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

MASTER THREAD: 2019.40.2 - FSD AutoSteer Stop Sign Warning and Adjacent Lane Speeds

This site may earn commission on affiliate links.
OK, guys, chew on this. I hadn't had any pingponging or wandering at all with 2019.40.2.1

BUT I always drive with a small counterweight on the right side of the wheel. 150g, say 5 Oz of BBs in a lil'bag held with velcro at the right cross-member. That's not enough to defeat the nags. I started doing this because it lets me satisfy the hands on wheel torque request by lightly moving my hand the way I hold it on the right. Without the counterweight I was too easily jerking out of AP, and that was just plain unsafe for me.

Today, for the obvious test, I removed it. With no counterweight the ride was entirely different - all sorts of drifting in the lane and other problems. After I put it back, everything is fine again. It's a striking difference. Fellow tweaks can try it and report.

It's possible that people who haven't seen drifting are unconsciously applying a little pull to the wheel the whole time. So that the "sometimes it's better/worse" reports come from how we hold the wheel?

bag_20191217_185243s.jpg
 
Last edited:
  • Helpful
Reactions: scottf200
I always drive with a small counterweight on the right side of the wheel.
Not sure what you're advocating here. So... I typically drive with enough torque on the wheel to make sure the car knows I'm engaged in the driving task. But not so much as to override AutoSteer. But somehow adding just a little bit more torque tames an inherently underdamped steering controller? If this is true, how would I exploit this to make AP handle better? Shouldn't we just submit bug reports on the underperforming system so that Tesla can fix it?
 
I have a question for discussion.

What is the purpose of having the traffic light and stop sign "warning" as opposed to simply showing us the stop sign or light when they are detected?

Now we have people attempting to go thru red lights and stop signs to "test" this feature. Dangerous for sure, and not desirable imo.

My suggestion is to simply have the stop and sign and/or traffic light display on the screen when they are detected. If one of them shows up when a sign or light is not viewed by the driver, they could flag a bug report. When a sign or light exists and does not display, a bug report could also be sent.

Would not that providde more useful data for training the neural net, and be far less dangerous?

What say you folks? Thanks.
 
  • Like
Reactions: jedi2b
I have a question for discussion.

What is the purpose of having the traffic light and stop sign "warning" as opposed to simply showing us the stop sign or light when they are detected?

Safety- it's providing a (sometimes works) alert you're about to make a dangerous mistake.


My suggestion is to simply have the stop and sign and/or traffic light display on the screen when they are detected. If one of them shows up when a sign or light is not viewed by the driver, they could flag a bug report. When a sign or light exists and does not display, a bug report could also be sent.

Would not that providde more useful data for training the neural net, and be far less dangerous?

Given bug reports stay on the car and are only reviewed if you schedule the car for service- no.

Instead the car alerts when it thinks you're about to run a light/sign- and the code can be programmed to upload to tesla when it DOES think it detects one but you don't stop for it- so they can check if they got a false positive.
 
Are you in city or on highway? Are you using NOA,Nav without autopilot, cruise control and/or auto-steering?
I had the alerts on local roads when using just AP. Stop signs detection if very rare and I only been able to get the alert twice when I tested it by slowing down the AP to about 3mph as I was approaching a stop sign. It only alerted as the car was about to cross past the sign, so at this time it's is pretty much useless.

The traffic red light works a little better, but still needs more ML training.
 
Given bug reports stay on the car and are only reviewed if you schedule the car for service- no.
Do we know for a fact* that this is the case? There have been several instances now of me "Bug Report"ing incorrect speed limits that appear to have had been corrected after I started reporting them. Others have reported same. I'll have to start monitoring my car's data usage when on my wireless network to see if outbound traffic goes up after using that tool.

* A SC tech claiming this is definitely not proof of 'fact', as there are a zillion cases of SC techs giving blatantly and clearly incorrect "information" when asked about myriad things.
 
  • Like
Reactions: MP3Mike
Do we know for a fact* that this is the case? There have been several instances now of me "Bug Report"ing incorrect speed limits that appear to have had been corrected after I started reporting them.

And 100 times more cases where they weren't corrected- Suggesting any where they were were coincidence of just getting fixed in a normal map update.

A moments thought will tell you there's no way Tesla who doesn't have enough people to even answer phones has enough people to be listening to every bug report from a fleet of what's rapidly approaching half a million cars- and likely to approach 1 million in another year.[/QUOTE]
 
I have a question for discussion.

What is the purpose of having the traffic light and stop sign "warning" as opposed to simply showing us the stop sign or light when they are detected? To alert the driver.

Now we have people attempting to go thru red lights and stop signs to "test" this feature. Dangerous for sure, and not desirable imo. Agree

My suggestion is to simply have the stop and sign and/or traffic light display on the screen when they are detected. If one of them shows up when a sign or light is not viewed by the driver, they could flag a bug report. When a sign or light exists and does not display, a bug report could also be sent. If driver's not responding, an alert should be shown. A bug report would be after the fact and not help the driver at the time.

Would not that provide more useful data for training the neural net, and be far less dangerous? What?

What say you folks? Thanks.

See above
 
Please explain and describe "ping-pong".

Constant subtle but noticeable corrections left and right between the lane. It used to drive dead center straight in the middle of the lane, now it constantly swerves left and right and if you hold the wheel just tight enough you feel the constant corrections in both ways.

I suspect this is something to do with side sensors and trying to give more clearance away from a wall or big truck. I noticed red side bumping animations when it is doing this. Especially noticeable on hov lane with divider close to the left side.
 
I've been driving with 40.2.1 for 4 days. I've used it with and without NOA in city and on highway. I have yet to see a Stop Light or Stop Sign alert. I still have HW 2.5, is that why?

I tried autosteer on a residential street with 20 speed and it clearly displayed a red stop sign under the speedometer and started beeping until I hit the brakes. I have HW 2.5 too.
 
I have finally been able to fully test the new 40.2.1 and I have to say in some ways it is a step backwards. For informational purposes, I live in central Texas and periodically travel fm 2657 which is a windy road leading down to Austin. Let me say from the start that at least in my experience, my tesla M3 LDDM does not and has never read speed limit signs or stop lights. Nor does the new 40.2.1 read stop signs. It appears that these functions are driven by Google maps. If Google does not show it, apparently tesla does not know it. As for my last trip, the car no longer slows down on AP for sharp curves and I experienced two phantom breaking going up hill with no traffic coming.
it has to be said clearly that I really like my M3. I love using the safety features as I am an old man. But as I discovered as a programmer when young, when trying to fix one problem, you may very well create unattended problems elsewhere. Be safe out there. Use all the safety features but do not rely on them. Keep aware.
 
  • Like
Reactions: Kabillyhop
And 100 times more cases where they weren't corrected- Suggesting any where they were were coincidence of just getting fixed in a normal map update.

A moments thought will tell you there's no way Tesla who doesn't have enough people to even answer phones has enough people to be listening to every bug report from a fleet of what's rapidly approaching half a million cars- and likely to approach 1 million in another year.
I wouldn't even begin to suggest that my specific bug reports are being looked at by a Mk. I Meatbag in the Tesla Mothership every time I make one.

But it's absolutely possible that every BR is dumped into data aggregation, and if they see a "hot spot" of bug reports all sent in at one approximate geolocation, that batch could be scrutinized for trends.

All I'm saying (asking, really) in my above post is- do we know that bug reports are only fetched pursuant to a Service engagement? How do we know this? If the data are uploaded via the car's LTE connection, how would be know? If they're uploaded when on a Wifi network, the administrator could definitely see whether the car is uploading more data than are typically sent.
 
Not sure what you're advocating here. ... But somehow adding just a little bit more torque tames an inherently underdamped steering controller? If this is true, how would I exploit this to make AP handle better? Shouldn't we just submit bug reports on the underperforming system so that Tesla can fix it?

A car under autopilot acts as one very complex servo-system, and like all servo-systems it is affected by every component, often in dramatic and unexpected ways. What I'm advocating is

1) Test whether many of us find that a 5 oz counterweight on the right side of the steering wheel stabilizes what in this version behaves as "an inherently underdamped steering controller". Let us know here so we can REPORT TO TESLA so they know the number of such cases as they work to better tune the servo and correct the problem.

2) For those who do confirm this, adding a little weight to the steering wheel might well be a practical and harmless temporary workaround until TESLA corrects the problem.

For me anyway, it sure beats this:

puke-gif.488829


As to the stoplight and red light detection, FWIW, I have HW2.5 EAP and FSD, and I've never ever seen it show up or do anything.

Regen braking in HOLD mode worked pretty well in version 36, allowing single pedal driving. I has indeed changed to the point where approaching a stop in manual from e.g. 25 MPH now ALWAYS requires some use of the brakes.
.
 
Last edited:
  • Like
Reactions: scottf200
Constant subtle but noticeable corrections left and right between the lane. It used to drive dead center straight in the middle of the lane, now it constantly swerves left and right and if you hold the wheel just tight enough you feel the constant corrections in both ways.

I suspect this is something to do with side sensors and trying to give more clearance away from a wall or big truck. I noticed red side bumping animations when it is doing this. Especially noticeable on hov lane with divider close to the left side.

I just got back from a long drive (over 5 hours on NOA) and there was a very noticeable ping pong effect going on. It was really annoying.