The sad thing is, since @strangecosmos has @Bladerskb (and many others like me) on ignore as per his own statement, he will never read that quote from SAE proving him absolutely wrong on doubting the Level 4ness of Waymo. I expect the misinformation to continue unfortunately and actual conversation to suffer.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla autopilot HW3
- Thread starter verygreen
- Start date
He actually does not as he has repeated some of my statements verbatim in other settings, including in a tweet to Amnon. He only pretends to put people on ignore in other to convince others to.
I need to stop taking people at their word.
Last time I did that I ended up with this thing called FSD.
But I will be merciful and end this with one post.
"It is incorrect to classify a level 4 design-intended ADS feature equipped on a test vehicle as level 2 simply because on-road testing requires a test driver to supervise the feature while engaged, and to intervene if necessary to maintain safe operation." - SAE Document
Waymo are Level 4 because its features are designed to be a L4 while Tesla is L2 because its features are designed to be a L2.
- You have no basis to say that Tesla is designed (long term) to be L2.
- You are ignoring that the SAE quote is for a "test vehicle". Not a consumer vehicle.
- If Tesla intends L4, then with a test driver on a test vehicle it would be L4.
I'll break it down for you. I said:
Then you come back with
See the contradiction now?
lunitiks
Cool James & Black Teacher
There’s so much noise and confusion in here about SAEs levels it’s unbearable.
J3016 is a taxonomy. It deals with the question: To which degree is the human expected to perform the driving tasks? And it lets us assign a Level from 0-5. "0" meaning the human is absolutely necessary in all respects, "5" meaning the human is not expected to do anything, anytime, anywhere.
J3016, chapter 8.1:
As for mr. Lex Friedman, he said that J3016 is “the most widely accepted taxonomization of autonomy”. And that it’s useful for describing the system, for legal discussion and for policy making. Which, of course, is right.
But then Friedman goes on to say that he thinks/feels J3016 is not useful for “design and engineering” because he thinks/feels there are/should be only two “real” categories – one where the system “is not fully responsible”, and one where the system “is fully responsible”.
Which is where he messes up.
First of all, responsibility is a legal matter - so it’s exactly what Friedman began saying J3016 is useful for describing.
Secondly, Friedman doesn’t seem to appreciate the fact that J3016 draws an unambiguous line between L0-L2 and L3-L5, where you are doing the driving in L0-L2, while you are not doing the driving in L3-L5. Because your car manufacturer has designed the system such that you're not supposed to be driving.
Which is why your car manufacturer must take legal responsibility for the car's actions in L3-L5. The manufacturer can’t tell you that you’re not supposed to be monitoring the car’s surroundings, and at the same time say you’re liable because you failed to monitor the car’s surroundings.
So J3016 is as much about manufacturer intent than anything else. Chapter 8.2:
So when Waymo publicly writes that their system is Level 4, it's because it's their design intent. It's what Waymo expects of it, and takes responsibility for.
Tesla on the other hand publicly makes it clear you must «keep your hands on the wheel at all times», that the car is not autonomous, and that YOU are responsible. So L2 is Tesla’s design intent with NoA, Auto Lane Change and Summon. They’re not intended to be, which is why they’re not, L3 features.
Get it? Good. Now wake me up when Tesla demonstrates anything L3.
J3016 is a taxonomy. It deals with the question: To which degree is the human expected to perform the driving tasks? And it lets us assign a Level from 0-5. "0" meaning the human is absolutely necessary in all respects, "5" meaning the human is not expected to do anything, anytime, anywhere.
J3016, chapter 8.1:
As for mr. Lex Friedman, he said that J3016 is “the most widely accepted taxonomization of autonomy”. And that it’s useful for describing the system, for legal discussion and for policy making. Which, of course, is right.
But then Friedman goes on to say that he thinks/feels J3016 is not useful for “design and engineering” because he thinks/feels there are/should be only two “real” categories – one where the system “is not fully responsible”, and one where the system “is fully responsible”.
Which is where he messes up.
First of all, responsibility is a legal matter - so it’s exactly what Friedman began saying J3016 is useful for describing.
Secondly, Friedman doesn’t seem to appreciate the fact that J3016 draws an unambiguous line between L0-L2 and L3-L5, where you are doing the driving in L0-L2, while you are not doing the driving in L3-L5. Because your car manufacturer has designed the system such that you're not supposed to be driving.
Which is why your car manufacturer must take legal responsibility for the car's actions in L3-L5. The manufacturer can’t tell you that you’re not supposed to be monitoring the car’s surroundings, and at the same time say you’re liable because you failed to monitor the car’s surroundings.
So J3016 is as much about manufacturer intent than anything else. Chapter 8.2:
So when Waymo publicly writes that their system is Level 4, it's because it's their design intent. It's what Waymo expects of it, and takes responsibility for.
Tesla on the other hand publicly makes it clear you must «keep your hands on the wheel at all times», that the car is not autonomous, and that YOU are responsible. So L2 is Tesla’s design intent with NoA, Auto Lane Change and Summon. They’re not intended to be, which is why they’re not, L3 features.
Get it? Good. Now wake me up when Tesla demonstrates anything L3.
It may their current implementation, but I don't think it is their long term "design intent".
I'm with strangecosmos, I think. I'd rather have a Tesla with the current Autopilot at whatever SAE level we want to call it than a level 4 NURO sidewalk delivery bot. Level 4 is not always better because it may come with limitations that degrade its value to you. Like the rail tracks of an autonomous airport shuttle.
Waymo's intent is to be level 4, and they look like they can come pretty close, but it's not a useful level 4 for us until it is actually approved for use by us for city or highway travel. So not quite level 4 yet. And so far unavailable to us. I'm not calling any self driving system level 4 until it proves itself in use. So far Waymo is useless to me.
All Tesla has to do is convince "regulators" that their FSD is some multiple safer than a human all by itself and they can call FSD L4. It may not have everything you would like in your autonomous car, but it might get my driving done. Even at whatever level it is now it's more useful than Waymo. And they're not done yet.
Waymo's intent is to be level 4, and they look like they can come pretty close, but it's not a useful level 4 for us until it is actually approved for use by us for city or highway travel. So not quite level 4 yet. And so far unavailable to us. I'm not calling any self driving system level 4 until it proves itself in use. So far Waymo is useless to me.
All Tesla has to do is convince "regulators" that their FSD is some multiple safer than a human all by itself and they can call FSD L4. It may not have everything you would like in your autonomous car, but it might get my driving done. Even at whatever level it is now it's more useful than Waymo. And they're not done yet.
I think people are failing to appreciate the different approaches that can be taken to developing a self-driving car.
The approach Waymo is taking is L4 over a limited geographical area. Even with the limited geographical area, and limited conditions it still hasn't been given the green light to drive truly autonomously. How much of that was due to the fatality accident involving Uber? How much of it was due to the car not interacting well with humans? Where humans have started to attack it. Waymo reveled a lot of challenges to L4 driving.
The approach Tesla is taking is an incremental approach where they're trying to grow a L2 system into a L3 system, and beyond. We don't know if they can get to L3. We also don't know if Tesla will assume the responsibility for an accident (while under L3) that other manufactures have pledged to do. Tesla can't leave it to an individuals automotive insurance because a lot of policies have changed to clarify that they won't cover the accident if the vehicle was in self-driving mode.
I'm pretty pessimistic about Tesla's approach, but that's mostly due to hardware the car doesn't have. Like it doesn't have Lidar or corner radars. It doesn't have things like Vehicle to Vehicle communication or infrastructure to vehicle communication. You might think this isn't important, but imagine intelligent stop lights that can tell cars what speed to go to hit greens through one after the other. That would be amazing even without autonomous driving. We already have around 4300 of those stop lights in the US so one would think Tesla would add hardware to work with them, but they haven't. Hopefully with time Tesla will learn from experiences, and make the required changes at the hardware level.
With all that being said the biggest obstacle to self-driving cars isn't the approach. It doesn't matter if you go bottom up or top down.
The biggest problem is standardization, and regulatory at the federal level.
We need to have freely available highly detailed maps that are updated constantly by all the players in the automotive, and road construction world. Where they are 3d maps (gotta think about drones too)
We need to have intelligent stop lights.
We need to improve our infrastructure. There are lots of areas where humans can't successfully figure out what to do.
We need to have federal standards that the states adopt. It can't be a state by state approval with different requirements in different states.
Audi has a limited L3 driving A8 that they won't even turn on it's L3 capabilities in the states because there is no clear regulatory pathway to do so.
If we don't do anything soon we might end up having to visit China to experience truly autonomous driving.
Why?
Cause even you disagree with their political policies you have to give them credit for getting stuff done.
They can make the roads smarter
They can force the cars to be compliant with whatever standard they want
They can force the citizens not to bully the robocars
In the US we have to teach our cars to out bully the bullies. Good luck with your self-driving car if it can't do that.
The approach Waymo is taking is L4 over a limited geographical area. Even with the limited geographical area, and limited conditions it still hasn't been given the green light to drive truly autonomously. How much of that was due to the fatality accident involving Uber? How much of it was due to the car not interacting well with humans? Where humans have started to attack it. Waymo reveled a lot of challenges to L4 driving.
The approach Tesla is taking is an incremental approach where they're trying to grow a L2 system into a L3 system, and beyond. We don't know if they can get to L3. We also don't know if Tesla will assume the responsibility for an accident (while under L3) that other manufactures have pledged to do. Tesla can't leave it to an individuals automotive insurance because a lot of policies have changed to clarify that they won't cover the accident if the vehicle was in self-driving mode.
I'm pretty pessimistic about Tesla's approach, but that's mostly due to hardware the car doesn't have. Like it doesn't have Lidar or corner radars. It doesn't have things like Vehicle to Vehicle communication or infrastructure to vehicle communication. You might think this isn't important, but imagine intelligent stop lights that can tell cars what speed to go to hit greens through one after the other. That would be amazing even without autonomous driving. We already have around 4300 of those stop lights in the US so one would think Tesla would add hardware to work with them, but they haven't. Hopefully with time Tesla will learn from experiences, and make the required changes at the hardware level.
With all that being said the biggest obstacle to self-driving cars isn't the approach. It doesn't matter if you go bottom up or top down.
The biggest problem is standardization, and regulatory at the federal level.
We need to have freely available highly detailed maps that are updated constantly by all the players in the automotive, and road construction world. Where they are 3d maps (gotta think about drones too)
We need to have intelligent stop lights.
We need to improve our infrastructure. There are lots of areas where humans can't successfully figure out what to do.
We need to have federal standards that the states adopt. It can't be a state by state approval with different requirements in different states.
Audi has a limited L3 driving A8 that they won't even turn on it's L3 capabilities in the states because there is no clear regulatory pathway to do so.
If we don't do anything soon we might end up having to visit China to experience truly autonomous driving.
Why?
Cause even you disagree with their political policies you have to give them credit for getting stuff done.
They can make the roads smarter
They can force the cars to be compliant with whatever standard they want
They can force the citizens not to bully the robocars
In the US we have to teach our cars to out bully the bullies. Good luck with your self-driving car if it can't do that.
Last edited:
Frankly, at this stage, we have no reason to believe Tesla is trying to grow an L2 system into al L3 system. Clearly we can forget about all the talk of Level 5 capable hardware, Tesla has stopped talking of Tesla Network (details to be released in 2017), Tesla has no known public autonomous driving data anywhere, and are on record as saying they won’t take responsibility for crashes with their self-driving cars (beyond bugs). It also does not seem like a current design intent on the roads of California where non-ADAS miles need to be reported.
Also, by all accounts Tesla’s Level 2 system is so far from autonomous level in reliability and range of features that all that is moot anyway.
What I would say would be a more realistic assesment:
The approach Tesla is taking is an incremental approach where they're trying to grow an L2 system.
All this talk of Levels 3 and above with Tesla are basically just wishful thinking — at least based on any data we can see. It seems at best more aspirational than anything like a current design intent.
As said, an advancing Level 2 system is still interesting for us consumers. But let’s call a spade a spade. Tesla currently is developing a Level 2 system.
What you are talking about basically already has a name: Level 5.
It is not about what anyone ”feels” is useful. It is about when the driver is not responsible for the drive. If never, it is Level 5. If never in limited scenarios, it is Level 4. If not for limited timespans, it is Level 3. If the driver is always responsible, it is Level 2 (or lower).
Your mistake is that you are inventing meanings for the Level 4. You can of course have any levels in your mind as you want, but SAE who defined the Levels specifically requires things that are different from your message.
The point with Level 3 and above is car-responsible driving. People do not seem to grasp how immense a change this is both from developer and user perspectives. Beyond that basic requirement, Levels 3-4 allow the car maker to set limits on when that car-responsible driving is available as both Waymo and Audi have done respectively in the most famous examples.
This is false. All Tesla has to do is designate their FSD as Level 4 and outline the geographical, speed and other potential limits to that — and then take responsibility for an important statement: driver is not responsible within those limits. (Certainly in some jurisdictions regulatory approval might be needed as well, but in fact in many jurisdictions Tesla could be doing this already.)
Of course they are not doing it, because they can not say under any circumstances: driver is not responsible, because they can’t handle that liability with their current technology. (In fact they are not even testing Levels 3+ in California their home state.) But this is fundamentally what the Levels are about. The confidence of the manufacturer and their stated guidance on how to use their features. They are not about what is useful, or even what is safe, they are about one thing and thing only: is the driver responsible for the drive or is the car.
Waymo is unambiguously Level 4. That they use safety drivers in testing and early deployment is allowed, as quoted, within SAE guidelines of what is Level 4. @strangecosmos is simply wrong.
Last edited:
If you have a look at the 2018 disengagement statistics UPDATE: Disengagement Reports 2018 – Final Results you will notice some companies have stopped to reporting their disengagements. Apple (because it was bashed because of being last ) now reformulated into reporting only "important" disengagements. Google likewise is said to encourage their safety drivers to disengage in known difficult situations, so the disengagement counts as a planned one and doesn't count towards the relevant statistic. Others moved their main testing to other states where reporting isn't necessary (Arizona, Michigan, ...). Tesla consider themselves to have found a loophole and argue they are only testing L2+ so they wouldn't need to report disengagements 
Obviously no-one could prove you are really testing level 4 because it all just come down to whether you trust the performance of your system enough that you are willing to take the responsibility for its behaviour.
Obviously no-one could prove you are really testing level 4 because it all just come down to whether you trust the performance of your system enough that you are willing to take the responsibility for its behaviour.The tech behind Falcon Wing Doors on my Tesla Model X is capable of doing much more, including flying, it chooses not to because we are waiting for regulators to approve!
Please, compare apples to apples. If we are comparing 'unreleased unavailable to customers' features, lets incluye the Roadster SpaceX package which will be cable to take off and fly small distances. Or the summon from the other side of the country. Elon is unbeatable on unreleased features.
Shock-On-T
⚡️⚡️⚡️⚡️⚡️
This isn’t speculation on HW3, it’s an argument about how various makers assign their technology to various categories.
For L3 I disagree because in terms of implementation it's an extension of a L2 system. An L3 system can be purposely limited in both where it can be used, and under what speed it can do. From a liability stand point I don't think Tesla's current view is workable, but I think that's why they want Tesla insurance. So if you get Tesla insurance you'll be able to use L3.
Now I still think to pull off L3 it will need the sensor hardware that's at least equivalent to what the Audi A8 has in Germany.
Tesla uses the "billion miles for validation before you can use it unsupervised" as a delay tactic. It's simply pushing the ball further down the road.
In short I think Tesla can pull off L3 in the next two years, but it won't be with the current sensor suite in the Model 3. If they do pull off L3 with the current sensor suite it will be in very limited weather conditions.
Keep in mind I'm only referring to undivided highways/freeways that's on-ramp to off-ramp.
Tesla also has another drawback in that they don't have a way of properly monitoring driver engagement. So Tesla will need to clear that up before they can activate L3 on any vehicle. An L3 vehicle can't work right if it can't give control back over to the driver. How do you give control back over if the driver is sleeping? An eye tracking solution like what Supercruise isn't that expensive.
L4 and above is where it gets tricky for everyone. I don't think the infrastructure is at a point where that's doable in any kind of real way. That's why Tesla likely shelved everything about the Tesla network, and snake chargers thing. They're incredibly far from having anything needed for L4.
I think the funnest thing about Tesla's approach is it causes the safety conscious people to lose their minds. Even an L3 system is kind of a bad idea (why some manufactures are skipping it). But, an advanced L2 system everywhere where the driver has to pay attention at all times? That's completely beyond the realm of anything that makes sense.
Despite my pessimism I hope it works.
L3 on the freeway is the only thing that can be even remotely safe. That's what NoA is intended for. Sure it sucks right now, but give it a few iterations. In my testing the worst part was the maps integration. It was buggy and didn't interpret the map right. It felt rushed, and undercooked. Keep in mind that the better it gets (while remaining an L2 system) the less safe it is. It's because when things work really well humans get overtrusting. We're likely going to go through phases of thinkings it's there, but then we'll get a wake up call that it isn't.
L2 on city streets is probably okay as the speeds are less, and people are likely going to have to take over constantly. To the point where they go "it seemed like a good idea at the time, but this is terrible".
The way I see it, is the journey to L3 with a Tesla with FSD is going to be an adventure.
The journey to L3 (or above) without it is going to be boring. I'd have to live my days from one tech demo to another.
For me I'm sticking with Tesla's approach even if it's completely insane way of going about it.
Last edited:
On the ARK podcast, Elon guessed at sleeping in your moving car in 2020ish. You may doubt the timing, but doesn't that indicate a level 4-5 system?
When did Tesla / Elon ever say level 5 anyway?
As has been stated before, running with nags gets around reporting requirements as does operating in state without such requirements. Your lack of data is not proof of anything.
Of course Tesla is not currently taking responsibility for crashes, the current system is speced to be driver supervised. That has zero bearing on what their design goal is.
Currently released versions always lag development in terms of features and performance...
Design intent is an aspiration, until it becomes design release...
(Yes I get bladerskb SAE quote nuance) They are developing toward effective L4. Testing means they are not there yet.
So if Tesla removed nags and used test drivers and test cars as defined by CA, you would call the current system L4? That is really all it would take for you to stop calling it L2?
lunitiks
Cool James & Black Teacher
When J3016 speaks of "design intent", it's not about aspirations. It's not about what lies in the future; what the manufacturer hopes to accomplish in a year or ten year.
It's about classifying current features. Current design. What is the human's intended role (if any), with the current system/feature?
It's about classifying current features. Current design. What is the human's intended role (if any), with the current system/feature?
I’m looking at Tesla
1) closing stores
2) taking away bonuses abruptly from the sales force
3) slashing some Model S/X prices by something like a third
4) while apparently raising money through AP/FSD price-cut sales lever
...and I’m thinking to myself, no way am I getting that free HW3 upgrade anytime soon. Such a major retrofit project would simply kill any benefit from this exercise and its scale likely just became immensely larger due to the FSD price-cuts.
In fact, for the first time ever, I have to put some (no matter how low) weight on the possibility that there will be no HW3 retrofit. And some even to the possibility that there will be no Tesla.
Mind you, I don’t believe in the last two. But I think only a fool would assume they are impossible looking at this on going mayhem.
1) closing stores
2) taking away bonuses abruptly from the sales force
3) slashing some Model S/X prices by something like a third
4) while apparently raising money through AP/FSD price-cut sales lever
...and I’m thinking to myself, no way am I getting that free HW3 upgrade anytime soon. Such a major retrofit project would simply kill any benefit from this exercise and its scale likely just became immensely larger due to the FSD price-cuts.
In fact, for the first time ever, I have to put some (no matter how low) weight on the possibility that there will be no HW3 retrofit. And some even to the possibility that there will be no Tesla.
Mind you, I don’t believe in the last two. But I think only a fool would assume they are impossible looking at this on going mayhem.
Last edited:
Psst. I said hw3 isn’t coming to us two weeks ago, like I said we are f-cked....
Similar threads
- Article
