Poll : What is your disengagement rate with FSD Beta ?

[Mardak]

Today's weekend 50mi routine had a system forced disengagement from main camera blinded by the sun although I believe previous FSD Beta versions show a "blinded" message instead of the big red "take over immediately." Last week this route had a a convenience disengagement to prevent it from incorrectly getting into a right-turn only lane for a parking lot, but today it did fine as there was no lead vehicle blocking view of the desired intersection, so theoretically the neural network can make the correct prediction even when occluded.

Looking at the whole week's usual trips, I'm averaging around 30 miles for each convenience disengagement. I do see a lot more safety related issues when intentionally testing with difficult routes though.

 

[Electroman]

Interesting, no update for the last 3 months with anecdotes?

 

[EVNow]

May be I should start a fresh poll.

 

[Twiglett]

May be I should start a fresh poll.

definitely worth it, to see if the numbers change. My overall disengagement rate has decreased, but of that maybe because I don't bother with the beta in areas where it is consistently obnoxiously wrong

 

[Z_Lynx]

Yes. Fresh poll (things have "changed").

Repeat what you don't want included. You mentioned interventions like pressing the accelerator or changing speed limit. Maybe you have other things to add?

Would coming to almost a complete stop triggered by an approaching car count as an intervention if you let it painfully play out (i.e. no intervention but not acceptable)? Would accelerating to break the hard deceleration count? I can understand not counting acceleration to get through turns or stops.

I think it could also help to know how heavily people are curating their drives to avoid disengagements and only use FSD where it is likely to be successful (what Twiglett is doing - me too).

Number of disengements is one metric that can be open to interpretation.

 

[momo3605]

Since you last posted we were on 10.10.2? We’ve only moved up 1 dot version since then. Imo nothing much has changed. 10.11.2 is marginally better and worse in certain instances but I’d probably say disengagement rate is about the same

 

[t3sl4drvr]

Here is 2021 data from California

 

[t3sl4drvr]

Interesting to see if DMV will be able to force Tesla to start reporting... According to this poll, their numbers might not be that impressive.