Will Mercedes jump to level 3 before Tesla? Looks like it.

[Daniel in SD]

Absolutely the opposite.

No supervision (i.e. Tesla taking legal liability) is very bad from investor POV. Tesla is on the hook for millions of $ in liability but get nothing extra in return than they are already getting.

Some tiny ODD unsupervised AP (like < 40 mph on highways) to me is worthless. I'll almost never use it compared to using FSD Beta 90%+ of the time.

I’m not an investor. It sounds like you have a conflict of interest.
I wouldn’t buy the Mercedes system either because the freeways are rarely congested on my commute. If Mercedes gets it working at full speed that would be very valuable to me.
I wish we had usage stats on FSD beta. To me it seems useless but there seem to be many people who find it valuable.

 

[EVNow]

I’m not an investor. It sounds like you have a conflict of interest.

It is - thats why I clearly state the POV and also note in my Sig that I'm long TSLA.

BTW, since you don't have FSD Beta, isn't it also conflict of interest for you ?

I wouldn’t buy the Mercedes system either because the freeways are rarely congested on my commute. If Mercedes gets it working at full speed that would be very valuable to me.

Absolutely - I hardly ever drive on congested highways. So, it is absolutely useless to me. Esp. because congested traffic doesn't constantly remain that way ... atleast on roads I drive - the speeds vary greatly - from 60 mph to 20 mph, frequently. So, by the time I "open my laptop" and login - I've to close it and take over.

So, as I said, both from investor POV and customer POV, it would be bad.

I wish we had usage stats on FSD beta. To me it seems useless but there seem to be many people who find it valuable.

Oh yes - if only I can get my hands on all the FSD Beta stats ....

 

[EVNow]

Actually Tesla would get something in return. A much larger than 10% take rate on FSD. Say take rate goes up to 25%, or say 100,000 more cars/yr. That’d be an extra $1 billion. Investors would be very happy.

Not really. Would you pay $12k to get unsupervised highway driving but below 38 mph ? If rest of FSD doesn't interest me, this certainly won't.

There might be a small % of people who are fence sitters who might find it useful enough to switch. I have no doubt, though, that the increase won't be anywhere close to 150% (are you kidding me ?!).

 

[nvx1977]

AP has worked incredibly well in bumper to bumper traffic for years. I have an 80-mi round trip commute into Boston, and that commute was torture in my old ICE car. AP is heavenly. Been using it since 2018. It's only failed me one time (I had to intervene). I reported it to Tesla's AP team, and they actually reached out to me because it was such a rare one-off situation that caused the problem.

I really think AP in bumper-to-bumper (whatever low speed threshold they decide) could be sold as L3 and there would be very few liability accidents. I just don't think Tesla cares about L3. The goal has always been "Full" self driving.

Note: in a more recent update last year, AP accelerates hard from a stop and brakes hard while in bumper-to-bumper traffic. It's not every stop that it does this, but enough to be annoying. There's no indication Tesla will fix this in the old code branch. Likely will see it resolved with single stack.

 

[Daniel in SD]

It is - thats why I clearly state the POV and also note in my Sig that I'm long TSLA.

BTW, since you don't have FSD Beta, isn't it also conflict of interest for you ?


Absolutely - I hardly ever drive on congested highways. So, it is absolutely useless to me. Esp. because congested traffic doesn't constantly remain that way ... atleast on roads I drive - the speeds vary greatly - from 60 mph to 20 mph, frequently. So, by the time I "open my laptop" and login - I've to close it and take over.

So, as I said, both from investor POV and customer POV, it would be bad.


Oh yes - if only I can get my hands on all the FSD Beta stats ....

I don’t even like auto lane change because it’s unpredictable (and makes me look like an idiot) so I doubt I’d like FSD Beta. My friend has it and it doesn’t seem like something I’d enjoy.
If Tesla sold Drive Pilot they’d promise that it would soon work at high speeds which would increase the take rate.

 

[AlanSubie4Life]

To me it seems useless but there seem to be many people who find it valuable.

I don't believe anyone who has FSD Beta who claims it provides functional utility above the standard 2018 EAP feature set. It's a curiosity and a science project right now, nothing more. It may well provide entertainment utility and novelty value to some users but that's a different topic.

don’t even like auto lane change because it’s unpredictable (and makes me look like an idiot) so I doubt I’d like FSD Beta

I do wonder if the performance is crap with your archaic hardware, but really hard to know. Or are you talking about the full auto lane change (which is awful and completely useless and has no contextual awareness), and not the auto lane change on request (which is fine now that it doesn't have weird abort behavior as long as you're not trying to make congested lane changes, or anything other than plain vanilla lane changes)? I think you'd use FSD Beta and laugh at it if you had it. Not sure you'd like it though (would depend strongly on how much you paid for it).

 

[Goose66]

I generally agree but I think there would be some benefit for Tesla to be able to say they have true level 3 on highways. I think you're right, though - the financial liability is still a significant concern for investors.

Honestly, 'plain' AP on highways like Mercedes is doing I can't remember the last time I needed to intervene. Even up to normal highway speeds it does incredibly well.

I have some problems with lane changes near exits. The car gets squirely when a car in front is taking an exit ramp - like it needs to go around... no the car is in another lane... no it has to go around... wait... etc. Other than that, with Speed-based Lane Changes on Mild, I feel like I would trust Navigate on AutoPilot in L3 if it had a fallback routine built-in (in other words, it didn't just say "Take Over Immediately" and relinquish control.

Oh yeah, and emergency vehicles stopped on side of the road - still needs work there too.

 

[Daniel in SD]

I don't believe anyone who has FSD Beta who claims it provides functional utility above the standard 2018 EAP feature set. It's a curiosity and a science project right now, nothing more. It may well provide entertainment utility and novelty value to some users but that's a different topic.

You have to believe in The Mission.

I do wonder if the performance is crap with your archaic hardware, but really hard to know. Or are you talking about the full auto lane change (which is awful and completely useless and has no contextual awareness), and not the auto lane change on request (which is fine now that it doesn't have weird abort behavior as long as you're not trying to make congested lane changes, or anything other than plain vanilla lane changes)? I think you'd use FSD Beta and laugh at it if you had it. Not sure you'd like it though.

I do use it because disengaging and reengaging Autosteer is annoying. I just really hate when it aborts for no reason or just sits there with the signal on. I try to recognize situations where it might have trouble but am not always successful. Maybe Tesla Vision fixes it, should have bought FSD for $2k when I had the chance!

 

[AlanSubie4Life]

I do use it because disengaging and reengaging Autosteer is annoying. I just really hate when it aborts for no reason or just sits there with the signal on.

This never happens to me anymore (it used to). I'm excluding cases where there are cars which could conceivably be close enough or moving fast enough to interfere with its ponderous decision making. Tesla Vision FTW. It's a miracle.

 

[Daniel in SD]

It's funny when people use their personal experience to claim that Autopilot would have acceptable safety without supervision. How many miles between collisions is acceptable? How many miles have you driven without a safety disengagement?

 

[sleepydoc]

It's funny when people use their personal experience to claim that Autopilot would have acceptable safety without supervision. How many miles between collisions is acceptable? How many miles have you driven without a safety disengagement?

And just what else are we supposed to do in the absence of other data? Do you have some mysterious trove of data on accidents and disengagements in various circumstances that you can share with us?

 

[sleepydoc]

I have some problems with lane changes near exits. The car gets squirely when a car in front is taking an exit ramp - like it needs to go around... no the car is in another lane... no it has to go around... wait... etc. Other than that, with Speed-based Lane Changes on Mild, I feel like I would trust Navigate on AutoPilot in L3 if it had a fallback routine built-in (in other words, it didn't just say "Take Over Immediately" and relinquish control.

Oh yeah, and emergency vehicles stopped on side of the road - still needs work there too.

Except if we're talking about Mercedes' system, there are no lane changes and you just pop an alert up when you see an emergency vehicle saying 'take over.' That's the thing - the Mercedes system is still so limited that it doesn't have to worry about any of these situations.

All of my issues with cars exiting is more with FSD when a car is moving into a turn lane and FSD is too conservative, waiting until the car's been in the turn lane for a bit before it even starts to accelerate and pass it. This isn't technically wrong or unsafe, though, just not what most people do. I've never experienced AP do anything unsafe on the highway in these circumstances.

 

[Dewg]

This isn't technically wrong or unsafe, though, just not what most people do. I've never experienced AP do anything unsafe on the highway in these circumstances.

This is the problem with all L2-L5 systems. Most people, especially in urban areas, don't drive to the letter of the law. They speed, do rolling stops, make unsafe lane changes, turn when a pedestrian is in a crosswalk, pass a school bus when it's lights are flashing, etc. And for some reason they expect everyone else to do the same and get annoyed or feel inconvenienced when someone drives "correctly".

I've been honked at while on FSD Beta for my car driving correctly and to the law. Most recently it was in the right lane of 2 left turn lanes. In California, the left lane of 2 left turn lanes must turn into the left-most lane. The right lane can turn into the #2 lane or #3 lane (if there are 3+ lanes, and it is safe to do so). But the guy to my left wanted to be in the #2 lane, so when my car turned perfectly into the #2 lane, he got really close as he was drifting over (like so many people do - oversteer across the lines) and honked at me and gave me a dirty look. Dude - my car did EXACTLY what it was supposed to do, and drove EXACTLY to the letter of traffic law, but Oooohhhhh no, that just won't do...

 

[nvx1977]

This is the problem with all L2-L5 systems. Most people, especially in urban areas, don't drive to the letter of the law. They speed, do rolling stops, make unsafe lane changes, turn when a pedestrian is in a crosswalk, pass a school bus when it's lights are flashing, etc. And for some reason they expect everyone else to do the same and get annoyed or feel inconvenienced when someone drives "correctly".

I've been honked at while on FSD Beta for my car driving correctly and to the law. Most recently it was in the right lane of 2 left turn lanes. In California, the left lane of 2 left turn lanes must turn into the left-most lane. The right lane can turn into the #2 lane or #3 lane (if there are 3+ lanes, and it is safe to do so). But the guy to my left wanted to be in the #2 lane, so when my car turned perfectly into the #2 lane, he got really close as he was drifting over (like so many people do - oversteer across the lines) and honked at me and gave me a dirty look. Dude - my car did EXACTLY what it was supposed to do, and drove EXACTLY to the letter of traffic law, but Oooohhhhh no, that just won't do...

those same people are indignant when they are issued a citation after they cause an accident.

 

[Daniel in SD]

And just what else are we supposed to do in the absence of other data? Do you have some mysterious trove of data on accidents and disengagements in various circumstances that you can share with us?

I have had safety disengagements. That's why I can say with high confidence that it's not safe enough. The chances of me just being unlucky are extraordinarily small. However if I had driven 20,000 miles on Autopilot and had zero safety disengagements that would be prove nothing. There's a 50% chance of that happening if the collision rate were 1 per 20k miles!

Except if we're talking about Mercedes' system, there are no lane changes and you just pop an alert up when you see an emergency vehicle saying 'take over.' That's the thing - the Mercedes system is still so limited that it doesn't have to worry about any of these situations.

It has to be able to recognize those situations and deal with them for 10 seconds. That doesn't sound easy to me (and for all I know Drive Pilot will be a horrible failure!).

 

[nvx1977]

Tesla is ok lengthening the nag duration at slow speeds on the highway because they know it is a much lower-risk situation, and accidents when they occur will be less costly from a liability perspective. Not sure how Merc is calculating the risk here, but they clearly think they can handle the liability.

On the consumer side, that is little comfort. Even minor accidents are a major inconvenience. Given the complexity of the problem, I have serious doubts that Merc is somehow way ahead in its capability. If it were me behind the wheel, I'd be treating the system as if it were L2.

 

[sleepydoc]

I have had safety disengagements. That's why I can say with high confidence that it's not safe enough. The chances of me just being unlucky are extraordinarily small. However if I had driven 20,000 miles on Autopilot and had zero safety disengagements that would be prove nothing. There's a 50% chance of that happening if the collision rate were 1 per 20k miles!


It has to be able to recognize those situations and deal with them for 10 seconds. That doesn't sound easy to me (and for all I know Drive Pilot will be a horrible failure!).

What have your disengagements been, and what were the circumstances? The only times I've had to disengage autopilot on the highway was for routing issues. (non-limited access highways, county roads, etc are a different story.) Again, if we're talking about the level of the Mercedes system, pretty much all it has to do is stay in the lane.

IME, Tesla can freak out and ask you to take over pretty quickly. I have no idea what analysis is being done but really, all we're talking about here is "I'm not 100% sure- bail!" type decision.

 

[AlanSubie4Life]

. There's a 50% chance of that happening if the collision rate were 1 per 20k miles!

Isn't it (probability of zero safety disengagements in your stated interval) approximately 1/e, or 36.8%, or am I thinking about this wrong? (1-1/20e3)^20e3. You'll have to show your work with probability pronouncements; I can never get it straight.

 

[Daniel in SD]

What have your disengagements been, and what were the circumstances? The only times I've had to disengage autopilot on the highway was for routing issues. (non-limited access highways, county roads, etc are a different story.) Again, if we're talking about the level of the Mercedes system, pretty much all it has to do is stay in the lane.

IME, Tesla can freak out and ask you to take over pretty quickly. I have no idea what analysis is being done but really, all we're talking about here is "I'm not 100% sure- bail!" type decision.

Most recent one: Real Autosteer Edge Case
Though it's quite possible at low speeds it wouldn't have been a problem. At lower speeds Autopilot seems to have no awareness of people merging into your lane (even semi trucks!). It surprises me that some people never have to disengage.

 

[Daniel in SD]

Isn't it (probability of zero safety disengagements in your stated interval) approximately 1/e, or 36.8%, or am I thinking about this wrong? (1-1/20e3)^20e3. You'll have to show your work with probability pronouncements; I can never get it straight.

Haha. Yes, must be statistically rigorous here on TMC. It is 36.8%. Need to brush up on my statistics I guess...