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

Tesla Begrudgingly “Recalls” FSD Beta for NHTSA

This site may earn commission on affiliate links.
I'm sure this will be a sticky on all of the vehicle forums shortly:


(moderator note: related threads here…)
FSD Recall? in Software
Recall FUD in Uk

46071715365_d36a6e2bf4_b (1).jpg

"Full Self Driving Tesla" by rulenumberone2 is licensed under CC BY 2.0.
Admin note: Image added for Blog Feed thumbnail
 
Last edited by a moderator:
Beta has been speeding up to stop signs and trying to run though them.

This is latest version in areas that beta has been running since first public beta.

Residential area sees stop signs and starts speeding up towards sign well past speed limit. I have to disengage and hard break. Any attempt to reengage will result in the same thing over and over.

In other areas it’s functioning normally.

Not sure what’s going on. New map data messing things up perhaps.
 
NHTSA has found a few things they want fixed, end of story, but it's a big story "because it's a Tesla".
And no, nobody knows what the V11 update will include, let alone the recall issue.
Too much speculation on here...
I very much expect it's just beginning of story. The NHTSA still has numerous investigations and engineering analysis open into AP and FSDb is rife with bad behavior that they could pick on. If they are serious that is.
 
If they can't "fix" FSDb to satisfy the requirements of the recall then they can't release FSD to any HW3 buyers. Plus there are many other bad things about FSDb that aren't even addressed in this recall. Maybe there will be other recalls issued to address those bad things too.

However the recall doesn't really say anything about how good it has to be, it's a pretty lame recall from the NHTSA in terms of setting any standards. Even Tesla just says they are going to "improve" it, they don't say they are going to fix it. So who knows if the NHTSA is going to say great, at least you're trying, or are they going to be more harsh and say no, it's still doing those bad things.
I don't see the recall as a big deal, it's a few relatively minor things that even NHTSA does not think will take much to fix. Again, I see this akin to the rolling stop recall. Some people here characterize this recall as if NHTSA is telling them to scrap the whole thing or as if it requires "finishing" FSD to address, when that is very much not the case.
Tesla could back down on the scope of FSD and just go with a basic L2 system that meets with the approval of the NHTSA, which IMO is all they will ever get from HW3. Or, they can keep trying with FSDb, but I'm still not sure what they are hoping to achieve anyway. If they make something that has many of the apparent features of an L4 system but is called an L2 system then that's just going to encourage people to drive with blind reliance on a system that is not intended for that.

Well, let's see what 11.3 does. [Which doesn't really seem like it's addressing the recall, that is just another update. Are they really making the update address the specific terms of the recall? Again, I guess we'll see.]
A door-to-door L2 would be able to do all the functions of a city L4 system, but it would still require the driver to pay attention and be ready to take over at any time. Tesla's not the only one that will have it, I've posted other examples of upcoming systems (Mobileye Supervision L2 in Zeekr 001, GM's Ultracruise, XPeng City NGP). Some of these may even have hands-free operation (given many of them have infrared camera monitoring). As long as there is a robust driver monitoring system, and the agreement for activation makes it extremely clear the driver must be ready to take over, then it shouldn't be an issue.
 
  • Like
Reactions: GSP
This is absolutely just the tip of the iceberg RE; FSD and NHTSA actions..
Exactly

The behaviors NHTSA called out are behaviors that have existed with FSD and been complained about by users since the beginning of time.
If they were so easy to fix, they'd have been fixed.
Some of them get fixed an un-fixed in subsequent releases.
Some of this is the nature of AI/ML models and tuning parameters / probabilistic behavior.

There's a laundry list of other suboptimal behaviors that NHTSA can also pick on.
Tesla largely brought this on themselves by removing the test gate to FSD while continuing to call it beta, so theres now far more users with it on the road.
 
  • Informative
Reactions: 2101Guy
Does anyone know what the burden of evidence is for the NHTSA? Do they have a car with FSDb that they have taken to a particular place and verified its behavior or is it just based on reports? Even with my own car, I have trouble reproducing a particular behavior due to traffic or lighting or weather being different. I have a hard time believing they have a constant repro and can go to Tesla and say this is dangerous, but I have a bigger problem if they issue a recall based on something someone said happened once etc.
 
Does anyone know what the burden of evidence is for the NHTSA? Do they have a car with FSDb that they have taken to a particular place and verified its behavior or is it just based on reports? Even with my own car, I have trouble reproducing a particular behavior due to traffic or lighting or weather being different. I have a hard time believing they have a constant repro and can go to Tesla and say this is dangerous, but I have a bigger problem if they issue a recall based on something someone said happened once etc.

One report isn't a problem.

There isn't any pattern of bogus reports triggering recalls. They're not dumb. Even the big unintended acceleration cases only triggered a recall when they found an actual small defect.

This isn't about FSD mistakes. It's about how it executes maneuvers. It believes that FSD executes some maneuvers dangerously and wants them changed.
 
I think a fun test would be if each state DMV took an FSD Tesla through their license road test and scored it like a human.
Given the probabilistic nature of FSD behavior, it's a coin toss if any single shot test would pass or fail.

So let's say the DMV could require it pass say 7 out of 10 road tests.
In most states, FSD would fail this arguably low bar.

"But it's just ADAS / they don't do this to other makers!"
Sorry but Musky can't have it both ways.

Either its FULL SELF driving, and works everywhere including on city streets, with no explicit feature bounds.. costs $15k, and is miles ahead of the competition who he can't see in the rear view even with a telescope..

OR - it's just ADAS, and should have explicit bounds of how/where it works (divided highway only, handling only lane keeping / lane changing / TACC), be labelled as assist, not be regulated, and cost something like $1-2k like everyone else.
 
I think a fun test would be if each state DMV took an FSD Tesla through their license road test and scored it like a human.
Given the probabilistic nature of FSD behavior, it's a coin toss if any single shot test would pass or fail.

So let's say the DMV could require it pass say 7 out of 10 road tests.
In most states, FSD would fail this arguably low bar.

"But it's just ADAS / they don't do this to other makers!"
Sorry but Musky can't have it both ways.

Either its FULL SELF driving, and works everywhere including on city streets, with no explicit feature bounds.. costs $15k, and is miles ahead of the competition who he can't see in the rear view even with a telescope..

OR - it's just ADAS, and should have explicit bounds of how/where it works (divided highway only, handling only lane keeping / lane changing / TACC), be labelled as assist, not be regulated, and cost something like $1-2k like everyone else.
You are setting a false dilemma. There is zero indication NHTSA is treating this as anything other than a L2 feature.

There is no limit to how high you can set the cost of an L2 feature. For example even though on paper the Super Cruise option was only $2500, the minimal options to get it was $8,740 (including $2500 for a different set of driver assistance package that is prerequisite to it).

 
Full Driver Assist would have been a more fortunate moniker, for sure, with the promise of a regulated and complete FSD at a not-defined future date.
The moniker is a huge part of the problem and perception.
Correct. But calling it FULL SELF DRIVING along with making repeated false promises about what it could and would do, and by when, resulted in many millions more $ to Tesla via persons paying thousands of dollars for the promises than if the statements would have been truthful and calling it Driver Assist.
 
  • Like
Reactions: COS Blue and sjg98
NHTSA has found a few things they want fixed, end of story, but it's a big story "because it's a Tesla".
And no, nobody knows what the V11 update will include, let alone the recall issue.
Too much speculation on here...
Its a story because these are things that have been issues since early releases of FSD, but Tesla chose to concentrate on fixing issues highlighted by youtube "influencers" instead. The problem was that they didn't think they were important. They seem to be prioritized on it being an interesting problem, not because it causes issues for people driving. Heck the latest releases have removed the feedback option.
Now an influencer of of the regulatory kind has given them a list of things to fix - its just that they should have been fixed months/years ago.

Regarding speculation . . .
You know you're on TMC right? We are the epicenter of rampant, completely biased Tesla speculation, its our superpower :cool:
 
Does anyone know what the burden of evidence is for the NHTSA? Do they have a car with FSDb that they have taken to a particular place and verified its behavior or is it just based on reports? Even with my own car, I have trouble reproducing a particular behavior due to traffic or lighting or weather being different. I have a hard time believing they have a constant repro and can go to Tesla and say this is dangerous, but I have a bigger problem if they issue a recall based on something someone said happened once etc.
they only have to read through this forum. FSDb is pretty darned good at some things and comedically bad at others.
The main problem seems to be the prioritization of issues isn't based on providing a good experience for owners, who knows what the goal is.
One thing we know is that it isn't L5 robotaxi autonomy any longer.
 
  • Like
Reactions: 2101Guy
Given the probabilistic nature of FSD behavior, it's a coin toss if any single shot test would pass or fail.

So let's say the DMV could require it pass say 7 out of 10 road tests.
Lol.

More like 0 for 10. Not really a coin toss, at all. Every. Single. Time.

Have to be very careful about overstating FSD capabilities. I may be being too generous here.
 
  • Like
Reactions: sjg98 and Twiglett
Lol.

More like 0 for 10. Not really a coin toss, at all. Every. Single. Time.

Have to be very careful about overstating FSD capabilities. I may be being too generous here.
we have to be careful with that capabilities word, it might imply more or less than people think.
Roll on the continued mandates to fix things, the more the better, we might get something that works then.
 
[rambling thoughts]I remember reading (and my memory is well....... 🤔 ) that Tesla must notify about the Recall on April 25 (or sometime around this). In the past most OT Recalls have been done without a letter being sent. Wonder if the stipulation is that if the Recall is "fixed" and completed before the notification date (April 25) that Tesla (or any auto maker) can avoid sending out letters. This would actually save a chunk of change since 400k is a huge costs.
 
[rambling thoughts]I remember reading (and my memory is well....... 🤔 ) that Tesla must notify about the Recall on April 25 (or sometime around this). In the past most OT Recalls have been done without a letter being sent. Wonder if the stipulation is that if the Recall is "fixed" and completed before the notification date (April 25) that Tesla (or any auto maker) can avoid sending out letters. This would actually save a chunk of change since 400k is a huge costs.
While it would reduce paper, that does not seem to be the process.
How Will I Be Notified If a Recall Is Ordered or Initiated? Within a reasonable time after the determination of a safety defect or noncompliance, manufacturers must notify, by first-class mail, all registered owners and purchasers of the affected vehicles of the existence of the problem and give an evaluation of its risk to motor vehicle safety . The manufacturer must explain to consumers the potential safety hazards presented by the problem . Names of vehicle owners are obtained from State motor vehicle offices . The letter must also instruct consumers on how to get the problem corrected, remind them that corrections are to be made at no charge, inform them when the remedy will be available, how long the remedy will take to perform, and whom to contact if there is a problem in obtaining the free recall work . If you do not receive a letter of notification from the vehicle manufacturer but think that your vehicle might be involved in a recall campaign, call the Vehicle Safety Hotline at 888-327-4236 or 800-424-9393, visit the NHTSA website at www .safercar .gov/vin or contact the manufacturer or your dealer .
49 CFR § 577.5 - Notification pursuant to a manufacturer's decision.