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

Poll : What is your disengagement rate with FSD Beta ?

What is your estimated disengagement rate with FSD Beta ?

  • > 1 per mile

    Votes: 12 16.9%
  • 1 every 1 to 5 per miles

    Votes: 31 43.7%
  • 1 every 6 to 10 per miles

    Votes: 14 19.7%
  • 1 every 11 to 20 per miles

    Votes: 8 11.3%
  • 1 every 21 to 50 per miles

    Votes: 3 4.2%
  • 1 every 51 to 100 per miles

    Votes: 3 4.2%
  • 1 every > 100 miles

    Votes: 0 0.0%

  • Total voters
    71
  • Poll closed .
This site may earn commission on affiliate links.
Is turning at the wrong place, e.g., into a parking lot or wrong street, considered "uncomfortable?" FSD Beta can still be driving safely but not immediately making progress to the destination.
 
Is turning at the wrong place, e.g., into a parking lot or wrong street, considered "uncomfortable?" FSD Beta can still be driving safely but not immediately making progress to the destination.
Personal Preference I would say. If you find it uncomfortable, and you jerk the wheel to follow the proper street, then yes, mark it as a disengagement. But if you don't mind car taking a detour and taking slow path to your destination, then you don't have to mark as a disengagement.
 
In the city it is at least 1 per mile unless I pick an easy route with minimal construction, congestion and edge cases. If I go out to the burbs I can improve that probably close to 10x.
Yea I think it depends a ton on where and what time of day you're driving. In Burbs with mixed city/highway, i've done 20-30 miles without disengagement. But all city streets in burbs, maybe 1 for every 5-10 miles, and i'm sure if I went up to SF, it would be unusable haha.
 
But, do not include interventions like pressing the accelerator or changing speed limit.
What about pressing the accelerator because the car insists on creeping through a perfectly clear intersection and you have a driver on your tail behind you? At a minimum this is uncomfortable driving, but in some cases I think it is dangerous and need to press the accelerator to get the car to actually get going.

I do agree with not counting using the accelerator because the car is simply slow to accelerate though.
 
In the city it is at least 1 per mile unless I pick an easy route with minimal construction, congestion and edge cases. If I go out to the burbs I can improve that probably close to 10x.
Same experience in the Philadelphia area.

Suburbs and rural is roughly 1 disengagement every 10 to 20 miles. In Center City Philadelphia it is a disengagement every 2 to 3 blocks. Strong shadows from tall buildings on a sunny day confuse it along with tight streets with parking, outdoor dining and construction on both sides.

On a positive note, FSD 10.9 has very few disengagement on city streets with dedicated clearly painted bike lanes.
 
It’s actually a little difficult to quantify. I disengage when the car is going to cause other drivers problems. If it’s too careful or too slow to react etc. and, of course when it does something dangerous or stupid. It’s far from ready for prime time regardless of what some think.

I use it every time I drive to one extent or another, but long strait stretches are a joy and where it shines. Traffic and complex actions are pretty abysmal.
 
For comparison, here is the disengagement rate of the top 10 companies in California.

2020 Calif. Self-Driving Car Disengagements​

Company2020 Miles Driven2020 Disengagements2020 Miles/Disengagement2019 Miles/Disengagement
Waymo628,8392129,944.6913219.43
Cruise770,0492728,520.3412221.17
AutoX40,734220,367.0010684.67
Pony.ai225,4962110,737.906475.75
Argo AI21,037210,518.59N/A
WeRide13,01426,507.00151.72
DiDi10,40125,200.75N/A
Nuro55,370115,033.622022.41
DeepRoute10,01833,339.33N/A
Zoox102,521631,627.321595.6


 
  • Informative
  • Like
Reactions: t3sl4drvr and EVNow
For comparison, here is the disengagement rate of the top 10 companies in California.
Poor comparison. We are not talking about carefully selected few miles of roads with Lidar & HD maps etc. They don't have the technology to scale to the size Tesla is aiming for. Basically a different market.

Let me know if you have any comparison to a company that has an ODD of all mapped roads in US.
 
Last edited:
I don't think we can compare to other services because they are geo fenced to certain "blessed" areas. If I only engage FSD in places/conditions that I know it works well, it would be near 100% perfect. Tesla's approach works everywhere so we encounter many more and much harder conditions.
 
When I drop my kids at school (my most frequent FSD trip) - it is just a 5 mile round trip. That involves 2 disengagements minimum because of school zone and one unmapped roundabout. So, my disengagement rate would be 1 per "1 to 5 miles". But I've also done 15 mile drives with no disengagements ...
 
  • Like
Reactions: momo3605
I use the beta as much as possible but never in certain situations, like when my wife is in the car. I am still cautious with trusting it to get in the turn lanes soon enough. Left turns better with 10.9, so is phantom braking.
Same for me. You learn quickly where it works and where it doesn't. For example, when pulling out of my neighborhood, it thinks the shoulder is a lane. Because of this, I never engage until later in my drive. When a new FSD version is released, I do a couple of tests in the problem areas to see if they have improved or not but other than that, I'm careful about where I use it.
 
Poor comparison. We are not talking about carefully selected few miles of roads with Lidar & HD maps etc. They don't have the technology to scale to the size Tesla is aiming for. Basically a different market.

Let me know if you have any comparison to a company that has an ODD of all mapped roads in US.
There are fair number of FSD Beta testers driving the same roads in San Francisco as Waymo and Cruise are. Surely they will select the "over 100 miles" choice on your poll?
 
That involves 2 disengagements minimum because of school zone and one unmapped roundabout
What kind of disengagements are these? For the school zone, is it because an intervention of lowering the set speed doesn't slow the car down fast enough? And the roundabout is that it doesn't correctly yield (requiring brake) or that it needs accelerator intervention to move smoothly through?

But in any case, it seems like "school zone" behavior doesn't exist yet for FSD Beta. California's typically are indicated with "WHEN CHILDREN ARE PRESENT" white signage while other states might have a yellow "SCHOOL" sign attached to the speed limit sign that also says "WHEN FLASHING," so there's some additional complexity of getting it to work for the whole FSD Beta audience.

I believe even production Autopilot knows these are special conditional speed limit signs and tries to ignore them, so it's probably more of some engineer writing the logic to make use of what the neural networks already predict and verifying they are good enough to change driving behavior. And this also would only happen if it gets prioritized enough over other functionality that might be more critical while FSD Beta is driver assistance. Tesla most likely knows people disengage for school zones, but maybe it's not as critical in addressing right now as maybe there's more pressing safety issues.

This might be even lower priority if Tesla is confident that FSD Beta can safety drive with children around anyway. There's definitely a different set of things to prioritize if the goal is to be "safer" than a human and not so much "follow all laws."

However, once FSD Beta has the school zone behavior added and is working consistently, then sounds like your average miles between disengagements would double.
 
Same for me. You learn quickly where it works and where it doesn't. For example, when pulling out of my neighborhood, it thinks the shoulder is a lane. Because of this, I never engage until later in my drive. When a new FSD version is released, I do a couple of tests in the problem areas to see if they have improved or not but other than that, I'm careful about where I use it.

In which case our (and Tesla's) disengagement metrics are skewed, because we intentionally don't engage FSD beta in "tricky" situations. Therefore if Tesla is measuring quality of the FSD system based on disengagements, the number of disengagements will appear lower than what it would really be.
 
  • Love
  • Like
Reactions: hellocar and Matias
Would be beneficial to have Tesla set up focus groups with drivers who are using FSD Beta. Many improvements are needed to help lower the rate if necessary disengagement, but not sure if the data we send via our camera taps for FSD issues are sufficient for engineers to fully understand the issues we may be having.
 
There are fair number of FSD Beta testers driving the same roads in San Francisco as Waymo and Cruise are. Surely they will select the "over 100 miles" choice on your poll?
But Tesla is not geohacked to SF like those others.

BTW, Waymo seems to be 3x better than humans and yet doesn’t dare to open the service for non-nda public. What are they hiding ?
 
Last edited:
What kind of disengagements are these? For the school zone, is it because an intervention of lowering the set speed doesn't slow the car down fast enough? And the roundabout is that it doesn't correctly yield (requiring brake) or that it needs accelerator intervention to move smoothly through?

School zone disengagement is because of multiple things, but mainly not reducing speed quickly and close to getting into school parking lot anyway. Roundabout is tiny and FSD doesn’t slow down and I think it will hit the curb, don’t want to take chances.

Coming to the overall disengagement rate - I think it will be lower for long straight suburban driving and high for neighborhood driving. There are a few cases that if Tesla handles will make neighborhood driving better. Then comes the long tail. Tesla still has a lot of low hanging fruits to pick.