolutely get a pretty decent understanding of the Tesla or Waymo approach. So I reject this notion that we just can't really know how they work. That just seems intellectually lazy to me.
To suggest that the only two approaches is to either drive only on accurate HD maps or drive with vision like humans is a gross mischaracterization. Companies who use HD maps, like Waymo, rely heavily on camera vision, lidar and radar to do real-time perception. They do not just navigate with HD maps.
It is also a gross oversimplification to say that there are only 2 broad approaches. There are many different approaches. Vision-only in modular components (perception, prediction, planning are separate), vision-only in end-to-end stack, vision+radar, vision+lidar, vision+lidar+radar (early sensor fusion) with HD maps, vision+lidar+radar (late sensor fusion) with HD maps.
I would also point out that there different types of maps for AVs. There are the basic nav maps, medium definition maps and HD maps. So it is not HD maps or nothing. In fact, all AVs use maps, even Tesla. It is just a matter of how detailed they are.
So it is wrong to say that the only approaches are either hyper accurate HD maps or no maps at all.
Not every company uses HD maps the same. Some might heavily rely on HD maps for certain cases. But others may use HD maps more lightly. I know Waymo has stated that they use HD maps as a prior and do not heavily rely on them. Waymo is designed to rely primarily on real-time perception and Waymo can drive even when the maps are wrong or missing.
diplomat33
Average guy who loves autonomous vehicles
That seems like a convenient excuse since you can easily dismiss anything from Waymo since you can't personally test it. But I think it is disingenuous. Do you not believe anything in the news since you did not personally witness it? Of course not! There are lots of things in life you believe because it comes from a reliable source, but you never personally experienced it. So why not Waymo? You can't believe the videos of people riding in a driverless Waymo unless you personally ride in one? You believe Tesla AI Day even though you were not personally in attendance but you can't believe a Waymo talk?
willow_hiller
Well-Known Member
It's not disbelief in general. It's disbelief in claims they make where it's impossible for anyone outside of the company to prove it, and where it would be advantageous for them to lie by omission.
They have every incentive to make custom versions of the Waymo driver firmware for every city. Why wouldn't they? There are no plans for a SF Waymo driver to ever drive to Phoenix or vice versa. And it's consistent with the amount of time it takes them to adapt the Waymo driver to new cities. If it was solely a matter of making a new map, they would have services in many more cities by now.
So it's entirety possible that when they say "it's the same driver" between cities, they're omitting important details about how they may be adjusting the firmware to suit specific cities.
Similar threads
