I guess my point was that if the car can do what I want it to do, where I want it to do it, I don't care what they call the classification.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
I guess my point was that if the car can do what I want it to do, where I want it to do it, I don't care what they call the classification.
diplomat33
Average guy who loves autonomous vehicles
Here is Tom Boyd's answer from the podcast about when he thinks Cruise will achieve full autonomy:
"I think I'm so close to it, and I’m thinking so fast about all of it that I don't trust my sense of that. I think that because I don't trust my sense, I fall back and look at what actually the combined journalism that's approaching the topic looks at and say like, “Okay. What is the scientist outside of me think?”
But I think it's near. I don't want to really talk about timing and dates that much, because I'd like to not make up my own answers that might overlap Cruise’s different answers. But it's really close. Before COVID, I was riding around in these cars every day and we had some rides where the car would drive to Chinatown, which is mainly pedestrian. It’s little tiny bit of room for cars to go through, then drive down Market Street and then go drive – And there are whole drives for the vehicle operators, the AVT didn't have to take over. The car just knew what to do. I just see the rapid amount of change Cruise is making as they just brute force bang their heads on all these problems and solve them. I think it's closer than people think, but I'm going to be careful. I don't really want to commit to what, but I can say is really important to Cruise. I want to say, if you don't mind, which is we will release our vehicles to drive themselves when they're safe. That's the number one thing. They have to be safe. The word we use internally is superhuman. We are working to statistically measure how to prove that our cars are a number of times safer than human drivers before we’ll let them on the road. You'll see it’s been driven by safety and that's going to be what lets us decide."
https://softwareengineeringdaily.com/wp-content/uploads/2020/06/SED1089-Cruise-Simulation.pdf
I like that he does not try to give a specific date. But he does seem to think Cruise is close to full autonomy which I assume means L5. I also like that the focus is on safety. I think we can assume from his answer that Cruise is waiting for their FSD to achieve a certain safety metric and then they will deploy L5 autonomous cars to the public. It's just a question of when Cruise achieves that safety metric.
"I think I'm so close to it, and I’m thinking so fast about all of it that I don't trust my sense of that. I think that because I don't trust my sense, I fall back and look at what actually the combined journalism that's approaching the topic looks at and say like, “Okay. What is the scientist outside of me think?”
But I think it's near. I don't want to really talk about timing and dates that much, because I'd like to not make up my own answers that might overlap Cruise’s different answers. But it's really close. Before COVID, I was riding around in these cars every day and we had some rides where the car would drive to Chinatown, which is mainly pedestrian. It’s little tiny bit of room for cars to go through, then drive down Market Street and then go drive – And there are whole drives for the vehicle operators, the AVT didn't have to take over. The car just knew what to do. I just see the rapid amount of change Cruise is making as they just brute force bang their heads on all these problems and solve them. I think it's closer than people think, but I'm going to be careful. I don't really want to commit to what, but I can say is really important to Cruise. I want to say, if you don't mind, which is we will release our vehicles to drive themselves when they're safe. That's the number one thing. They have to be safe. The word we use internally is superhuman. We are working to statistically measure how to prove that our cars are a number of times safer than human drivers before we’ll let them on the road. You'll see it’s been driven by safety and that's going to be what lets us decide."
https://softwareengineeringdaily.com/wp-content/uploads/2020/06/SED1089-Cruise-Simulation.pdf
I like that he does not try to give a specific date. But he does seem to think Cruise is close to full autonomy which I assume means L5. I also like that the focus is on safety. I think we can assume from his answer that Cruise is waiting for their FSD to achieve a certain safety metric and then they will deploy L5 autonomous cars to the public. It's just a question of when Cruise achieves that safety metric.
Last edited:
Soda Popinski
Member
From the February Karpathy talk, it seemed like they're hoping to have planning also NN based at some point in the future based on telemetry from Tesla owner's driving behaviors.
diplomat33
Average guy who loves autonomous vehicles
Thanks. That is my impression as well. I was just wondering how much of it have they done so far. Have they started any work on it yet?
The reason I ask is because it would be another data point to measure Tesla's FSD progress. If Tesla has started some work on planning, that would be a good sign.
Soda Popinski
Member
I wouldn't be surprised if they implement this on the next next re-write, or along with HW4 or something. I can see a very cautious Feature Complete FSD that will drive like a granny. Similar to the way Nav on AP has all the functions for highway FSD, but really doesn't cut it in Los Angeles traffic.
I'm disappointed that GM, owner of Cruise and a company I deeply distrust and dislike based on decades of bad corporate citizenship, seems to be ahead of Tesla on this. I'm glad that at least it seems that the Bolt would be the platform for their autonomous technology since I'd be seriously conflicted if the first fully autonomous car were a stinker.
If Tesla is first I'd probably only wait a month after introduction before buying a Level 4 car. If GM is first, it would take longer for me to trust the car.
All Teslas (except the Roadster, which was built on a rolling chassis not built by Tesla) have gotten five stars for safety in every category. Tesla has demonstrated that safety is its top concern. GM has demonstrated that profit is its top concern and it will make its cars only as safe as it must to maximize profit. It will adjust quality control for maximum profit. I don't trust public statements coming out of a company whose slogan was once "What's good for GM is good for the country."
If Tesla is first I'd probably only wait a month after introduction before buying a Level 4 car. If GM is first, it would take longer for me to trust the car.
All Teslas (except the Roadster, which was built on a rolling chassis not built by Tesla) have gotten five stars for safety in every category. Tesla has demonstrated that safety is its top concern. GM has demonstrated that profit is its top concern and it will make its cars only as safe as it must to maximize profit. It will adjust quality control for maximum profit. I don't trust public statements coming out of a company whose slogan was once "What's good for GM is good for the country."
Soda Popinski
Member
diplomat33
Average guy who loves autonomous vehicles
IMO, there is very little chance that Tesla will be first at full autonomy. Not that they can't eventually do it, just that they won't be first.
Why? Because Tesla is still doing perception and has apparently barely even started on planning. They recently released a beta traffic light response feature that requires driver confirmation and have yet to even release any basic "turning at intersections" feature.
On the other hand, companies like Cruise have advanced perception and advanced planning and have been testing L4 cars for years. And they have L4 cars that handle difficult driving scenarios like busy San Francisco driving in tight city streets for hours and hours with no driver intervention. Cruise is just waiting to increase safety to a certain x times better than humans and they will have full autonomy done.
For Tesla to be first, they would have to finish perception, do all the planning and validate it in every case, that it is so good that the driver never has to pay attention, not even in difficult city driving cases, and all in like the next 1-2 years. And Cruise would have to stumble and not be able to reach the needed safety in that time. That is unlikely.
By the way, in April, Cruise announced that they are the first autonomous car company to be 100% EV and their fleet in SF is 100% powered from renewable energy.
You can read more about it here:
Cruise Becomes First Self-Driving Company to Power Vehicles With 100% Renewable Energy
Last edited:
Hi
IMO, there is very little chance that Tesla will be first at full autonomy. Not that they can't eventually do it, just that they won't be first.
Why? Because Tesla is still doing perception and has apparently barely even started on planning. They recently released a beta traffic light response feature that requires driver confirmation and have yet to even release any basic "turning at intersections" feature.
On the other hand, companies like Cruise have advanced perception and advanced planning and have been testing L4 cars for years. And they have L4 cars that handle difficult driving scenarios like busy San Francisco driving in tight city streets for hours and hours with no driver intervention. Cruise is just waiting to increase safety to a certain x times better than humans and they will have full autonomy done.
For Tesla to be first, they would have to finish perception, do all the planning and validate it in every case, that it is so good that the driver never has to pay attention, not even in difficult city driving cases, and all in like the next 1-2 years. And Cruise would have to stumble and not be able to reach the needed safety in that time. That is unlikely.
By the way, in April, Cruise announced that they are the first autonomous car company to be 100% EV and their fleet in SF is 100% powered from renewable energy.
You can read more about it here:
Cruise Becomes First Self-Driving Company to Power Vehicles With 100% Renewable Energy
I'm not sure it's fair to compare the Cruise test cars with Tesla's consumer cars. I may be out of date, but I don't think you can buy a car even as advanced as my EAP-equipped Model 3 from any other car company. And to the extent that the FSD package is now marginally beyond EAP, it's the only thing you can buy that's better than what I have.
Two separate questions:
1. Who will be first to Level 4 or Level 5? We agree that it probably won't be Tesla, though I want it to be Tesla.
2. When will an ordinary consumer be able to buy a car with self-driving features more advanced than you can buy at that time from Tesla?
What Cruise (and Waymo, and MobilEye) is (are) doing is impressive and encouraging. But what matters most to me is, When can I buy it? I'm fascinated by what those test cars are doing. But it doesn't help me until I can buy one. Tesla is behind, but we're driving what they have.
And how can Cruise claim to be the first all-electric autonomous-car company when Tesla has always been all-electric? Unless they're claiming that being further ahead than Tesla means that Tesla doesn't count.
diplomat33
Average guy who loves autonomous vehicles
First of all, achieving L4 and achieving L5 are quite different in terms of difficulty. It's like saying who will win a 100 m race and who will win a 3 km race. So I think they need to be separate questions. We also need to break down achieving L4 versus deploying L4 to the public versus selling L4 to the public because those are very different milestones as well.
Who will achieve L4 first?
Waymo was the first to achieve L4. They were the first to demonstrate L4 autonomy in a car on public roads. I think they did that back in 2015 if memory serves.
Who will deploy L4 en masse?
Waymo has deployed L4 robotaxis that the public can use but it's very limited. So Waymo has partially achieved this goal already. And Cruise has plans, I think next year, to deploy robotaxis in SF that the public can use. They even unveiled the production version of the Origin robotaxi earlier this year. So I suspect either Cruise or Waymo will be first to deploy L4 robotaxis in larger numbers, probably in 2021 or 2022 if I had to guess.
Who will sell L4 first?
This is a hard one because it depends a lot on the business models of companies. Cruise or Waymo could sell their L4 to an automaker once it is reliable enough. It's hard to say who they might sell it to. Of course, Cruise could give it to GM which could put it in consumer cars.
Who will achieve L5 first?
If I had to guess, I would say Cruise or Waymo. They are the current leaders in FSD. They already have L4 that is great on highways and city streets. They just need to improve it more until it can handle every case super reliably and then it will be L5.
Hard to say. It depends on what Tesla and other auto makers do. But probably 2021 or 2022. I know there are automakers working on L3, set to be released in 2021 if memory serves.
I think 2021 or 2022 will probably be the year that someone has L3 or above that is ready to go into consumer cars. But keep in mind that just because an auto maker has L3, does not necessarily mean they will actually put in a consumer car. I think Audi said that they had L3 for Europe at one point but then at the last minute, cancelled it, claiming something about regulators blocking it. So I think your question is more a business question than a FSD question.
It is worth noting that many companies have really good L4 now. The big thing that is blocking automakers from putting it in consumer cars is that it is not deemed safe enough. Auto makers want their autonomous driving to be many times safer than humans before they put it in a consumer car. So achieving that "x times better than humans" safety rating is the tricky part, depending on what value you assign to "x".
Because Tesla cars are not autonomous cars. Sure, Elon claims the hardware is capable of full autonomous driving but the software is not autonomous yet. Teslas require that the driver monitor the environment and provide input so they are not autonomous. Cruise cars are fully autonomous since the cars actually drive completely on their own with no driver input. So Tesla meets the first criteria of being 100% electric but does not meet the other criteria of being autonomous. Only Cruise meets both criteria at the same time, being 100% electric and 100% autonomous.
powertoold
Active Member
I'm confident that none of these companies are anywhere close to achieving a profitable ride-sharing FSD service.
It's simple: 99.999% vision (or whatever 9s you like) needs to be achieved in order to achieve FSD. There's no way around it. Lidar won't help with a paper bag on the road. It also won't help with the stop sign corner cases documented in Karpathy's talk (or emergency vehicles or human traffic controllers). List goes on. No one can say FSD anything until they can show they have the vision achieved.
Again again again: if we are to believe that vision is an absolute necessity for FSD. Tesla is still ahead.
It's simple: 99.999% vision (or whatever 9s you like) needs to be achieved in order to achieve FSD. There's no way around it. Lidar won't help with a paper bag on the road. It also won't help with the stop sign corner cases documented in Karpathy's talk (or emergency vehicles or human traffic controllers). List goes on. No one can say FSD anything until they can show they have the vision achieved.
Again again again: if we are to believe that vision is an absolute necessity for FSD. Tesla is still ahead.
Soda Popinski
Member
I would argue bigger reasons companies won't sell it in consumer cars is because it would be prohibitively expensive and only be useful in a very limited market, due to being geofenced to HD mapped places. Maybe with the exception of MobilEye if they went camera only for cost, but that still limits use to Jerusalem.
diplomat33
Average guy who loves autonomous vehicles
No, Tesla is not ahead on vision either. Cruise, Waymo, Mobileye have better vision. Watch their presentations. They have vision that can see if a door on a parked car is open (I've not seen Tesla Vision do that), emergency vehicles and more. And they have pseudo-lidar already, something that Tesla seems to have just now discovered. The fact is that Cruise, Waymo etc are at the fore front of machine learning methods, are experts in machine learning, and have been working on camera vision for years. They probably have the best camera vision possible today.
Just to backup my case, here is a 14 mn clip from a Mobileye presentation from 2019 where they show the deep neural networks that they have for camera vision and you can see how advanced it is:
But you do realize that "solving vision" is not enough, right? Even if you do "solve vision", the car still needs to figure what to do with that vision. Seeing the environment is great and necessary but the car still has to figure out how to drive in that environment. And that's not trivial. You don't want your car to drive in an unsafe way, or be uncomfortable for passengers or annoying to other cars on the road. There are a ton of complicated city driving scenarios where you need to teach the car how to handle in a smooth and safe way. So even after "solving vision", you still have a lot of work to do.
No, it's not a small thing. That's why they have not deployed their FSD yet to the general public. They still have all kinds of driving scenarios that they still need to work out.
Cost is already starting to come down. This Lidar Is So Cheap It Could Make Self-Driving a Reality
HD mapping is also becoming easier. Nvidia has scalable, end-to-end mapping: End-to-end HD Mapping for Self-Driving Cars from NVIDIA Automotive
And Toyota is developing a way to build and update HD maps with satellite imagery and dash cams in cars that is much faster and cheaper than the old way: Toyota and its Partners Are Building Maps for Autonomous Vehicles Using Satellite Images & Dash Cam Footage
Mobileye is not limited to just Jerusalem. They are currently crowd sourcing from cars equipped with cameras and their Mobileye chips to build HD maps from all over the place:
FloridaJohn
Member
To me, this sounds a lot like an Elon Musk quote. Sure, he has a lot more disclaimers than Elon would have, but at the end of the day, he is claiming that "full self driving" is right around the corner and is pretty much done.
It wouldn't surprise me if Cruise is ahead of Tesla, but I think in this field, you get to a point where things are going swimmingly, and you identify problems and solve them pretty quickly, and all seems to be going according to plan. Then you start testing in other environments and all of a sudden you realize that you have merely optimized your system for the environment around the engineers working on the problem. Then the corner cases become obvious, and then there are more than you thought, and then you extrapolate out based on previous experience of quickly solving those problems, and then you start missing deadlines, and then you have a bunch of people on internet forums who have no idea what they are talking about acting as if they know how hard it was to get something so simple to work, and people are now demanding you to live up to your earlier predictions. Oh how naive you were back then!
diplomat33
Average guy who loves autonomous vehicles
To me, this sounds a lot like an Elon Musk quote. Sure, he has a lot more disclaimers than Elon would have, but at the end of the day, he is claiming that "full self driving" is right around the corner and is pretty much done.
It wouldn't surprise me if Cruise is ahead of Tesla, but I think in this field, you get to a point where things are going swimmingly, and you identify problems and solve them pretty quickly, and all seems to be going according to plan. Then you start testing in other environments and all of a sudden you realize that you have merely optimized your system for the environment around the engineers working on the problem. Then the corner cases become obvious, and then there are more than you thought, and then you extrapolate out based on previous experience of quickly solving those problems, and then you start missing deadlines, and then you have a bunch of people on internet forums who have no idea what they are talking about acting as if they know how hard it was to get something so simple to work, and people are now demanding you to live up to your earlier predictions. Oh how naive you were back then!
Sure, it is always possible that Cruise might encounter unexpected setbacks.
But there is a big difference between claiming you are close when you have something to show for it versus claiming you are close when you don't have much to show for it. It's like saying you are almost done writing a novel when you have a manuscript that is 90% done versus saying you are almost done when you've only written the first chapter.
Cruise has demonstrated L4 FSD that works in almost all cases. So it certainly is more plausible when Cruise says that FSD is almost done. In 2015, when Elon was saying that FSD would be "easy" and is basically "like a solved problem", Tesla had no FSD at all. And even now, Tesla has only released some preliminary FSD features. It is much less plausible to say that FSD is almost done when you are still in the early stages.
Last edited:
I am probably missing something in this conversation but what evidence do you have that cruise is l4?
It seems like Tesla has a marketing disadvantage against companies like Cruise. Cruise can claim a feature in a one off demonstration, whereas when Tesla release a feature it has to be reliable and safe because it's going to be tested religiously from the go by hundreds of thousands of customers.
That said it seems like Tesla are some way behind Waymo and probably behind cruise, but I wouldn't count them out yet, in the last year they have largely cracked computer vision, they have streamlined their training process, they have rewritten the code and had a hiring binge.
I think that Tesla are going to release new features fairly quickly from here on out and it's not that far from where they are now to full self driving(can drive anywhere but needs supervision.)
It seems like Tesla has a marketing disadvantage against companies like Cruise. Cruise can claim a feature in a one off demonstration, whereas when Tesla release a feature it has to be reliable and safe because it's going to be tested religiously from the go by hundreds of thousands of customers.
That said it seems like Tesla are some way behind Waymo and probably behind cruise, but I wouldn't count them out yet, in the last year they have largely cracked computer vision, they have streamlined their training process, they have rewritten the code and had a hiring binge.
I think that Tesla are going to release new features fairly quickly from here on out and it's not that far from where they are now to full self driving(can drive anywhere but needs supervision.)
diplomat33
Average guy who loves autonomous vehicles
Based on the definition of L4: Cruise has FSD that works in an entire city with the human driver rarely touching the controls.
The Cruise fleet did over 800,000 miles completely autonomously last year in SF with the safety driver only taking over about once every 10,000 miles. For example, Cruise cars do what you see in this video for thousands of miles with the safety driver never touching the steering wheel or pedals or stalk.
It's time compressed but what you see in this video is the car doing all the driving and making all the decisions for 1 hour:
This is not just a demo of a one off feature. This is what Cruise cars can do with NO driver intervention for thousands of miles.
Yes, I am sure that Tesla will release new features and that is awesome. I can't wait. But like you said, Tesla's FSD will require driver supervision at first, especially in tricky city driving. Cruise can already do FSD in very difficult city driving without driver supervision for thousands of miles. That's the difference. That is why they are ahead of Tesla.
Hopefully you see the difference between releasing a feature like traffic light response where the driver needs to tap the stalk to continue and the car being able to handle city driving for an hour with no driver supervision like in the Cruise video.
Now you might ask: if Cruise's FSD is so great, why don't they release it already? And the answer is because Cruise does not consider full self-driving for 10,000 miles without driver supervision to be good enough. They want better before they release it to the public. There are still some things that trip up Cruise and they need to improve that.
Last edited:
All excellent points.
The real problem with L4 as a category is that it really doesn't tell me what I need to know: Will it operate where I want to drive while I sleep in the back seat? So, yes, L4 and L5 are very different, but for my needs, and this is what I care most about, a Level 4 car that will operate on any road on Maui where I'm likely to want to go (never off-road, and never on the Road to Hana (I cannot go on that road because I get carsick)) is just as good as a Level 5 car. A car that is fully autonomous in San Francisco and does not require a driver there is of no use to me. Interesting, but does me no good.
Level 3 as a category has a similar problem: It's only useful to me if it operates where I want to drive and does not require me to take over excessively often.
diplomat33
Average guy who loves autonomous vehicles
Oh yes. I agree. Ultimately, what matters is whether the FSD will meet the needs of the owner. In the case of L4, the automaker will need to specify the ODD so that the owner will know if the L4 car is right for them.
I think the reason the SAE included L4 is because they needed an intermediary level. It makes to sense to have an intermediary level that represents "FSD but only in some cases" to cover systems where the FSD works but still has some limitations.
Same with L3. The SAE needed an intermediary level that represents systems that are sort of half between driver assist and autonomous driving.
Similar threads
- Article
