I believe it takes the exit, that's my experience, and in the exit it will do the 3 tones down (in freq) indicating it's coming off NoA. That's a small distinction and they have improved it in the more recent updates. Also, I wouldn't ascribe SAE levels as Tesla doesn't. I believe it simply says it's dropping out of NoA but AS is still active.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: "Feature complete for full self driving this year"
- Thread starter dckiwi
- Start date
M3BlueGeorgia
Active Member
That's very "constructive" of you.
Tesla appear to be very hardcore Agile, including all the way down to the manufacturing floor. And that's a good thing, BTW
My take is that Elon reached back in time to dredge up a old-style benchmark purely for marketing reasons.
Daniel in SD
(supervised)
This post perfectly illustrates the problem with "driver assist" systems and Tesla's strategy. Imagine the system improves to the point where it can go 10k miles without needing driver intervention, what level of attentiveness will @wcorey have?
To be clear: Level 3 cars have to be able to self-drive in every eventuality, barring a loud breakdown of the vehicle, including when they are exiting their operational design domain (ODD).
To be Level 3, the car will never hand over control instantly nor will they ever stop driving in a controlled, car responsible fashion — other than a controlled stop, that is, or the car suddenly falls into pieces in such a way the driver is clearly alerted anyway.
If a Level 3 car is leaving its ODD, it will alert the driver and request taking over. But it will still continue to be responsible for the drive even if the driver does not react at all. Such a drive will end in a controlled stop eventually for it to be Level 3.
It is true this makes it a hard problem to solve. Frankly I think for many traditional automotive companies saying they’ll skip it for Level 4 is just buying time. They are nowhere near Level 4 either which is an even harder problem because it requires every autonomous drive to end up in a minimal risk condition or MRC (like the side of the road or a parking spot).
diplomat33
Average guy who loves autonomous vehicles
No, Level 3 system is Level 3 in some places (the ODD) and the car can of course be Level 2 in others (outside the ODD when autonomous drive is not operational). Level 4 is different in its end condition.
As for what Level 3 needs to know in advance: It needs to know that you have entered its operational design domain (ODD) and accept autonomus control. Then it becomes responsible for the drive and the driver can open the book, the emails or the TV.
After that the car must handle absolutely everything, except a physical breakdown of the vehicle so loud the driver is alerted. When I say everything, I do mean everything, including an elephant and a construction site appearing in the middle of the highway, lane marking disappearing and so forth. It can ask for the driver to take over but it can not relinquish control until that happens — so it can not fall back on it.
What it can do, and will eventually upon exiting its ODD or driver refusing to respond to a control request, is stop the car in a controlled fashion. This separates it from Level 4 where the end condition is different.
For NoA to be Level 4 it would have to have two things beyond removal of nags:
1) It would not require driver supervision at all and in fact the driver could be outside of the car or asleep. (On Level 3 driver is allowed to read but not be absent or asleep.)
2) It would always have to end its autonomous drive in a minimal risk condition (MRC), ie side of the road or parking spot. (On Level 3 a controlled stop in-lane would be allowed.)
Last edited:
@diplomat33 I really recommed you guys read SAE J3016 instead of these summaries from a press release or Wikipedia. There is nothing vague about the levels in the full document.
diplomat33
Average guy who loves autonomous vehicles
I am not sure about that. By definition, ODD is the entire area where the car is designed to be self-driving. If the car leaves it's ODD, it can't be self-driving anymore. The self-driving car should notify the driver to take over before it leaves it's ODD but it does not need to continue to self-drive outside it's ODD where, by definition, it is not designed to be self-driving. That makes no sense. If the car can continue to self-drive outside of its ODD, then it is not actually outside of it's ODD.
Sure. But did I say something wrong about the levels of autonomy? Just curious.
You are mistaken. The ODD is not and can not be a fixed space in geography or time because the world is always in flux. The car will have to be able to handle it all.
For example, the ODD for the Audi Traffic-jam pilot is a highway traffic-jam up to 60 kph with traffic on adjacent lanes of the car. This is its ODD where the Level 3 autonomous drive can be initiated in.
How would a car know beforehand when its ODD, as defined for example in this case, ends? It can not know that. It can only know if the conditions are met at the time of initiating the system.
If and when the Level 3 car accepts responsibility for the drive, it means accepting it until a driver takes control with sufficient warning (eg 10 seconds or other such time defined). If a driver does not take control, the car can and must make a controlled stop.
Of course the car can plan ahead such as initiating a request to take control before the highway system (if it is a highway Level 3 feature) is coming to an apparent end on the map, so it can stop before the road ends if need be. But it will need to be able to handle a situation where the road abruptly ends before that due to a construction site for example, until the driver takes control or a controlled stop is done.
Right there you just did, but overall I just think it is helpful to really understand the phenomenal and forward-looking work SAE has done. Any ambiguity in the Levels in my view usually comes from reading just summaries of them, instead of the real thing.
diplomat33
Average guy who loves autonomous vehicles
Yes and what does the car do when it is about to leave it's ODD (traffic starts to move and the speed is greater than 60 kph)? I am saying that the Audi will not continue to self-drive as L3 outside of it's ODD!
Yes it can very easily. When the speed is accelerating and is about to exceed 60 kph, I presume the Audi would notify the driver to take over or at least resume requiring the driver pay attention again (using eye monitoring camera to verify driver has eyes on road).
In the case of Tesla, NOA already alerts the driver when the car is about to reach the end of an off ramp. So, NOA is already capable of notifying the driver when it is about to leave it's ODD.
Obviously, I agree that it is always useful to go deeper rather than rely on summaries.
Yes it will — or it is not Level 3. It will continue to drive or come to a controlled stop. A Level 3 car never forces the control of the vehicle back to the driver (except if the car breaks down in a loud manner, say a major physical failure).
This is not how SAE Level 3 works. Yes, when the ODD is ending certainly the car will prompt for the driver to take over but there is no middle ground. Either the car is responsible for the drive or the driver has accepted control and is now responsible for the drive.
On Level 3 — and this is very important — the driver always (except if the car breaks down loudly) has the pre-defined amount of time to accept control but they also have the guarantee that if they won’t accept control, it will not be forced upon them.
On Level 3, the car has only three recourses once it has accepted responsibility for the drive: 1) driver accepts control back, 2) the car is brought to a controlled stop, 3) the car experiences a physical breakdown of such a nature that the human driver is bound to be alerted to it anyway (a loud force majeure type of situation).
Other than that the car must drive without requiring human supervision.
This feature has got very little to do with Level 3 given that NoA does not accept control of the car while in ODD — its design requires constant driver supervision, it can force the control back to the driver at any second and it is not designed to handle all eventualities within the ODD.
diplomat33
Average guy who loves autonomous vehicles
Not true. When a L3 car prompts the driver to take over, the driver MUST take over.
I think you might be confusing L3 with L4. If the car can self-drive everywhere and the driver does not need to take over even when prompted, that is L4.
By the way, here is a video of Audi's Traffic Jam Assist:
Check the 6:30 mn mark. Basically, the system just gives the driver a warning to hold the wheel again and bumps the driver out of Traffic Jam Assist. Traffic Jam Assist does NOT continue to operate outside of it's ODD.
@diplomat33
You are simply mistaken. When a Level 3 system prompts for control, the driver should take control, yes. But until they do the car remains in control until the driver accepts control or the car stops itself in a controlled manner.
Level 4 adds to this the requirement that the stop must happen in a minimal risk condition like the side of the road or other safe destination, Level 3 can stop in-lane.
As for that video, it is of Traffic-jam Assist, a Level 2 system. We are talking about Traffic-jam Pilot, the Level 3 system.
You are simply mistaken. When a Level 3 system prompts for control, the driver should take control, yes. But until they do the car remains in control until the driver accepts control or the car stops itself in a controlled manner.
Level 4 adds to this the requirement that the stop must happen in a minimal risk condition like the side of the road or other safe destination, Level 3 can stop in-lane.
As for that video, it is of Traffic-jam Assist, a Level 2 system. We are talking about Traffic-jam Pilot, the Level 3 system.
diplomat33
Average guy who loves autonomous vehicles
I think I see our miscommunication. You are talking about during the prompt to take over. Yes, absolutely, L3 must continue to self-drive until the driver takes control again. I was talking about AFTER the 10 second period to take over has expired. What I was saying is that the driver does need to take back control within 10 seconds of getting the prompt.
Here is a direct quote from page 7 of the SAE document:
"At level 3, an ADS is capable of continuing to perform the DDT for at least several seconds after providing the fallback-ready user with a request to intervene. The DDT fallback-ready user is then expected to achieve a minimal risk condition if s/he determines it to be necessary."
So yes, the L3 car will request the driver take over and give the driver enough time to take over. Yes, the L3 car will continue to self-drive until the driver has taken over. But the driver is expected to take over at some point.
Also, per your suggestion, I downloaded the full SAE document on levels of autonomy.
Here is an interesting bit from page 7:
NOA is clearly designed to be a L3+ feature but it not quite there yet. It can do lateral and longitudinal vehicle control, it can do most OEDR (Object & Event Detection & Response, ie monitoring and responding to the driving environment) and it can even do "destination and waypoint planning". It is only L2 right now because it can't do 100% of OEDR yet. In fact, that's probably why Tesla has the nags. But once NOA can handle all of OEDR then it will be L3+ according to the SAE definitions.
EVNow
Well-Known Member
It doesn't indicate a problem with current systems. It just shows the inadequacy of SAE levels.
The jump from 2 to 3 is too big. Real car companies (not waymo types) will slowly move from level 2 up, without declaring L3 for a long time (unless marketing types interpret L3 differently and claim no liability even with L3).
I think I see our miscommunication. You are talking about during the prompt to take over. Yes, absolutely, L3 must continue to self-drive until the driver takes control again. I was talking about AFTER the 10 second period to take over has expired. What I was saying is that the driver does need to take back control within 10 seconds of getting the prompt.
Also, per your suggestion, I downloaded the full SAE document on levels of autonomy.
No, you are still mistaken in the driver needing to take control after the 10 second period (or whatever time is defined). They don’t. They should, yes, but the car still can’t just give up even if they don’t. What the car can do, then, is make a controlled stop — at least in-lane. (Level 4, when leaving its ODD, must make a safe stop in a minimal risk condition and can’t stop in-lane.)
While Traffic-jam Pilot is a limited system regarding its rather slow and specific ODD, it is actually fairly clever in how it asks control of the driver and is thus a good example of what Level 3 means. This has been demoed to the press in Germany and all I describe works already in the system. Feature complete, shall we say.
When it is approaching the end of its ODD (or indeed has gone over it due to traffic conditions changing rapidly) the car pings the driver for control audiovisually. One end of the ODD would also be if it notices you falling asleep through driver monitoring. The driver then has 8-10 seconds — whatever the exact figure will be in production — to take control after being thus warned.
What happens if you don’t take control is the part where it really differentiates itself from Level 2. No, the control is not ”thrown at the driver” after that time forcibly. Instead Traffic-jam Pilot starts making more noise and pulling the driver’s seat belt in an effort to get them to accept control. All this time the car will continue driving and will continue being responsible for the drive. This process may well take longer than 10 seconds.
If the driver still doesn’t react, eventually the car assumes a medical emergency, turns on the hazards, slows down and stops in-lane in a controlled manner, turns on all lights, unlocks all doors and calls for help.
Assuming Audi can release this eventually as Level 3, this is how it should work in production too.
Daniel in SD
(supervised)
This isn't really about SAE levels, most consumers have no idea what SAE levels are. The issue is that users of level 2 systems are already starting to consciously exhibit "relaxed attentiveness" because they observe that the system is capable of controlling the vehicle in most common situations. This isn't an attack on @wcorey, it's simply human nature. Imagine if NoA gets so good that it requires user intervention only once a year for the average driver, will the average driver still be fully attentive? I'm skeptical.
EVNow
Well-Known Member
Lex Fridman's research says the opposite. You would have to provide evidence to prove him wrong - not just anecdotal observations.
diplomat33
Average guy who loves autonomous vehicles
No, you are still mistaken in the driver needing to take control after the 10 second period (or whatever time is defined). They don’t. They should, yes, but the car still can’t just give up even if they don’t. What the car can do, then, is make a controlled stop — at least in-lane. (Level 4, when leaving its ODD, must make a safe stop in a minimal risk condition and can’t stop in-lane.)
While Traffic-jam Pilot is a limited system regarding its rather slow and specific ODD, it is actually fairly clever in how it asks control of the driver and is thus a good example of what Level 3 means. This has been demoed to the press in Germany and all I describe works already in the system. Feature complete, shall we say.
When it is approaching the end of its ODD (or indeed has gone over it due to traffic conditions changing rapidly) the car pings the driver for control audiovisually. One end of the ODD would also be if it notices you falling asleep through driver monitoring. The driver then has 8-10 seconds — whatever the exact figure will be in production — to take control after being thus warned.
What happens if you don’t take control is the part where it really differentiates itself from Level 2. No, the control is not ”thrown at the driver” after that time forcibly. Instead Traffic-jam Pilot starts making more noise and pulling the driver’s seat belt in an effort to get them to accept control. All this time the car will continue driving and will continue being responsible for the drive. This process may well take longer than 10 seconds.
If the driver still doesn’t react, eventually the car assumes a medical emergency, turns on the hazards, slows down and stops in-lane in a controlled manner, turns on all lights, unlocks all doors and calls for help.
Assuming Audi can release this eventually as Level 3, this is how it should work in production too.
Thanks. That's helpful. It sounds like Audi did come up with a good solution for its L3 in cases when the driver is unresponsive.
SAE says:
"Thus, a level 3 ADS, which is capable of performing the entire DDT within its ODD, may not be capable of performing the DDT fallback in all situations that require it and thus will issue a request to intervene to the DDT fallback-ready user when necessary
At levels 4 and 5, the ADS must be capable of performing the DDT fallback and achieving a minimal risk condition. Level 4 and 5 ADS-equipped vehicles that are designed to also accommodate operation by a driver (whether conventional or remote) may allow a user to perform the DDT fallback if s/he chooses to do so. However, a level 4 or 5 ADS need not be designed to allow a user to perform DDT fallback and, indeed, may be designed to disallow it in order to reduce crash risk (see 8.9)."
So it seems like the big difference between L3 and L4 is that L3 will prompt the driver to take over but will pull over if the driver is unable to take over whereas a L4 car needs to be able to do the entire fallback itself and leave the driver out of the process entirely.
Daniel in SD
(supervised)
How do you explain the problems that companies that are actually testing much more capable autonomous vehicles are having in keeping their test drivers attentive?
There's already a huge number of posts of why your interpretation of Lex Fridman's research is flawed. Here's a selection of quotes from the paper contradicting your claim: What the chances Tesla cars will be self driving in 3 years? Why do you think that way?
Similar threads
