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

Software Update 2018.42.x

This site may earn commission on affiliate links.
I don't think the purpose of the confirmation is for lane change purposes.
It makes more sense for it to be for route purposes.
Few questions they should be able to answer based on the :
  • Did I suggest this lane change for taking an exit/following a route too early?
  • Did I pick the right of the two exit lanes?
  • Was the fastest lane suggestion really faster?
  • Is it a faster lane but there is another reason why the driver didn't accept it? (Bunch of trucks ahead on the faster lane)
I noticed that the distance from the exit where it asks you to change to the lane is not fixed.
It has varied from .9 miles to 2.5 miles

Sorry but none of the proposed questions can be answered by requesting acknowledgement of lane decisions made by the AI. Some will choose to override, some will accept no matter what and some will randomly choose between the two. As the human intervention cannot be counted on as required, nor the reason for the intervention determined, it cannot be used to determine optimal outcome, which is required to determine whether or not the AI decision was optimal and therefore used to improve the AI. You need to have a clear measure of the quality of the outcome for machine learning to work properly.
 
Sorry but none of the proposed questions can be answered by requesting acknowledgement of lane decisions made by the AI. Some will choose to override, some will accept no matter what and some will randomly choose between the two. As the human intervention cannot be counted on as required, nor the reason for the intervention determined, it cannot be used to determine optimal outcome, which is required to determine whether or not the AI decision was optimal and therefore used to improve the AI. You need to have a clear measure of the quality of the outcome for machine learning to work properly.
I think it will still suggest the lane change while someone is in your blind spot and it waits till they move before actually changing. DOn;t take my word for it though
 
If possible, both of you guys should go to a Service Center and connect to the Guest Wifi with your vehicle for a while. Their Wifi has caches of the latest software versions and force pushes the most current software to all vehicles connected. I did this last month while charging.

This is not true and hasn’t been in years.
 
Yes seems to, but was not always and the ding notification is similar to the warning to move the wheel while on AP so you end up looking down and trying to figure out why it is notifying you. Just adds stress to AP actually.


This is not really what I mean though. As you use it you will be checking to see if it wants to change lanes when you think it should a lot. This means looking down a lot.

YMMV, check it out when you get it.

I'm curious why you would look down to determine if a lane change should be made? Since it's clear you cannot see cars far enough behind you on the display to solely make the decision I always use the old fashioned "vision" method to determine whether it is safe or not. Now if you could see cars further behind you on the display I might consider rely solely on the display but until then not a chance I only look down. After all isn't the point right now to confirm this new EAP feature is working just as well or better then a regular driver? And a driver does this by looking out the window.
 
Last edited:
  • Like
Reactions: Nilnoc
I'm curious why you would look down to determine if a lane change should be made? Since it's clear you cannot see cars far enough behind you on the display to solely make the decision I always use the old fashioned "vision" method to determine whether it is safe or not. Now if you could see cars further behind you on the display I might consider rely solely on the display but until then not a chance I only look down. After all isn't the point right now to confirm this new EAP feature is working just as well or better then a regular driver? And a driver does this by looking out the window.

The proposed lane change that has to be acknowledge with Nav on AP is only displayed on the binnacle screen so if you want to acknowledge it you have to look to see what you are acknowledging I guess. You still need to check if the lane is clear, but without looking down you won't know which way the car plans to move and therefore which lanes to check. If they are not using the acknowledgement data, which I argue above is not useful, they should have left the feature in shadow mode and just track your decisions against the machine decisions without offering the machine alternative I guess.

In any case, if you try to use the feature as it is delivered you will find yourself looking down a lot to see if it is suggesting a lane change, but you could just ignore it completely as you suggest I guess.
 
  • Like
Reactions: MorrisonHiker
Sorry but none of the proposed questions can be answered by requesting acknowledgement of lane decisions made by the AI. Some will choose to override, some will accept no matter what and some will randomly choose between the two. As the human intervention cannot be counted on as required, nor the reason for the intervention determined, it cannot be used to determine optimal outcome, which is required to determine whether or not the AI decision was optimal and therefore used to improve the AI. You need to have a clear measure of the quality of the outcome for machine learning to work properly.

Ah, I was referring more to why are they asking for confirmation during lane changes. They are not using it for testing/verifying decisions of the lane change itself. Just the route portion or if the decision to go to a faster lane was correct. Can't really do that in shadow mode besides doesn't make a difference if its shadow mode or not might as well release it and let people also get the benefit out of it.
I have also been using it for a while and it suggests the lane change no matter what is on the lane.
Which is why I posted earlier that it seems the lane confirmations are more for route purposes and counting how many people are accepting it's suggestions vs ignoring/not doing anything, canceling.
 
Disagreed. The car would not even ask you to change lane if there is a car in your blind spot. Thus, your example is not a real life example.

Machine training is based on learning human behaviors under similar situations. Compare against the suggested behavior by the car is a critical step. Driver intervention is telling the car you don't like its suggestion. Repeat over and over again until a pattern is established. When your car prompts you to change lane, it means it believes it is safe. If you don't confirm, it should analyze the surrounding and try to understand why you don't agree...maybe it suddenly becomes unsafe, maybe too late, maybe the driver wants to go somewhere, etc. Or if you confirm and then you suddenly intervene because someone might have got into your lane at the last second, etc. So many possibilities. But it is a "teaching moment" for the car.

At the end, machine learning is not based on what you do. It is based on what vast majority of people would do under a specific scenario. This is exactly why Elon said it needed 10 mil driven miles before he can remove the lane confirmation restriction. Whether 10 mil driven miles is enough or not is up to the experts to decide. However, it is an unavoidable step. You may not like what it does at this time. New owners next year may love it because most of the problems may be resolved by then because we, the early owners, are part of the beta testing group. Remember, it is called Machine Learning....it takes time to "learn"

For those who don't want to be part of the journey, they really should not get EAP. We can go back and debate whether Tesla should have sold a unfinished feature, but it is a dead topic. It is what it is. We learned our lesson. At this point, we should wait for the feature to mature.

That's not machine learning. Theres no machine learning in the car. Come on people. Can we google before name dropping big words? Its embarrassing!
 
  • Like
Reactions: djlott and kavyboy
That's not machine learning. Theres no machine learning in the car. Come on people. Can we google before name dropping big words? Its embarrassing!
Where did he say anything about machine learning being done on car? The data you’re generating is being sent back to Tesla where machine learning is absolutely happening using the data you generated.
 
The proposed lane change that has to be acknowledge with Nav on AP is only displayed on the binnacle screen so if you want to acknowledge it you have to look to see what you are acknowledging I guess. You still need to check if the lane is clear, but without looking down you won't know which way the car plans to move and therefore which lanes to check. If they are not using the acknowledgement data, which I argue above is not useful, they should have left the feature in shadow mode and just track your decisions against the machine decisions without offering the machine alternative I guess.

In any case, if you try to use the feature as it is delivered you will find yourself looking down a lot to see if it is suggesting a lane change, but you could just ignore it completely as you suggest I guess.

Why do you need to constantly look down? Just wait for the alert tone and then look down. You’re overthinking this I think
 
  • Disagree
Reactions: PhilDavid
That's not machine learning. Theres no machine learning in the car. Come on people. Can we google before name dropping big words? Its embarrassing!

In a cloud connected device the machine learning element wouldn't typically be in the end device. It would be in the cloud where it gathered data from the entire fleet devices.

As to what this training is depends on what kind of model. Typically its Deep Learning stuff for classification/detection which is under the ML umbrella.

If my data from my car isn't being used for ML of some kind then I'd really like to know why Tesla is doing with all my data that my car consistently sends to them.
 
  • Like
Reactions: JohnnyG and Doctt
That's not machine learning. Theres no machine learning in the car. Come on people. Can we google before name dropping big words? Its embarrassing!
U are the one not knowing what u are talking about. Tesla is the one saying they are collecting data from the fleet. If it is not for machine learning, why do they need to collect tens of millions of miles of driving data before removing lane confirmation?
 
  • Disagree
Reactions: KArnold and BurntOC
That's not machine learning. Theres no machine learning in the car. Come on people. Can we google before name dropping big words? Its embarrassing!

I see no indication in that comment that he/she thinks learning is happening in the car. They specifically call out learning as happening from data aggregated across many users/cars, suggesting that wasn’t what they were saying.