Elon: "Feature complete for full self driving this year"

[S4WRXTTCS]

What happened to "automatically change lanes WITHOUT requiring driver input".

@S4WRXTTCS are you seeing what i'm seeing?

Yeah, I thought I commented on it. But, there are so many discussions on it that I lost track of where I responded.

In any case.

Before 2019.8.5 was announced/released I made a comment that said I didn't believe Tesla was ballsy enough to release unconfirmed lane changes any time soon.

Then shortly after the blog post someone pointed out that I was wrong.

But, was I wrong?

I would say no because like you pointed out the expectation was for no driver input. That's exactly why I thought ULC was too ballsy to actually release.

Now am I disappointed? Not at all. I didn't think it was going to happen, and it didn't happen. I'm still skeptical of ever having a true ULC without rear corner radars (or proven L3 driving).

The feature shouldn't have been promised in the first place. It's squarely an L3 feature because how is a human supposed to oversee something that happens either automatically or with a single beep? The implementation was the best Tesla could do realistically. I'm not sure European regulators would even allow ULC as it was listed on the EAP options page.

I think it comes down to not getting what I wanted, but what I needed or something like that.

What I wanted (as in what was promised) was probably a bad idea.

I do think it improves on NoA, and now they can focus on improving right lane performance and navigation along with refining things so it doesn't do stupid things like turning on the blinkers when there is a car next to me. Why would anyone want that? Unless the car absolutely needed to get over soon. I usually change lanes well ahead of where I absolutely need to do so I don't have to bug anyone. That's my nature, and I want my car to follow my nature.

 

[mhan00]

What I was trying to say is that I doubt Tesla will actually launch the Tesla Network on April 22. What Tesla will most likely do on April 22 is give some details of how they envision the Tesla Network working whenever it does launch.

I wonder if they might actually launch the Tesla Network, but not as a FSD thing at first but as a straightforward Lyft/Uber alternative where the owner is still the driver and does ride shares if he/she wants to? It would let them start working out the stuff for the app and help them figure out how to set up the logistical stuff like insurance, payment, etc while also getting their name out there as an Uber/Lyft alternative and hopefully build up a user base. Probably not, but something to think about while waiting the couple of weeks until the event.

 

[AlanSubie4Life]

I would say no because like you pointed out the expectation was for no driver input.

My opinion is having hands on the wheel is not driver input in a level 2 system.

It’s possible that if you look over arguments that only two members posting here are familiar with that have taken place over the prior 3 years you might be able to construe this differently.

But that is not the discussion here. Seamless NoAP does not require driver input for lane changes, as has been described here. I don’t know how one could interpret it differently. If I change my mind after trying it, I will correct my posts.

Am I missing something?

 

[Daniel in SD]

I wonder if they might actually launch the Tesla Network, but not as a FSD thing at first but as a straightforward Lyft/Uber alternative where the owner is still the driver and does ride shares if he/she wants to? It would let them start working out the stuff for the app and help them figure out how to set up the logistical stuff like insurance, payment, etc while also getting their name out there as an Uber/Lyft alternative and hopefully build up a user base. Probably not, but something to think about while waiting the couple of weeks until the event.

Or maybe the Tesla Network will be a car rental service like Turo. That makes more sense than getting into a money losing business like Uber/Lyft.

 

[mongo]

I don't think it would be that hard to remove sensors. I think you're underestimating how difficult the rest of the software is even if you've got a system with perfect "vision". Anyway it doesn't take that many miles to verify safety. Like I said humans get into accidents every 150k miles. A few million miles seems like it should be enough. If Tesla could train one thousand of their employees to be test drivers they could cover a million miles in a month. It just doesn't seem like a major hurdle given the ridiculous amount money one could make selling FSD vehicles.

Oh, I'm not saying it's hard to remove sensors. I'm saying it is hard to get enough miles on the sensor reduced system to convince people to use it instead.
Look at Tesla:
1. So so EAP with nags
2. Good EAP with nags
3. Really good EAP with nags (actually FSD code in final testing)
4. FSD
done

Now, if you are another vendor, how do you get millions of miles on new hardware? One thousand drivers does not a full data set make, nor does a sub-year data collection effort. And it still requires making a thousand cars with reduced functionality (nags).
Do you produce cars with the new hardware and force people to use nags until it is proven? If so, buyer will be annoyed that they do not have the features the older models do. Do you disable the sensors on existing cars? That also seems questionable. What if there is an issue with the new sensor suite? Recall all the cars?
Meanwhile, your gen 1 system is racking up more miles of verified goodness...

 

[mhan00]

Re: torque on the wheel as confirmation

It is, but I like the requirement for now. Requiring the driver to prove they have the wheel solidly in hand is smart since if something unexpected happens during a lane change, having hand on wheel already makes it much easier to take over than if the person was just doing the volume control scroll every so often with their hands on their lap the rest of the time like too many people seem to like to do. And if you’re like me, who keeps at least one hand on the wheel 99% of the time anyway (besides times when I scratch my ear or something, which is really no different than when I’m actually driving instead of using AP), it can often seem like no confirmation at all since I’m providing torque most of the time anyway. It’s another incremental step, but probably a necessary one and It was still pretty surreal and remarkable while I was using NOA last night, and I’m looking forward to a longer drive with it when I’m heading up to SF next weekend.

 

[S4WRXTTCS]

My opinion is having hands on the wheel is not driver input in a level 2 system.

It’s possible that if you look over arguments that only two members posting here are familiar with that have taken place over the prior 3 years you might be able to construe this differently.

But that is not the discussion here. Seamless NoAP does not require driver input for lane changes, as has been described here. If I change my mind after trying it I will correct these posts.

I think it's actually going to come down to which Tesla it is.

When I had a 2015 Model S I could easily hold the steering wheel with hardly ever getting nags. Whatever torque sensor it had was pretty sensitive.

With my new 2018 Model P3D I have a tough time holding the steering wheel without getting nagged. I've only found one position where it doesn't nag me, and it's terribly uncomfortable.

So for me ULC as implemented in 2018.8.5 will most definitely require confirmation more often than not.

At least that's my prediction, and from what I've heard/seen from Model 3 drivers. I'll try it myself this weekend as I got the update last night.

What's funny about this one is I actually want confirmation as it's implemented using the torque sensor. My prediction is I'll be pretty happy with Steering wheel vibrate for notification, and slight torque steer to confirm.

So happy with the implementation, and less happy about other things about NoA (driving policy, lack of good maps, lack of alternative routes, etc).

 

[Daniel in SD]

Oh, I'm not saying it's hard to remove sensors. I'm saying it is hard to get enough miles on the sensor reduced system to convince people to use it instead.
Look at Tesla:
1. So so EAP with nags
2. Good EAP with nags
3. Really good EAP with nags (actually FSD code in final testing)
4. FSD
done

Now, if you are another vendor, how do you get millions of miles on new hardware? One thousand drivers does not a full data set make, nor does a sub-year data collection effort. And it still requires making a thousand cars with reduced functionality (nags).
Do you produce cars with the new hardware and force people to use nags until it is proven? If so, buyer will be annoyed that they do not have the features the older models do. Do you disable the sensors on existing cars? That also seems questionable. What if there is an issue with the new sensor suite? Recall all the cars?
Meanwhile, your gen 1 system is racking up more miles of verified goodness...

You produce 1000 cars and pay people to drive them 100 miles a day for a month. Now you've got 3 million miles of data, enough to prove safety greater than a human driver. This is by far the easiest part of developing FSD, all you need is money, and not even that much money relative to the value of FSD ($80 billion has been invested so far!).
If in some bizarre scenario people insist on buy your more expensive 1st gen solution then that's a good problem to have, you make even more money!

 

[AlanSubie4Life]

With my new 2018 Model P3D I have a tough time holding the steering wheel without getting nagged. I've only found one position where it doesn't nag me, and it's terribly uncomfortable.

I see. I have had no such problems (other than on my very first couple drives as I learned the system) but obviously the resulting comfort would depend on how the driver is used to holding the wheel.

 

[Engr]

Yeah, I thought I commented on it. But, there are so many discussions on it that I lost track of where I responded.

In any case.

Before 2019.8.5 was announced/released I made a comment that said I didn't believe Tesla was ballsy enough to release unconfirmed lane changes any time soon.

Then shortly after the blog post someone pointed out that I was wrong.

But, was I wrong?

I would say no because like you pointed out the expectation was for no driver input. That's exactly why I thought ULC was too ballsy to actually release.

Now am I disappointed? Not at all. I didn't think it was going to happen, and it didn't happen. I'm still skeptical of ever having a true ULC without rear corner radars (or proven L3 driving).

The feature shouldn't have been promised in the first place. It's squarely an L3 feature because how is a human supposed to oversee something that happens either automatically or with a single beep? The implementation was the best Tesla could do realistically. I'm not sure European regulators would even allow ULC as it was listed on the EAP options page.

I think it comes down to not getting what I wanted, but what I needed or something like that.

What I wanted (as in what was promised) was probably a bad idea.

I do think it improves on NoA, and now they can focus on improving right lane performance and navigation along with refining things so it doesn't do stupid things like turning on the blinkers when there is a car next to me. Why would anyone want that? Unless the car absolutely needed to get over soon. I usually change lanes well ahead of where I absolutely need to do so I don't have to bug anyone. That's my nature, and I want my car to follow my nature.

About the only thing that will make @Bladerskb happy is if someone starts a thread about how awful ULC NoA is, but so far so good all we hear is people mostly saying they're happy with the improvements. Perhaps @jimmy_d will come to his rescue for a change, poor bladerskb!

 

[mhan00]

About the only thing that will make @Bladerskb happy is if someone starts a thread about how awful ULC NoA is, but so far so good all we hear is people mostly saying they're happy with the improvements. Perhaps @jimmy_d will come to his rescue for a change, poor bladerskb!

There’s a couple of people on reddit who aren’t that impressed. I think they’re mostly outside of CA. I know greentheonly has mentioned CA has better and more detailed mapping data in comparison to other states, so I’m wondering how much that has to do with any differences in opinions?

 

[Bladerskb]

Yeah, I thought I commented on it. But, there are so many discussions on it that I lost track of where I responded.

In any case. Before 2019.8.5 was announced/released I made a comment that said I didn't believe Tesla was ballsy enough to release unconfirmed lane changes any time soon. Then shortly after the blog post someone pointed out that I was wrong. But, was I wrong?

I would say no because like you pointed out the expectation was for no driver input. That's exactly why I thought ULC was too ballsy to actually release.

Now am I disappointed? Not at all. I didn't think it was going to happen, and it didn't happen. I'm still skeptical of ever having a true ULC without rear corner radars (or proven L3 driving).

The feature shouldn't have been promised in the first place. It's squarely an L3 feature because how is a human supposed to oversee something that happens either automatically or with a single beep? The implementation was the best Tesla could do realistically. I'm not sure European regulators would even allow ULC as it was listed on the EAP options page.

 

[mongo]

You produce 1000 cars and pay people to drive them 100 miles a day for a month. Now you've got 3 million miles of data, enough to prove safety greater than a human driver. This is by far the easiest part of developing FSD, all you need is money, and not even that much money relative to the value of FSD ($80 billion has been invested so far!).
If in some bizarre scenario people insist on buy your more expensive 1st gen solution then that's a good problem to have, you make even more money!

I see your point, but I'm not sure I agree, and that's ok.
That gives a mile count, sure. But does it give you the geographic, situational, and season coverage you need?
Ideally, they would have data from the initial testing that they could reuse. On the other hand, Tesla Model 3 already has a billion miles under its belt...

 

[AlanSubie4Life]

There is no additional driver input required to change lanes. That's why it's called an unconfirmed lane change. Some users may not like the way they have to keep their hand on the wheel, and it's possible there is a sensitivity change (I'll see), but no additional driver input is required for a lane change to occur. That's what everyone above has said. You can't claim that the torque sensor is a driver input because that is a prerequisite for being in NoA. Nothing additional is required to change lanes.

I don't know why you keep grasping on shreds of what people say to pound this obsession of yours. As I said, it seems like you have legitimate trolling which you can engage in with better factual backing. You're squandering credibility, as someone else mentioned.

Anyone trying to use EAP/AP without at least one hand on the wheel is really asking for trouble. The system is completely incapable of being used that way safely. So, ULC is perfect for the current capability. When (if) the hands can come off the wheel, the system will be level 3 at least.

 

[AlanSubie4Life]

if the person was just doing the volume control scroll every so often with their hands on their lap the rest of the time like too many people seem to like to do

Who are these people, and how did they survive long enough to buy a Tesla? They must be straight-up certifiable. Completely insane. Not only would it be incredibly annoying to use this procedure, it's also incredibly dangerous.

 

[diplomat33]

@Bladerskb Just for you, I did a test drive of the new NOA earlier today as soon as I got off work. Here is my preliminary report:

NOA indicated an upcoming left lane change in advance to get me around a slower car in front of me, automatically put my blinker on and when I tugged the wheel to dismiss the AP nag, it steered smoothly into the left lane. When the right lane was clear, it repeated the process to move me back into the right lane. So it was able to get me around slower cars with 2 smooth lane changes, with no issues. Lane changes were smart and easy.

Also, NOA does automatically take exits with no driver input at all. There was no nag or anything. The car notified me of an exit change, put the blinker on, steered onto the exit on its own and then turned the blinker off. Once in the exit, the car gives you a countdown notification (300 ft... 200 ft... 100 ft) that NOA will switch back to EAP. It was smooth and easy.

Overall, NOA is much better. Not having the stalk confirmation is better because instead of 2 different driver inputs, there is only 1 driver input (the tug) for all AP nags whether doing a lane change or not. So driver input required is more consistent and intuitive across different driving scenarios.

Overal, I would rate the new NOA as a big improvement.

Now to answer your question: yes there is a AP nag before the car actually steers into the adjacent lane. The driver must tug the wheel before the car will actually steer into the adjacent lane. So in that respect, I can see how it might be seen as another lane confirmation. However, I would consider it more of a fail safe than a confirmation. There is a big difference between the stalk confirmation and this "wheel confirmation" as you called it. In the previous NOA, the car would merely suggest a lane change and the driver needed to tap the stalk to agree with the lane change and then the car would execute the lane change. So the car was merely suggesting a lane change. In this new version, the car is actually capable of doing the entire lane change with no driver input. It sees surrounding cars, notifies the driver in advance of a lane change, automatically puts on the blinker and could execute the lane change itself on its own. The tug on the wheel is more of a fail safe than a confirmation because it is not asking for permission but instead checking that the driver has their hands on the wheel. In other words, the car is capable of doing the lane change on its own and prepares to do so on its own (driver notification, turning on blinker) but only executes the lane change when it knows the driver has their hands on the wheel. I feel like preparing to do a lane change and only executing the lane change when the driver has their hands on the wheel rather than just suggesting lane changes is an important distinction and a meaningful improvement in NOA!

Also, this "wheel confirmation" is better than the stalk confirmation because it is consistent with what the driver needs to do for other AP nags but also puts an emphasis on safety by ensuring that the driver has their hands on the wheel when the car actually steers into an adjacent lane.

At least in the scenarios that I tested it in, the new NOA was able to do all the highway driving on its own with no additional driver input other than ensuring that the driver has their hands on the wheel.

 

[AlanSubie4Life]

Now to answer your question: yes there is a AP nag before the car actually steers into the adjacent lane. The driver must tug the wheel before the car will actually steer into the adjacent lane.

Interesting. Somewhat different assessments from you vs. @Soda Popinski Although he did mention a possible change in sensitivity he didn't say it required an additional tug. Does the direction matter? (Does it require you to tug in the direction of the lane change? I assume not...)

I guess I'll have to try it myself to get a feel for it. Have the update but I'm waiting on a couple other things before installing it. I'm always tugging on the steering wheel with a significant portion of the weight of my arm, so I wonder how that will work. With any harder a tug from me I would disengage autosteer.

 

[Soda Popinski]

Who are these people, and how did they survive long enough to buy a Tesla? They must be straight-up certifiable. Completely insane. Not only would it be incredibly annoying to use this procedure, it's also incredibly dangerous.

I would never do this at speed. I'll admit I've rested my hands in my lap when in stop and go doing an average or 3 or 4 mph. Especially when at a complete stop for more than 15 seconds. Ah, the Sepulveda Pass.

That said, I'll add NoAP, while improved, isn't ready for Los Angeles traffic. It was barely adequate on my morning commute around 6:30am. I left work early to avoid the Trump traffic, and found the extra afternoon traffic at around 3pm to be too much for it. It hesitated far too long, resulting in me getting honked at. I stopped unsafely. I ended up canceling it, as I felt unsafe on the drive home.

 

[Bladerskb]

Honestly, I think you are kinda grasping at straws now. You are ignoring reviews that say NOA is really good. You are ignoring that the car does lane changes on its own with no stalk confirmation. You are just left with the fact that Tesla added an extra AP nag before lane changes to try to argue that it's not what Tesla promised because they promised "no driver input". Sure, but the AP nags are easy to dismiss. You don't even need to hold the wheel to do it, just pushing the left scroll wheel is good enough. NOA still handles all highway driving with no direct driver input other than dismissing AP nags from time to time.

But I got the update today so I will do an extensive test of the new NOA this weekend and give you a more detailed review if you like.

This is not true. If you looked around here you will find my view of Tesla compared to the competition (the entire field) which is actually more optimistic than most. Yet i have been bulls-eye. I have been bulls-eye because i look at the evidence before me rather than let my emotions cloud my judgment.

In the morning when i saw the video, my immediate post on reddit was:
"Tesla is definitely ahead of companies like Toyota, Honda, Mercedes, Hyundai, jaguar..."

Then hours later when asked the question in the quote below: i followed up with...

But behind GM, VW, Waymo, Zoox, and Ford ??

"
'But behind Mobileye, GM, VW/Audi, Waymo, Zoox, and BMW ??'

Precisely, Although i would remove Ford. Ford has not shown willingness to adopt ADAS (incremental autonomy sold to customers) and their L4 Taxi Argo AI is still a toss up to me, they haven't shown the public anything yet.

For the other list of who they (tesla) are better than, i forgot to add Bosch.

Basically anyone not on either list is an unknown, meaning we can't predict where they stand in relation to tesla. But this is a fast industry, the leadership board changes every 6 months.

Now this doesn't give credence to Elon's timelines and as i have said over and over again, is complete BS and made up. Proven by the fact that he's been saying 2 years since 2015 and have been wrong every year.

But the companies they are ahead of are so incompetent and behind. For example Toyota with its "we wont be developing self driving cars because humans are so good" or Mercedes with its yearly CES demo of riding around a parking lot and now this year punking to BMW and basically begging them so they can have a Level 3 car in 2025. 2025. Talk about being behind.

I could go on and on."

You can even check my posts in 2016 and early 2017, my views hasn't changed about where people stand. I always saw Toyota and Mercedes to be right at the end of the line and Tesla in the middle.

People look at me as an anti-Tesla, but actually i'm a realist who evaluates the facts daily and simply calls out Elon for his constant obvious BS and others too. You are not in reddit so you don't see it and think i'm only calling out Elon. Although Elon is the worst BSer, its not even close.

But what i dislike is the obvious contradiction from fans. Now its the whole demo shenanigans while before talking about a demo was considered by fans as a bannable offense. Now its being praised.

What people didn't realize was that demos/test rides of actual products in development are valuable because you see how the product have actually improved over time with each media/public rides. Tesla's 2016 demo was not a demo of an actual active development product, it was a one time photo-op, with a code that was deleted.

Fans refuse to accept that because they didn't have a progressive Tesla demo that they can point to that matched the complexity of what other companies were showing thier car tackling in their demos. So they denounced any kind of demo entirely.

Now that Tesla is doing demos again, its finally okay to love demos and the excuse is "well its done on production cars". Like you do realize the cars of Waymo, GM or Mobileye for example ARE production cars? Those are the cars they will use when they go commercial.

 

[diplomat33]

Interesting. Somewhat different assessments from you vs. @Soda Popinski Although he did mention a possible change in sensitivity he didn't say it required an additional tug. Does the direction matter?

I guess I'll have to try it myself to get a feel for it. Have the update but I'm waiting on a couple other things before installing it. I'm always tugging on the steering wheel with a significant portion of the weight of my arm, so I wonder how that will work. With any harder a tug from me I would disengage autosteer.

It worth noting that I did not have both hands on the wheel when the car wanted to make a lane change which might explain the strong AP nag. Maybe if I was holding the wheel firmly, the car would have made the lane change without giving me any AP nag.

No, you do not need tug in the direction of the lane change.

The sensitivity was definitely stronger in the AP nag during the lane change. Normally on EAP, I can easily dismiss the nag with one finger but during the lane change I needed to give a stronger tug. And the screen was flashing blue right away like when you miss the first nag.

Keep in mind that this was my first experience with the new NOA. This is just my very first impressions.