Tesla autopilot HW3

[Snuffysasa]

As I'm new, would you mind clarifying the difference. When the stop/start at stop signs/lights comes out, but before city driving, what will it not be able to do while doing the stopping and starting? Will it follow nav destination while doing this? Is it that it will just go down one street, stopping and starting, but not being able to make the turns - is that the main difference?

I assume at first they will release it where it still only goes straight, but doesn't make any turns.

Also I suspect they could do something where it will stop on its own... but to start back up again.. would require "stalk confirmation" for quite some time. (or some other kind of confirmation)

 

[tufa4311]

Also I suspect they could do something where it will stop on its own... but to start back up again.. would require "stalk confirmation" for quite some time. (or some other kind of confirmation)

So even when they have no stalk confirmation on NoA you think they'll still need it on city driving?

 

[diplomat33]

Also I suspect they could do something where it will stop on its own... but to start back up again.. would require "stalk confirmation" for quite some time. (or some other kind of confirmation)

Why not just use the accelerator to start up again? I think that would make more sense than the stalk. But again, I don't think there is a need for any confirmation. If the car is reliable enough to see a red light, it should be reliable enough to see a green light. And if the car has correctly stopped at a red light, it should be able to go again at a green without needing any special permission.

 

[Daniel in SD]

Why not just use the accelerator to start up again? I think that would make more sense than the stalk. But again, I don't think there is a need for any confirmation. If the car is reliable enough to see a red light, it should be reliable enough to see a green light. And if the car has correctly stopped at a red light, it should be able to go again at a green without needing any special permission.

This is why I would be very skeptical of FSD features. The first time the system makes a mistake and accelerates into an intersection when it is not safe Tesla is going to have a hard time pinning it on the driver.
Another issue is that some jurisdictions, like California, require a permit to test autonomous vehicles. It's not clear how close to an autonomous vehicle you can get without being subject to the testing rules. Uber tried to take a similar approach to developing FSD (claiming that they were merely developing driver assistance features) and the DMV revoked their vehicle registrations.

 

[diplomat33]

This is why I would be very skeptical of FSD features. The first time the system makes a mistake and accelerates into an intersection when it is not safe Tesla is going to have a hard time pinning it on the driver.

I think Tesla has 3 obvious options:
1) Implement a stalk confirmation or accelerator confirmation to ensure that the car is always starting on green correctly and not accidentally running a red light.
2) Wait until the system is so reliable that it does not make a mistake.
3) Warn the driver that they must be vigilant and keep their hands on the wheel. Maybe increase the standard AP nag at intersections. That way in the odd event that something goes wrong, the driver would be responsible.

 

[verygreen]

The first true FSD features (requiring the FSD computer) are coming later this year based on everything we know.

the traffic light stuff does not appear to require any "FSD computer" (this is what is promised by the EoY)

Stopping and starting at traffic lights and stop signs is the first future FSD feature listed on the website. In fact, we will probably get that particular feature this year

These already work though. In EAP even.

 

[Daniel in SD]

I think Tesla has 3 obvious options:
1) Implement a stalk confirmation or accelerator confirmation to ensure that the car is always starting on green correctly and not accidentally running a red light.
2) Wait until the system is so reliable that it does not make a mistake.
3) Warn the driver that they must be vigilant and keep their hands on the wheel. Maybe increase the standard AP nag at intersections. That way in the odd event that something goes wrong, the driver would be responsible.

It seems like a really hard problem. I think the reaction time required to stop the car quickly enough would be impossible for most people. The car moving forward just a couple feet at an intersection could be catastrophic so I think that requiring some sort of confirmation will be necessary for a level 2 system.

 

[verygreen]

That way in the odd event that something goes wrong, the driver would be responsible.

Driver is always responsible, nag or no nag.

 

[diplomat33]

the traffic light stuff does not appear to require any "FSD computer" (this is what is promised by the EoY)

These already work though. In EAP even.

Both the "stop light recognition" and the "automatic city driving" features are listed on the website as coming this year. But I agree that the "stop light recognition" probably does not require the FSD computer since we've seen it work on a "hacked" AP2 car so we will probably get that feature enabled first in the next few months and then later when the "automatic city driving" feature is ready, Tesla will start doing the AP3 upgrades.

 

[Snuffysasa]

So even when they have no stalk confirmation on NoA you think they'll still need it on city driving?

Yes, of course. them having no stalk confirmation for highway lane changes is entirely irrelevant to them doing in city streets.

First, they will do stop light stop signs in shadow mode (happening now)
Then, enable with some kind of confirmation
then, once there is enough validation, enable without confirmation

Why not just use the accelerator to start up again? I think that would make more sense than the stalk. But again, I don't think there is a need for any confirmation. If the car is reliable enough to see a red light, it should be reliable enough to see a green light. And if the car has correctly stopped at a red light, it should be able to go again at a green without needing any special permission.

That is certainly an option too. and there is the message saying "press accelerator to startup again"

And Maybe not need confirmation for traffic lights where they have right of way... I was more thinking like stop sign intersections... or taking a right on red...

Also, even for going straight through a regular stop light I still think they will need confirmation.... because... starting up again while stopped when it hasn't turned green yet... is a lot more dangerous then not stopping at red.... this might sound backwards, but I don't think so if you think about it.

This is why I would be very skeptical of FSD features. The first time the system makes a mistake and accelerates into an intersection when it is not safe Tesla is going to have a hard time pinning it on the driver.
Another issue is that some jurisdictions, like California, require a permit to test autonomous vehicles. It's not clear how close to an autonomous vehicle you can get without being subject to the testing rules. Uber tried to take a similar approach to developing FSD (claiming that they were merely developing driver assistance features) and the DMV revoked their vehicle registrations.

to your first statement... this is why I think they will require some kind of confirmation for a long time... however for the record... detecting red light turning into a green light after detecting the red light.... is so damn trivial. would have very high reliability.

secondly... in CA... consumer Tesla vehicles with autopilot are already testing autonomy..... and they haven't stepped in for that...

There is no difference between what Tesla is doing now, and if Tesla adds features to stop at red lights.

the traffic light stuff does not appear to require any "FSD computer" (this is what is promised by the EoY)

These already work though. In EAP even.

I was going to bring this up.

Thank you. No FSD computer necessary.

It seems like a really hard problem. I think the reaction time required to stop the car quickly enough would be impossible for most people. The car moving forward just a couple feet at an intersection could be catastrophic so I think that requiring some sort of confirmation will be necessary for a level 2 system.

I agree. especially for right on red... and stop sign intersections.

Both the "stop light recognition" and the "automatic city driving" features are listed on the website as coming this year. But I agree that the "stop light recognition" probably does not require the FSD computer since we've seen it work on a "hacked" AP2 car so we will probably get that feature enabled first in the next few months and then later when the "automatic city driving" feature is ready, Tesla will start doing the AP3 upgrades.

I'm not sure the AP3 will be necessary for auto city driving either.

Also, someone else asked.. but I'm curious what features do you mean in the "auto city driving" that would not be apart of responding to traffic lights and stop signs?

 

[Daniel in SD]

secondly... in CA... consumer Tesla vehicles with autopilot are already testing autonomy..... and they haven't stepped in for that...

There is no difference between what Tesla is doing now, and if Tesla adds features to stop at red lights.

Uber made exactly the same argument though. I suspect that the DMV will only step in if it appears like FSD features are increasing the number of accidents. I think that Tesla releasing "automatic city driving" to untrained customers would be dangerous. Hopefully I'll be proven wrong since it seems like a good way to get autonomous testing done with no cost (profit even!).

 

[Snuffysasa]

Uber made exactly the same argument though. I suspect that the DMV will only step in if it appears like FSD features are increasing the number of accidents. I think that Tesla releasing "automatic city driving" to untrained customers would be dangerous. Hopefully I'll be proven wrong since it seems like a good way to get autonomous testing done with no cost (profit even!).

I agree Uber made the exact same argument.

And I don't think FSD features will increase number of accidents... so then there should one no problem.

I'm just curious and I possibly agree, but what makes you think Tesla releasing automatic driving on city streets is any more dangerous than that they have already released?

for years now you could turn on autopilot on city streets...

 

[verygreen]

for years now you could turn on autopilot on city streets...

Having lived through false positive for a green signal while shootign my video.... The current autosteer does not pretend to stop for anything but cars.

Now with stopping for red - the car slows down for unreadable and red, it lulls you into complacemency, and then it decides that "hey, it's actually green, let's go!" and actually picks up speed. Not at 100% throttle, but still noticeably enough (more so if the slowing down did not slow all that much yet). If you are not prepared - very easy to run a red light/end up on the intersection at a wrong time.

 

[Snuffysasa]

Having lived through false positive for a green signal while shootign my video.... The current autosteer does not pretend to stop for anything but cars.

Now with stopping for red - the car slows down for unreadable and red, it lulls you into complacemency, and then it decides that "hey, it's actually green, let's go!" and actually picks up speed. Not at 100% throttle, but still noticeably enough (more so if the slowing down did not slow all that much yet). If you are not prepared - very easy to run a red light/end up on the intersection at a wrong time.

ahh, I can imagine this scenario. And I can see why that would add some risk.

But it could be easily addressed by requiring accelerator pedal confirmation to accelerate when it thinks it's green... even if it hasn't came to a complete stop yet and has only started to slow.... if this happens often it would be annoying of course.

Also be careful man! Tesla forums can't lose this information source!

 

[Daniel in SD]

And I don't think FSD features will increase number of accidents... so then there should one no problem.

Only one way to find out!

I'm just curious and I possibly agree, but what makes you think Tesla releasing automatic driving on city streets is any more dangerous than that they have already released?

I feel like the AP doesn't encourage complacency since it's basically just radar cruise control with lane tracking. On city streets you've got so many more variables and situations where the driver would have to intervene quickly. Also, I'm very suspicious of Tesla's statistics on AP safety. They don't try to correct for the roads and situations where AP is used and the accident rate is implausibly low for both AP on and AP off.

 

[Snuffysasa]

Only one way to find out!

I feel like the AP doesn't encourage complacency since it's basically just radar cruise control with lane tracking. On city streets you've got so many more variables and situations where the driver would have to intervene quickly. Also, I'm very suspicious of Tesla's statistics on AP safety. They don't try to correct for the roads and situations where AP is used and the accident rate is implausibly low for both AP on and AP off.

Of course Tesla's stats are misleading.

And I guess my thinking is that if they include the feature to where it comes to stop at red lights even when they are the first car at the light.. that it still wouldn't encourage complacency because as you say its still just adaptive cruise control, lane centering, and also stopping at red lights even when you are first vehicle at the light.

It just seems to me like adding that tweak wouldn't increase complacency much. However, I could be wrong.

I wonder if Tesla could test complacency in Shadow mode somehow.

 

[CK_Stuggi]

What exactly did Tesla show on autonomy day? All I saw was a bit of advanced summon and NoA. The video Tesla later released on YouTube was a preprogrammed round trip.
Better quality but actually less challenging traffic wise than a similar video I saw in October 2016.

Now - having seen the progress AP2+ made in 2018 I'm hopeful this time may be different, so I'm hoping to get the FSD computer sooner rather than later, if it's only for DashCam.
After all, 2017 was a desaster so far removed from the Elon claim of "feature parity" end of 2016, which has not been reached to this day.
The absence of speed sign recognition is apalling, and as a feature available in very basic cheapo cars and coming from a country with no stop signs or traffic lights on highways, it's quite telling that Tesla still can't do this (or requires much more powerful hardware than everybody else).

So, yeah, I want the new hardware, in the hopes they finally deliver what they promised to achieve 2.5 years ago. But I'll only believe it when I see it.
A staged YouTube video is not enough.

I bought my AP2 car and took a strong hit between my old AP1 85D and the new, pricier AP2 90D.
So, yeah, it's definitely fool me twice territory to me, and I'm not going there.

 

[EVNow]

Only one way to find out!

I feel like the AP doesn't encourage complacency since it's basically just radar cruise control with lane tracking. On city streets you've got so many more variables and situations where the driver would have to intervene quickly. Also, I'm very suspicious of Tesla's statistics on AP safety. They don't try to correct for the roads and situations where AP is used and the accident rate is implausibly low for both AP on and AP off.

Well, NOA is lot more than just lane keeping.

Also, at highway speeds there is less time for corrective action. I’ still not used NOA for this reason.

In city roads, for eg. If they introduce NOA, you can monitor whether the car is doing the right thing and intervene - is it slowing down before a stop sign or before a turn. Just the way you would know whether it picked the fact that the car in front is slowing in TACC.

 

[electronblue]

1) I get that it sucks for customers who had a lease and never got FSD but as the saying goes, "better late than never". Tesla does not have a time machine to go back and do things differently, so at this point, the best that Tesla can do is work hard and deliver FSD to customers going forward which they are doing.

The part I disagree with is how this all reflects on Tesla and how to read them. Of course we can not change the past but we can hold Tesla responsible — and most importantly we can use the past as a guide on how to take Tesla’s word this time around. Thus for example excitement of others may mislead people unless realistic expectations are factored in.

2) At the risk of sounding callous, leases are short term, 3 years. Folks who got a lease when AP2 was first released, had to know that it was a huge long shot to get FSD before their lease was up.

With FSD we were buying eight active cameras against four active cameras with EAP. This is how it was sold. Then in January 2017 Tesla’s CEO says we’ll definitely have differentiating features for FSD (something unique for us who bought that) by summer 2017. It is May 2019 and there are still no features unique to the FSD option we purchased.

Some of this is a choice by Tesla to give it out some features free to everyone instead. How could any of us have foreseen this in 2016? Well, in hindsight Tesla’s deception with P85D HP, P90DL performance figures and broken AP1 promises should have told us but Tesla used to still be a respectable company not so far back, back in 2014...

Finally, Tesla was marketing coast to coast FSD ride by the end of 2017. Back then we had no real information on how Tesla’s AP2 program was faring internally. We had the October 2016 video that was by that day’s standards no better or worse than the demos of April 22nd. Did we really have a realistic way of knowing nothing would be out by May 2019?

Even looking beyond that it still a significant time period in car ownership, that three years — especially with premium cars often used for business and frequently upgraded. Taking longer than that on shipping features clearly means a lot of folks will be missing out. They must have known that at Tesla too.

Tesla could have fixed this by offering refunds for early FSD buyers. Instead they offered discounts to everyone else on FSD and new cars further diminishing the value those early buyers put it.

Now, folks who outright purchased their AP2 cars will get the AP3 upgrade and will get FSD features. So they are good.

I don’t consider any of this good and I bought mine. But as I said I don’t think about the car itself anymore really. It is interesting to follow the news and discuss this on Reddit and here and elsewhere. Any excitement for Tesla has been obliterated by the acts of the company. The car itself is good and I have no reason to sell it or buy another, nor do I expect to.

It is what it is and, realistically, I don’t expect much from it. When something actually gets installed in my car I will use it and take if from there.

 

[diplomat33]

Also, someone else asked.. but I'm curious what features do you mean in the "auto city driving" that would not be apart of responding to traffic lights and stop signs?

Basically making any 90 degree turn like at intersections but also auto lane changes without confirmation on local roads to follow navigation. Like we saw a bit of in the FSD demo, Tesla is planning to use NOA on city streets too where we get the 1 blue line on city streets and the car follows your navigation direction, making turns to stay on the right road.