Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Autonomous Car Progress

This site may earn commission on affiliate links.
@powertoold is making a big deal about nothing. Remote assistance is not a big deal. You have to sometimes give the autonomous car a little hint because it can get stuck

Cool, so we've established that a full-time iPad guy is still allowed for level 4. Woohoo!

Every day, my mind is opening up to the smoke and mirrors created by all the other fsd developers.

It is so so obvious to me now who is actually ahead of the game.
 
Last edited:
  • Like
Reactions: mark95476
Holy cowww, what the heck are those two white lines doing? Is this a comedy?

We get to see *one* non-affiliated video about Mobileye, and it turns out to be hilarious.

SmartSelect_20210921-171745_YouTube.gif
 
Didn't take @powertoold long to come up with FUD.

If there was any disengagement on that drive (there wasn't based on the vid) it certainly wasn't the timeline he pointed out.

A cut doesn't equate to a disengagement when there's like 50 cuts in the entire vid which are acting as progression for the author's commentary.
The cut at 6:03 is to showcase the neural network output. The cut at 5:43 is to showcase the turn.

Lets take a close look at cut 5:43 for example. The claim that there was a disengagement there is completely unfounded.

  • Mobileye's system always pre-empt their turn by turning the wheel so the cut could have given the illusion that the car was attempting to drive infront of the upcoming cars.
  • But Mobileye's system had already marked those two upcoming cars as RED which means GIVE WAY. It would not drive infront of them because its giving them way.
  • The speedometer was 0 while the wheel was slowly turning anticipating those two cars passing.
  • The cut happened. Those two cars did pass and the car kept waiting for more cars to pass before eventually making the turn.
  • The drivers hands (which was making the okay hand signal) and feet which was slanted over NEVER moved before and after the cut. They stayed at the exact same spot on his jeans, with the same hand signal and his feet at the exact same spot. The driver couldn't have possibly taken over.
So no there was no disengagement based on those timelines.

All @powertoold has left is FUD after claiming multiple times that Tesla would have L5 safer than human drivers 6-9 months from October 2020 and then months later claiming by the end of the year.
 
Last edited:
Cool, so we've established that a full-time iPad guy is still allowed for level 4. Woohoo!

Every day, my mind is opening up to the smoke and mirrors created by all the other fsd developers.

It is so so obvious to me now who is actually ahead of the game.
It is just important to really understand that L4 is not necessarily better than L2. It depends on what you need it for. L5 is the holy grail, and everything below that is just different versions of trying to get to L5.

For instance, if you own your car, you probably love tesla's approach since you can use it anywhere you live(once it actually releases to the public). But you still have to pay attention. It would really suck to own a car in NY that only drives itself in Phoenix or SF.

But for robotaxis, all that matters is getting around the city you are in, so waymo/mobileeye are much better options.

Ultimately, competition is great, and it is fun seeing the different approaches to a similar problem. There is a ton of merit in all of them. As a tesla owner, I much prefer the tesla method because I can actually use it.

I would much rather a L2 everywhere over a L4 that is super limited.

And really, L2, L3, and L4 are all have enough similarities that you can "cheat" your way up the scale and deliver no actual usability if it is limited enough but still meets L4. For instance, someone could make a car that is fully L4 in a certain neighborhood in the US and it doesn't work anywhere else. Technically it is L4 in that area, but it is useless to most of the population.

I'd suggest stop worrying about what level is what and worry about the technology and what can the car actually do.
 
  • Like
Reactions: Bladerskb
My understanding of L4 is that if it cannot figure out what to do, it must be able to safely stop, in a safe spot, and can then ask a human to take over.

So if a robotaxi finds a safe place to stop and then queries a remote supervisor for instructions, it can be considered L4. However, if a remote supervisor must take over on the fly, or in any way intervene while the car is not stopped in a safe place, then it's not L4. If the car must ask for help while driving, or otherwise in traffic, then it's L3, not L4.

That said, I don't think it's terribly interesting to argue about whether Waymo or Mobileye is L3 or L4. Both are very impressive, not yet ready for wide-area deployment or consumer purchase, but could get there one day.

Tesla is wide-area but is L2 and nowhere near ready for even L3.
 
My understanding of L4 is that if it cannot figure out what to do, it must be able to safely stop, in a safe spot, and can then ask a human to take over.

So if a robotaxi finds a safe place to stop and then queries a remote supervisor for instructions, it can be considered L4. However, if a remote supervisor must take over on the fly, or in any way intervene while the car is not stopped in a safe place, then it's not L4. If the car must ask for help while driving, or otherwise in traffic, then it's L3, not L4.

Yes, L4 can safely pull over and ask a human to take over. But remember that remote supervisors never take over, they merely give extra information to the car and let the car decide what to do. The car is still in control and makes all the driving decisions. So that is also L4 since the remote supervisors do not take over in any way.

The car can ask for help while driving or in traffic and it is still L4 if the car stays in autonomous mode the whole time and the remote supervisor does not tell the car what to do or control the car.

Remember there is a big difference between actually taking over and just giving the car extra information.
 
My understanding of L5 is that it requires no human intervention ever. An L5 car can have driver controls, but it never needs them. So it could have iPad control, but if it ever needs someone to intervene via the iPad, it's not L5. (It only needs a human to give it a destination.)

Yup, that's why I said level 5 *can* have a full-time iPad guy. And it'd be okay if the developer of the level 5 feature *required* the iPad guy to be there all the time.

Basically everyone here won't understand what I'm saying lol.

The levels only define the software feature, not what the developer tells you to do or not.

If you disagree with me, go read the definition again. Look at how the definition talks about the roles and software features of a given level.
 
  • Disagree
Reactions: diplomat33
Yup, that's why I said level 5 can have a full-time iPad guy. And it'd be okay if the developer of the level 5 feature *required* the iPad guy to be there all the time.

Why are you still harping about the "full time ipad guy"?

And that is wrong. It would only still be L5 if the "full time ipad guy" just gave the car information and nothing more, and did not take over, take control, disengage, or tell the car what to do.
 
Yup, that's why I said level 5 *can* have a full-time iPad guy. And it'd be okay if the developer of the level 5 feature *required* the iPad guy to be there all the time.

Basically everyone here won't understand what I'm saying lol.

The levels only define the software feature, not what the developer tells you to do or not.

If you disagree with me, go read the definition again. Look at how the definition talks about the roles and software features of a given level.
The system you're proposing makes no sense. Why would someone design a system like that?
 
Holy coww, what the heck is FSD beta doing? Is this a comedy? Can't FSD beta see cones? Why does FSD Beta try to hit cones at the end?

Looks like it's trying to follow the original road, but the cones directed it to an opposing lane, so it was confused. Basically the existing lane structure overrode the cones for the pathing decision.
 
  • Like
Reactions: diplomat33
The system you're proposing makes no sense. Why would someone design a system like that?

The point was that we see all these "fully autonomous" drive videos from fsd developers, but in actuality, they could be having iPad guys making some of the decisions.

In a nutshell, the levels mean nothing to me anymore. Their definitions are well-understood by very few, and they don't help gauge progress or performance whatsoever, so what's the whole point of using them if what we really care about is how the fsd performs in the real world (without human decision making or interaction)?
 
I'm probably being optimistic, but it seems like ADAS put-up or shut-up is coming to a head. I imagine Tesla's very pubic FSDbeta roll-out and rapid iterations have been driving things along.

Every day, my mind is opening up to the smoke and mirrors created by all the other fsd developers.

It is so so obvious to me now who is actually ahead of the game.
 
  • Like
Reactions: powertoold
Yes, L4 can safely pull over and ask a human to take over. But remember that remote supervisors never take over, they merely give extra information to the car and let the car decide what to do. The car is still in control and makes all the driving decisions. So that is also L4 since the remote supervisors do not take over in any way.

The car can ask for help while driving or in traffic and it is still L4 if the car stays in autonomous mode the whole time and the remote supervisor does not tell the car what to do or control the car.

Remember there is a big difference between actually taking over and just giving the car extra information.

I acknowledge that you are more conversant than I am with the fine details of the SAE level definitions, but I feel that a human providing any input to the car while it is driving or otherwise not safely out of traffic goes against the whole concept of "autonomy," which is a machine doing a job on its own. This includes relying on its own mechanized sensing as well as calculations and decision-making. If it needs input from a human to operate safely and properly, then it's not autonomous. Level 4 is not full autonomy, but as I understand the spirit of L4, it differs from full autonomy only in that there are situations when it must find a safe place to park and only then can it ask for human assistance. If it needs human assistance on the go, then it's level 3.

Yup, that's why I said level 5 *can* have a full-time iPad guy. And it'd be okay if the developer of the level 5 feature *required* the iPad guy to be there all the time.

Basically everyone here won't understand what I'm saying lol.

The levels only define the software feature, not what the developer tells you to do or not.

If you disagree with me, go read the definition again. Look at how the definition talks about the roles and software features of a given level.

As above, I disagree with this for the same reasons stated above: The spirit of autonomy is a machine doing a job on its own. A Level 5 car is allowed to have driver controls and be operated by a human at the human's discretion, and this could be using an iPad. But then it's not in L5 (fully autonomous) operation.

An iPad could be a solution for a Level 4 car that has so few limitations to its autonomy that it only needs a "driver" so rarely that for normal road driving it doesn't need a steering wheel. But it's still just Level 4. And if it needs human input while driving, then it's only Level 3.

Just as my car performs all the driving tasks while it's in autosteer, but it's only Level 2 because I'm responsible for disengaging it before it makes a mistake. Level 3 performs all the driving tasks while engaged, but can request human assistance while driving, with sufficient advance warning to the driver. Level 4 can request human assistance, but only after parking safely on its own. And Level 5 never needs human assistance other than providing it with a destination before the ride.
 
As above, I disagree with this for the same reasons stated above: The spirit of autonomy is a machine doing a job on its own. A Level 5 car is allowed to have driver controls and be operated by a human at the human's discretion, and this could be using an iPad. But then it's not in L5 (fully autonomous) operation.

As long as the feature fulfills the level 5 specifications, then it's level 5. It doesn't matter what the spirit is or what you think or want lol. That's why I think the definitions aren't useful and have been saying it for a while now. The specifications don't define any safety or performance threshold as it relates to driving.

What people misunderstand all the time is that the levels define the specifications of a software feature. That's all. It doesn't define what the developer says or tells you. The developer can tell you to always supervise the feature, but as long as the feature doesn't require / expect you to be a fallback driver, then it can be level 4/5...
 
  • Funny
Reactions: AlanSubie4Life
The point was that we see all these "fully autonomous" drive videos from fsd developers, but in actuality, they could be having iPad guys making some of the decisions.

In a nutshell, the levels mean nothing to me anymore. Their definitions are well-understood by very few, and they don't help gauge progress or performance whatsoever, so what's the whole point of using them if what we really care about is how the fsd performs in the real world (without human decision making or interaction)?
All these developers had impressive demo videos way before they started using remote assistance. What matters is if they can achieve the required level of safety without a driver in the car. The problem with demo videos is that they're at most an hour long and the car needs to drive well over 10,000 hours without a collision!
I don't think there will ever be an L4 car without remote assistance of some sort so it seems silly to talk about adding more levels to measure "progress".
 
  • Like
Reactions: diplomat33