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

No-confirmation NOAP is here for the masses!

This site may earn commission on affiliate links.
I'm now of the opinion that it's likely going to be a significant improvement with little in the way of actual danger. Of course it's not quite what was promised, but that was likely shutdown by regulatory agencies.

People keep talking about "regulatory agencies". There are none for autonomous driving. Pilot testing processes barely exist. There are no regulatory approval processes or guidelines in place under existing agencies like the NHTSA. It's mostly a law-free sandbox at this point. Tesla doesn't take new AP code to the State or Federal government and say, "Please review and approve these changes pursuant to regulatory process 451-2" or anything like that AND THEN push it out to customers.
 
People keep talking about "regulatory agencies". There are none for autonomous driving. Pilot testing processes barely exist. There are no regulatory approval processes or guidelines in place under existing agencies like the NHTSA. It's mostly a law-free sandbox at this point. Tesla doesn't take new AP code to the State or Federal government and say, "Please review and approve these changes pursuant to regulatory process 451-2" or anything like that AND THEN push it out to customers.
They're leaving it up to the states right now. There are in fact many regulations in California, Tesla's biggest market, regarding FSD. To release a level 3-5 system here Tesla would need regulatory approval.
I expect that at the April 22nd announcement Elon will announce that they have started the approval process. Haha.
Here are the rules: https://www.dmv.ca.gov/portal/wcm/c...essAV_Adopted_Regulatory_Text.pdf?MOD=AJPERES
Here is the application: https://www.dmv.ca.gov/portal/wcm/connect/64d2b714-0f76-442c-89c5-b2bc7de3fd2c/ol321.pdf?MOD=AJPERES
 
Last edited:
  • Informative
Reactions: diplomat33
Be careful....it thinks very "linearly". I've been trying it out.

For example:

If your exit is 2.1 miles away and you are in the middle lane.....and the car wants to change lanes to the left ( to speed up to match the speed you set ).

Then it will try to change lanes....even if it takes a mile to do so.

THEN.

It will say - "Hey....we need to get off the expressway in a mile. Let's start to immediately start merging back into the right hand lanes".

Other drivers will probably think something is wrong with you. - bouncing back and forth between lanes.
This is called greedy in algorithm industry.
 
Good to know. Though over time it seems like the longer-term logic will be fairly easy to add, i.e. don't need to go to the fast lane if exit is < x miles away and traffic is moving < y mph.
The correct way to solve this is to use "dynamic programming". Current greedy solution is to choose whatever lane appears faster now. While using dynamic programming, it makes decision based on total time to your exit. Say left lane appears faster, but changing the lane two extra times may cost more time than going in the left lane would saved.

The core difficulty here is to evaluate the lane changing time consumption based on how many cars the computer sees on the road. I believe it is doable as they have plenty of learning data.
 
  • Like
Reactions: woodisgood
The correct way to solve this is to use "dynamic programming". Current greedy solution is to choose whatever lane appears faster now. While using dynamic programming, it makes decision based on total time to your exit. Say left lane appears faster, but changing the lane two extra times may cost more time than going in the left lane would saved.

The core difficulty here is to evaluate the lane changing time consumption based on how many cars the computer sees on the road. I believe it is doable as they have plenty of learning data.
I look forward to improved decision making. When I've tried NoA in busy traffic, it almost always wants me to move into the passing lane. I know from my daily commute that at one highway entrance, the "passing" lane can be backed up for half a mile or more while the other lane is nearly empty up until the point where cars are actually zipper-merging onto the highway. I know to almost always stay in the right lane but NoA would send me to the passing lane and I would be stuck there.

On another nearby highway, it is three lanes and the center lane is usually slower than the right lane which is 4 miles long but eventually becomes an exit-only lane. NoA wants me to get in the left lane but then I would have to cross over two lanes to get to the exit-only lane to exit. NoA would have me in the (non) fast passing lane for 3 of those 4 miles but it's way faster to just stay in the exit-only lane for the entire 4 miles.
 
The correct way to solve this is to use "dynamic programming". Current greedy solution is to choose whatever lane appears faster now. While using dynamic programming, it makes decision based on total time to your exit. Say left lane appears faster, but changing the lane two extra times may cost more time than going in the left lane would saved.

The core difficulty here is to evaluate the lane changing time consumption based on how many cars the computer sees on the road. I believe it is doable as they have plenty of learning data.

Makes sense to me as the obvious way to go here. I'm not in software (got as far as Matlab in college, lol), but can appreciate the different types of programming at Tesla's disposal (as well as the wealth of data they can collect from the fleet) to solve some of these problems. I assume it's simply a matter of priorities and limited resources. Over time it will just get better and better.
 
The correct way to solve this is to use "dynamic programming". Current greedy solution is to choose whatever lane appears faster now. While using dynamic programming, it makes decision based on total time to your exit. Say left lane appears faster, but changing the lane two extra times may cost more time than going in the left lane would saved.

The core difficulty here is to evaluate the lane changing time consumption based on how many cars the computer sees on the road. I believe it is doable as they have plenty of learning data.

I really wonder if the geniuses over at Tesla don't already know how to think and program. Hmmmmm…. ( of course they do ).

I wonder if the geniuses over at Tesla have gotten us this far....hmmmm ( of course they have ).

I believe the solutoin might be something more difficult than what is stated.
 
I really wonder if the geniuses over at Tesla don't already know how to think and program. Hmmmmm…. ( of course they do ).

I wonder if the geniuses over at Tesla have gotten us this far....hmmmm ( of course they have ).

I believe the solutoin might be something more difficult than what is stated.


Or maybe the computer simply isn't powerful enough to solve the problem the correct way so they've had to hack together some kludgely solution that does not work well....

Of course if that were the case they'd certainly have a much more powerful computer lined up for the future... oh...wait....
 
Just got this update last night and tried it this morning. It's kind of amazing what a difference it makes not having to pull on the stalk. Yes you do need your hands on the wheel but it's a significantly different (and sort of amazing) feeling. That said I think Tesla really has to start watching other cars indicator lights and processing them (assuming it doesn't today). A large truck in the lane next to mine indicated to pull into mine, it had plenty of room but started pulling over very slowly. As a human I could see the indicators and anticipate what was going to happen but the car didn't. It should have slowed down a little or prepared to move into another lane but it kept going. By the time it would have realised the truck was moving into my lane it would not have had time to slow down (or would have had to break very hard). Otherwise.. amazing features
It sounds like you applied the brake yourself here. If so, thank you so much for training the NN. :)
 
People keep talking about "regulatory agencies". There are none for autonomous driving. Pilot testing processes barely exist. There are no regulatory approval processes or guidelines in place under existing agencies like the NHTSA. It's mostly a law-free sandbox at this point. Tesla doesn't take new AP code to the State or Federal government and say, "Please review and approve these changes pursuant to regulatory process 451-2" or anything like that AND THEN push it out to customers.

Then what exactly is Elon referring to when he says things "got regulatory approval", and things like "got some regulatory pushback"?

On his tweets I've asked, and many others have asked but he never clarifies by saying who. One of them happened when the US government was shutdown. So I doubt it was at the federal level.

In any case I think Tesla tries to come up with generic solutions that fit multiple regions whenever possible. Especially in cases where a regulatory concern brought up is not just necessary for that region, but a damn good idea.

One comment on reddit said the "hands detected on the steering wheel" requirement of NoA Auto-lane change was driven not by any push back by US based regulatory bodies, but European. I can't say if that was correct or not, but it seems plausible to me.
 
Last edited:
I wish there was a way to have it speed up a little bit when making a pass on the left. Lots of times the left lane traffic flow is going a lot faster than the middle lane. i.e. If I'm set for 68 mph and come up on a car going 67 it will make a past to the left, but only go 68. It can take a while to get around the car in the middle lane. (can be even worse if that person subconsciously speeds up a bit as you try to pass them, which is a major pet peeve of mine while driving... damn human nature) Meanwhile other cars in the left lane want to go 75.
Can't you just push on the accelerator pedal?
 
  • Like
Reactions: DR61
It was very strange as it has never had an issue here before. I had a number of new issues this morning and will pose them to the SC, but when it was working it was fine in regular areas. Just not sure why it kept kicking on and off.
It may be a faulty ultrasonic sensor or dirt obstructing one - the symptons would include having NoA switching on and off and inability to change lanes
 
I love it so far.

Thankfully it appears to follow the true rules of driving. ( unlike humans ).

For example: it will normally ONLY change lanes to pass on the left.

Anyone remember that from the "rules of the road"?


I don't believe EAP or FSD will be designed to "drive like humans" but rather to "drive like humans are supposed to drive".

I guess this would leave out pulling up on the bumper of some green weenie jackwad, clogging up the left lane at the speed limit in their Prius, and high beaming them while laying on the horn to get them out of the fast / passing lane when they are neither going fast or passing?

I understand rules of the road, but when you're dealing with an idiot, you got to change the rules a bit.
 
  • Funny
Reactions: jrad6515 and Richt
Today on NoA, the car never asked to change Lane when we were 0.2 miles from the exit. We had to change the Lane and exit. It is not very helpful.
FSD seems so far away.

Was the second right lane for exit AND continuing on the interstate?
NoA likes to stay in the left exit lane and doesn’t usually (there are exceptions) move into the “exit only” lane.

It works for me but I would sometimes move to a different lane; that’s what happens if you have two “drivers” driving the same car at the same time...