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

2020.12.5.6: Traffic Light & Stop Sign Control

This site may earn commission on affiliate links.
OK, I guess it would be more precise to say “desired” rather than “correct.” In other words, the ideal behavior for a human driver when approaching a yield sign would be to check for traffic, then go ahead without stopping if the way is clear. In the context of this feature, that would be more like the behavior at traffic lights (go ahead if the driver gives confirmation) than stip signs (always stop completely before allowing the driver to confirm going ahead).

Plus, it seems dangerous to stop at a yield when there’s no traffic if there’s another car behind you!

Lets say that this feature doesn't drive like many of us. Ok. fine. Yield signs aren't even mentioned in the release notes. Can you imagine if it was? This thread would be another 25 pages of statements like "worthless", "I can't use this", "What is Elon thinking", "dangerous".

Folks, this is the absolute first iteration of Traffic Light & Stop Sign Control. Its not baked yet.

Does anyone remember when the first release of EAP? Now you can't even find a thread about it.
 
Last edited:
Aside from initial activation I am pretty sure TACC has never adjusted for the speed limit. That has been something that required Autosteer.

I believe this is correct, and has been correct for the entirety of TACC/AP. At least I've never seen it adjust for the speed limit. I've always had to either restart TACC or adjust the speed manually.

It's never been much of an issue because where I live the freeways/highways are pretty consistent in speed. They don't bounce up and down like the +/-5MPH rollercoaster that is I5 around Portland.

This is the first time where adjusting the set speed for the speed limit would be really handy. That way I could easily use this feature without much issue. Sure it doesn't always stop like I would, and it pauses for too long after the person in front of me has turned. It basically turns the accelerator into a "no, I want you to go" input device. Where a single momentary press is to tell it not to stop at the light coming up, and a longer press is to fix cases where its too slow. The only time its too fast is corners before the traffic light. It doesn't quite understand that you're not supposed to barrel around a blind corner up to a crowded stop light.

Personally I think the speed setting should be derived from both relative to the speed limit, and from a shadow mode that overrules that. That way it minimizes the "oh, no you don't moments of having to take over.
 
  • Like
Reactions: C141medic
Ok I finally got a chance to drive a decent amount to test. Here are my findings:

  • When approaching a traffic light that the car knows about well ahead of reaching it (presumably from map data), typically 600ft or so before the light, but before detecting the color of the traffic light, it presents the option to "Use accelerator or gear stalk to continue". In this case, this message appears with the symbol of a 3-light signal where all lights are off, but tapping the gear stalk or accelerator does dismiss the message.

  • When a light is detected as green, the same message appears and the traffic light symbol shows a green light. Dismissing the message with a tap of the accelerator or gear stalk allows the car to go through the traffic light, unless it turns yellow well ahead of reaching it. If no tap is given, the car starts to slow down about 200ft before reaching the light (I was going 35mph, the speed limit, on Autosteer at the time).

  • When a light turns from green to yellow after the car had already detected it as green, the car will continue through the intersection if a tap was previously provided for the green light AND the car is close enough to the yellow. I don't know the exact distance but I would estimate my furthest example was about 50ft from the light, again going 35mph. Otherwise, the car will brake for the yellow, and the message "Stopping for light control in XXXft" appears with no option to "Use accelerator or gear stalk to continue".

  • If a light is detected as red, the message "Stopping for light control in XXXft" appears with no option to "Use accelerator or gear stalk to continue". So tapping the accelerator or gear stalk obviously does not let the car go through the traffic light. However it is possible to hold the accelerator and go through the red light, but the car will yell at you loudly.

  • When approaching a stop sign or blinking red light, the message "Stopping for light control in XXXft" appears with no option to "Use accelerator or gear stalk to continue". So tapping the accelerator or gear stalk obviously does not let the car go through the stop sign or blinking red light when going over 5mph. HOWEVER when reaching 5mph and lower, a tap of the gear stalk or accelerator will allow the car to go through the intersection (albeit at glacial acceleration). So it is possible to "California-roll", for those curious :)

Some screenshots of the various messages in different conditions:


IMG_3074.jpeg



IMG_3078.jpeg



IMG_3076.jpeg



IMG_3085.jpeg
 
Lets say that this feature doesn't drive like many of us. Ok. fine. Yield signs aren't even mentioned in the release notes. Can you imagine if it was? This thread would be another 25 pages of statements like "worthless", "I can't use this", "What is Elon thinking", "dangerous".

Folks, this is the absolute first iteration of Traffic Light & Stop Sign Control. Its not baked yet.

Does anyone remember when the first release of EAP? Now you can't even find a thread about it.
I think many are just waiting for the re-write rather then continue to post what is wrong with EAP. At least that is my approach. I cannot honestly say anything has really improved for me in the last 6 months and if anything I use EAP less now. Hey, I'd just like to be able to summon my car into my garage on one try like it used to. Now it takes 3 tries so why bother.
 
I think many are just waiting for the re-write rather then continue to post what is wrong with EAP. At least that is my approach. I cannot honestly say anything has really improved for me in the last 6 months and if anything I use EAP less now. Hey, I'd just like to be able to summon my car into my garage on one try like it used to. Now it takes 3 tries so why bother.

I don't think people are waiting for a re-write at all. A re-write isn't a re-function. Everything is going to work the same way.



I use EAP every day..... as much as I can.
 
I don't think people are waiting for a re-write at all. A re-write isn't a re-function. Everything is going to work the same way.
I use EAP every day..... as much as I can.
The changes are certainly going to be important for improved functions and to think otherwise is silly. We already know it will include the stiching together in real time of a 360 degreee birds eye view to make better driving decisions. @20:00
Code 2 is referenced at 18:35. This talk is excellent and the reason for being patient and waiting for improved functions since so many of the problems I see haven't improved over many releases.
 
For me, this has worked pretty well...

The only major issue I have run into is blinking YELLOW lights...

In 50% of my cases, although the car has been given the OK to proceeded through, it will force a stop... Pretty dangerous in the area that I have this blinking yellow. This blinking yellow is on a stretch of road that has a 50mph speed limit.

I have been lucky to not have anyone behind me when approaching this light, so other than scaring the hell out of me, I have aborted the process when approaching this light...

Again, this behavior has only occured about 50% of the time and thus I have learned to work around it.
 
The changes are certainly going to be important for improved functions and to think otherwise is silly. We already know it will include the stiching together in real time of a 360 degreee birds eye view to make better driving decisions. @20:00
Code 2 is referenced at 18:35. This talk is excellent and the reason for being patient and waiting for improved functions since so many of the problems I see haven't improved over many releases.

Wow...this is a new record time in personal insults.
 
That would mean it was a tremendous waste of time and effort then.

It's FUNDAMENTALLY changing how the entire core system sees the world and will operate based on that.

Check out the video posted a couple back of the guy spearheading the effort

It will NOT be a waste of time and effort for Tesla.

I'm not speaking on behalf of Tesla when I say that it won't operate any differently. I write code for a living. Some of my re-writes updates ONLY affect how easy it will be for me to write code in the future. My customer won't see any changes at the time of my re-write. All of their functions are the same.

For example: If they give us a re-write TODAY ( right now ) do you think we will see any changes to what software we have purchased?

Do I disagree with a software re-write? no
Will a software re-write allow for easier changes in the software in the future? hopefully...it looks like it should / maybe.

That video almost exclusively talks about the benefits to AI....which may make things easier...in the future...but I don't know. Will it help produce the final version of FSD? that I purchased? hopefully.
 
Last edited:
For me, this has worked pretty well...

The only major issue I have run into is blinking YELLOW lights...

In 50% of my cases, although the car has been given the OK to proceeded through, it will force a stop... Pretty dangerous in the area that I have this blinking yellow. This blinking yellow is on a stretch of road that has a 50mph speed limit.

I have been lucky to not have anyone behind me when approaching this light, so other than scaring the hell out of me, I have aborted the process when approaching this light...

Again, this behavior has only occured about 50% of the time and thus I have learned to work around it.

I wonder if you’re giving it the ok to proceed when the blinking light is in “off” state?
Or, maybe the car thinks there is still time to stop at the line for this light, expecting the yellow will turn red.

If you happen to go through this light again, would you be able to safely take a short video of the screen?
I’m very curious whether the message says “Use accelerator or gear stalk to continue” when the light symbol is on.
 
I wonder if you’re giving it the ok to proceed when the blinking light is in “off” state?
Or, maybe the car thinks there is still time to stop at the line for this light, expecting the yellow will turn red.

If you happen to go through this light again, would you be able to safely take a short video of the screen?
I’m very curious whether the message says “Use accelerator or gear stalk to continue” when the light symbol is on.

Geeezzz dude....get the feature and try it.
 
It will NOT be a waste of time and effort for Tesla.

I'm not speaking on behalf of Tesla when I say that it won't operate any differently.

Clearly.


For example: If they give us a re-write TODAY ( right now ) do you think we will see any changes to what software we have purchased?

Obviously, yes.

The ability to understand a 360 realtime view would significantly enhance its ability to understand its environment and make correct decisions.

Know how right now sometimes your screen will show you a vehicle to the side that not only might jump around- but might change vehicle type?

That's because you have 8 different cameras all individually seeing some/part of a vehicle....and not all of it together in real time (since it was written for HW2 that can't process almost 1/3rd of the frames the cameras produce)... rather than 1 coherent view of the whole picture and vehicle that HW3 can enable via the re-write.

It's like the "describing an elephant" problem if you're familiar with it.

Likewise with a "whole picture" view I'd expect it'd immediately be better in things like figuring out what lane something's actually in and intending to be in and a number of other areas... (understanding what light goes with what lane better is another good one)

So yes I'd absolutely expect an immediate improvement in the software from such a re-write even on EXISTING features.



That video almost exclusively talks about the benefits to AI....which may make things easier...in the future...

Or also improve them right now. Which seems more likely.
 
Clearly.




Obviously, yes.

The ability to understand a 360 realtime view would significantly enhance its ability to understand its environment and make correct decisions.

Know how right now sometimes your screen will show you a vehicle to the side that not only might jump around- but might change vehicle type?

That's because you have 8 different cameras all individually seeing some/part of a vehicle....and not all of it together in real time (since it was written for HW2 that can't process almost 1/3rd of the frames the cameras produce)... rather than 1 coherent view of the whole picture and vehicle that HW3 can enable via the re-write.

It's like the "describing an elephant" problem if you're familiar with it.

Likewise with a "whole picture" view I'd expect it'd immediately be better in things like figuring out what lane something's actually in and intending to be in and a number of other areas... (understanding what light goes with what lane better is another good one)

So yes I'd absolutely expect an immediate improvement in the software from such a re-write even on EXISTING features.





Or also improve them right now. Which seems more likely.

I have an issue believing Tesla's timeline as they state it. I don't have an issue believing that Tesla "will" do what they say, but they don't have a great track record in the "right now" category.

FSD is supposed to have been "by now" already, however I ain't mad at them. I'm happily waiting.

I hope you are correct about the "right now"....however its based on Tesla....which reduces my confidence in what you are saying.
 
I have an issue believing Tesla's timeline as they state it. I don't have an issue believing that Tesla "will" do what they say, but they don't have a great track record in the "right now" category.

FSD is supposed to have been "by now" already, however I ain't mad at them. I'm happily waiting.

I hope you are correct about the "right now"....however its based on Tesla....which reduces my confidence in what you are saying.



To be clear- I don't think the re-write will enable huge leaps in NEW functionality immediately.

But they pretty obviously ought to allow significant improvements in EXISTING functionality just based on the car having a lot better picture of what's actually going on around it immediately.
 
Here are my observations with trying this feature using TACC.

  • The stops are half way decent. They could definitely slow down a little sooner, and smooth out the stopping.
  • It doesn't automatically change the TACC set speed when you go from one speed limit to another. This significantly hampers this feature, and makes it way harder to use than it really needs to be.
  • If you're behind a vehicle that stops for a stop sign it just leaves you sitting there. This creates a UX/UI issue because in this circumstance I just want to move up to the sign, and not go through the stop. If I recall correctly I hit the accelerator and then stopped manually.
  • It doesn't accelerate from a stop nearly as quickly as it should. So using it from a stop is a bit awkward, but I do acknowledge it's worse for me than the average person since I tend to take off quick. That was kind of the point of getting an EV for me.
  • Simply taping, and releasing the accelerator lightly makes the upcoming light notification go away. The only annoying parts were where the green light decided to go yellow just after doing this. I didn't test to see if the car would override my confirmation due to the lights changing color.
  • I wish that the car could learn from my driving behavior what speed I liked going on certain roads. I'm not someone who always goes 5mph over or 10mph over. It depends on what type of road it is, and where it's at. Like in my residential area its 25mph, but I never go above 25. There are too many kids, and it's too condensed. So I'm typically going 22 or less. Then outside of my residential area it's 35mph where I travel 40mph before encountering a 25mph section. Now this 25mph section used to be a 35mph section, and you'd never know it was 25mph if it wasn't for massive signage. There is also a lot of enforcement so lots of people do 25mph, but I do 35mph. Even on my pedal assist bike I'm doing 30mph, and it's actually my favorite road since it's fun to laugh at the cars stuck behind someone doing 25mph while I'm doing 30mph in the bike lane just up ahead. Although when that happens I get into the lane due to wanting to be further away from debris in the bike lane. Normally it's pretty clean as it's near rich people (hence getting it lowered to 25mph).
All in all with some improvements this feature will be pretty nifty. Now I'm not sure what point there would be in city TACC, but hey at least it's something new and interesting.
 
So, just returned from a 350 mile trip and had the opportunity to test out the new FW update. My 2 cents:

Traffic Lights:

1. For majority of traffic lights it worked perfectly and as others have described.

2. For more complicated intersections, with multiple traffic lights in close proximity to another intersection with traffic control, the car seemed to balk a bit. But I think this was mainly due to having to confirm for each traffic control. Once the confirmations are no longer needed I think this shouldn’t be an issue.

3. Turning from a traffic light-this was not so smooth in my opinion. So, I was stopped at the light and when it turned green, I lightly tapped on the accelerator and rather than turning at a safe pace it immediately began to increase speed to match the speed (40MPH) of the road I turned into.

4. This was all using TACC. I didn’t test the FW feature using auto steer. For all traffic lights, I used a light tap on the accelerator rather than the stalk to confirm.

Stop Signs:

In my opinion, this is not quite ready for prime time at the current level of the FW (yes I’m aware it’s BETA) and requires more attention to detail than driving through traffic lights. There were a few issues with this that I experienced.

1. It’s stopping way too early at stop signs. I experienced at least 10-15 feet before. Because of that I wasn’t able to see oncoming traffic (due to not being able to see past blind spots since it stopped so far behind the sign). So the only way to fix that was to move forward. However, if you move forward via tapping the accelerator to confirm, the car continues past the stop instead of actually stopping again once you can see past the blind spots. This really needs to be fixed sooner than later. Here’s a pic I took referencing this:
View attachment 538256

It was late at night so I experimented with it a little since there was no one else on the road. Immediately after this screenshot, and while on TACC set to about 10MPH, I tapped the accelerator and I turned left. The car turned left and then came to an immediate stop right in the middle of the turn. Very strange.

In summary, this is an exciting FW to test because it’s a prelude to more refined features that I’m sure will be forthcoming. I for one will NOT continue to use/test it at stop signs at the current version of the FW. Perhaps if there’s some improvement later on, I’ll try it. I will however continue to use it on major roads where there are only traffic lights.
Great post. Another interesting thing I found was that if I was going over a flyover (short bridge/overpass over train tracks), where a stopsign exists on a parallel street below, it thought there was stopsign up top. This is presumably due to the fact that technically the 2D gps location shows a stopsign is there (and along my direction), where a piece of height datum and reading would show it was not.

There was no stopsign visible, but it showed one on-screen exactly where it would have had I been on the street below. The car slowed a very slight bit upon approach, but did not stop. Their weighting algorithm is very good and seemingly uses gps as a suggestion, a visual que as primary, and final user input upon visual confirmation to stop/go. I didn’t have to do anything in this case as it caught itself, quickly, but very cool

I submitted a voice bug report (push the voice input wheel/button) “bug report, there is a stopsign below this bridge, but none up here on top” - not sure if the ai was updated or the car just knew to ignore it later on, but this was fixed after about 3 days. Absolutely blew my mind.
 
Last edited: