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

Autosteer on local Roads Videos (17.7.2)

This site may earn commission on affiliate links.
What part of the Tesla manual did you not understand?

Quote:
Autosteer is intended for use only on highways and limited-access roads with a fully attentive driver. When using Autosteer, hold the steering wheel and be mindful of road conditions and surrounding tra c. Do not use Autosteer on city streets or in areas where bicyclists or pedestrians may be present. Never depend on Autosteer to determine an appropriate driving path. Always be prepared to take immediate action. Failure to follow these instructions could cause serious property damage, injury or death.

It's like reading people complain that they can't use their toaster while taking bath... except the toaster only kills the idiot who fails to understand that electricity and water doesn't mix.
What part of the update did you not understand?

- What's new in this update:

"AUTOSTEER WHILE ON LOCAL ROADS"

Please, be careful with your toaster.
 
Watching these videos, it makes me think. What about that video Tesla it showing on their website? It shows the car driving so well. I get when they released the video they had an excuse that they needed government approvals, but they are now shipping AP2 functionality so approvals are in place, so why can't it work as well as the Tesla demo video? If I was to guess, the video was staged or hard-coded - software only works on that specific stretch of road, possibly remote controlled too.

I wondered this as well.
 
I wondered this as well.
As someone else noted before, it's fully possible/probable that the networks that actually perform the driving/steering aren't in place in production cars yet. That would mean they just use a rules-based architecture for actually driving and neural networks for, at this point, detecting cars and lane markers.

The reason to do so would be if the former works well in very particular conditions with some very bad failure cases when you get outside those conditions. Depending on their network architecture, it may not even be possible to split off segments of it to handle keeping in the same lane without also trying to follow navigation directions.
 
Which is??? Which way?

The instructions in the release notes, the manual, and that pop up every time you activate auto steering clearly say you should keep your hands on the wheel at all times and be ready to take over. The person in that video, instead, left the car to its own devices entirely and waited until it put them in an incredibly dangerous position(particularly with the oncoming car) before doing so.
 
What part of the update did you not understand?

- What's new in this update:

"AUTOSTEER WHILE ON LOCAL ROADS"

Please, be careful with your toaster.

The software update screen tells you that speed is limited to 35mph on local roads with AP2. Which is different to AP1, where speed on local roads does not suffer the same limitation. But that still not overrules the manual. AP is meant for highway driving. Take it out on local roads at your own risk.

I start to understand why all plastic bags in the US carries suffocation warnings...
 
The software update screen tells you that speed is limited to 35mph on local roads with AP2. Which is different to AP1, where speed on local roads does not suffer the same limitation. But that still not overrules the manual. AP is meant for highway driving. Take it out on local roads at your own risk.

I start to understand why all plastic bags in the US carries suffocation warnings...
So even though the release notes for our EAP update EXPLICITLY tell us that we now have autosteer on local roads up to 35mph, we shouldn't use it because the manual -- which was written before the update -- says that we don't have local road autosteer?

Take me to your leader.
 
Watching these videos, it makes me think. What about that video Tesla it showing on their website? It shows the car driving so well. I get when they released the video they had an excuse that they needed government approvals, but they are now shipping AP2 functionality so approvals are in place, so why can't it work as well as the Tesla demo video? If I was to guess, the video was staged or hard-coded - software only works on that specific stretch of road, possibly remote controlled too.
From my understanding, also based on the FSD info they had to release to CA, it appears those X's they used had a combined 500 miles of FSD likely on one or two routes, just to get that video.

The difference of how Nvidia does it's learning versus how Mobileye did theirs.
 
Just tried it this morning and yeah below beta quality. I really hope it's automatically sending feedback whenever you disengage it by grabbing the wheel, hell I really hope it's sending the feedback when it's not active and it would've made a different decision than the one you make.
 
pretty alarming video

Blader,
Thanks for posting this. WRT determining whether AP 2.0 is doing any on the fly "learning" or whether it only improves with subsequent releases, we need more videos like this. And what is specifically needed, is for someone to drive the same road under very similar conditions frequently, and post the resulting videos. Only then will there be any solid evidence as to the rate at which the software is improving. Empirical evidence is what is needed.

On the 5-6 threads here where people report on AP 2.0 issues, this kind of repeatable testing is sorely lacking. I'm surprised no one has started a thread with that level of testing in mind. It could be something as simple as recording your daily trip into work, pointing out where the AP 2.0 gets confused, and seeing whether it gets confused in the same place over time.

RT
 
  • Like
Reactions: NerdUno
Blader,
Thanks for posting this. WRT determining whether AP 2.0 is doing any on the fly "learning" or whether it only improves with subsequent releases, we need more videos like this. And what is specifically needed, is for someone to drive the same road under very similar conditions frequently, and post the resulting videos. Only then will there be any solid evidence as to the rate at which the software is improving. Empirical evidence is what is needed.

On the 5-6 threads here where people report on AP 2.0 issues, this kind of repeatable testing is sorely lacking. I'm surprised no one has started a thread with that level of testing in mind. It could be something as simple as recording your daily trip into work, pointing out where the AP 2.0 gets confused, and seeing whether it gets confused in the same place over time.

RT

I agree completely. I also thank you for posting this video. First, it gives me confidence that my car is not the only one that is operating so poorly on local roads. We all want Tesla to get their software right and feedback like this can only help - assuming anyone from Tesla is listening.

I completely disagree with the attitude that we are not supposed to provide negative feedback because the manual warns us to keep our hands on the wheel.

If Tesla gave us a mechanism to provide feedback, I would gladly take time from my day to provide it on a regular basis. As it is, we just have to hope they have an efficient means of gathering the data.

And lastly, I also agree with the posts that note that Tesla sold something completely different from what they have delivered so far. Since then, they have changed the language on their website which promised enhanced auto pilot in Dec 2016. With that said, I'm very happy with my car and I will continue to contribute if it helps.
 
I agree completely. I also thank you for posting this video. First, it gives me confidence that my car is not the only one that is operating so poorly on local roads. We all want Tesla to get their software right and feedback like this can only help - assuming anyone from Tesla is listening.

I completely disagree with the attitude that we are not supposed to provide negative feedback because the manual warns us to keep our hands on the wheel.

If Tesla gave us a mechanism to provide feedback, I would gladly take time from my day to provide it on a regular basis. As it is, we just have to hope they have an efficient means of gathering the data.

And lastly, I also agree with the posts that note that Tesla sold something completely different from what they have delivered so far. Since then, they have changed the language on their website which promised enhanced auto pilot in Dec 2016. With that said, I'm very happy with my car and I will continue to contribute if it helps.
If I am not mistaken you can leave feedback. Press the voice button and then say feedback (I think) and leave your feedback. No clue if they listen to it though.
 
And lastly, I also agree with the posts that note that Tesla sold something completely different from what they have delivered so far. Since then, they have changed the language on their website which promised enhanced auto pilot in Dec 2016. With that said, I'm very happy with my car and I will continue to contribute if it helps.

EAP "is expected to" not "promissed".
"Forward-looking statement" that might or might not happen in a predicted time.

auto-pilot-before-and-after-jpg.217303
 
For me, its scary to use AP2 on anything other than the following
- Stop go traffic on freeway
- Single lane highway with 50mph speeds

Just about all other scenarios, even with my hands on the wheel are a bit scary and nervous. The car drifts, darts unexpectedly and just feels nervous and unsure. There are so many times the lane markers have gaps, but instead of sticking to the line or following the straight line on one side, the car seems to get confused. I would not trust the car at 55mph on a freeway at this point, I have no idea why they are releasing that now when it feels so unstable and unsafe.

For stop and go in freeway during commute hours and traffic though, its really nice and works very well though so it is useful even in its current state, but the speed increases seem pointless when the overall quality does not seem to be improving as such.