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

FSD Beta 10.10 Release Notes

This site may earn commission on affiliate links.
Today's FSD joy was the minor PB event accompanied by screaming beeping. While the PB was no big deal, the loud beeping with no notification on the screen is a real mystery to me. If the car has decided, for whatever reason to sound an alarm, you'd think the car could proved some clue as to what the problem is.
Probably forward collision warning? Doesn't always mark what it's beeping at in red for long enough. Also I remember the same logic applies for situations where the car thinks it's running out of drivable space which also may not show up on screen. With 10.10.2 mine definitely throws random forward collision warnings at objects both real and imaginary, usually during left turns, but sometimes when going straight.
 
Probably forward collision warning? Doesn't always mark what it's beeping at in red for long enough. Also I remember the same logic applies for situations where the car thinks it's running out of drivable space which also may not show up on screen. With 10.10.2 mine definitely throws random forward collision warnings at objects both real and imaginary, usually during left turns, but sometimes when going straight.
I've had false AEB alerts before. Those post a notice on the display that also goes in the notification history. This was like the typical major PB events that I used to get with regularity last year. The car slows significantly accompanied be beeping, but no indication what the issue is. There's nothing on the display colored red and in this case, no other cars close to me up ahead. Obviously, the car concluded that there was a safety issue.

I understand and accept that the current state of FSD is not perfect and that PB events are a part of the work in progress. My gripe is that it should post a notification as to what danger the car detected. There was nothing. FSD continued to plod merrily away once the beep sequence was completed and even sped back up on its own.
 
  • Like
Reactions: VanFriscia
I've had false AEB alerts before. Those post a notice on the display that also goes in the notification history. This was like the typical major PB events that I used to get with regularity last year. The car slows significantly accompanied be beeping, but no indication what the issue is. There's nothing on the display colored red and in this case, no other cars close to me up ahead. Obviously, the car concluded that there was a safety issue.
Hm, not every forward collision warning shows a warning in my experience. I only remember one that had a warning that might have also ended up in the notification history (though I didn't check); this was pre-beta and actually on the freeway. But with 10.10.2 some don't even mark anything as red; in those situations I'm guessing they are FCWs based purely on the tone used.

It can also happen when it thinks something is coming from the side - mine beeped at a parked car for this reason just the other day.
 
So far in my case it seems 10.10.2 is a major step backwards.

Just on my 15 min drive over to the supercharger it tried to run an obvious four way stop with clear and visible signs and markings.

It also tried to swerve into last min right turns (one into oncoming traffic) when in both cases it was supposed to go straight and the navigation route showed as much.

It to mention it’s also not handing long night driving or light sprinkles as well as it used to. Keeps throwing up error messages that conditions aren’t good and sometimes disengaging.
 
So far in my case it seems 10.10.2 is a major step backwards.
the first drive (or maybe even day of drives) was pretty terrible for me. Today however, 10 miles to work, 4 miles at lunch, 10 miles home not one single disengagement. It was a night and day difference. It was even sprinkling on the way home from work and FSD kept chugging right along!
 
the first drive (or maybe even day of drives) was pretty terrible for me. Today however, 10 miles to work, 4 miles at lunch, 10 miles home not one single disengagement. It was a night and day difference. It was even sprinkling on the way home from work and FSD kept chugging right along!

We’re in the same area so we had a similar baseline today.

One thing I would say is there’s a fork in the rolling hills around my house and it’s never got it right on any version of beta before. Today it did it almost perfect. First time ever.
 
I understand and accept that the current state of FSD is not perfect and that PB events are a part of the work in progress. My gripe is that it should post a notification as to what danger the car detected. There was nothing. FSD continued to plod merrily away once the beep sequence was completed and even sped back up on its own.
Agreed, this drives me nuts. If you're going to give me the chimes of doom, at least let me know what you THOUGHT the trouble was! Though this sometimes happens to me when I'm driving manually too. Kinda weird.

And then there's the case yesterday, where I'm driving through an empty parking lot that's barely marked and I get the chimes of doom, along with a message that it applied corrective steering for my safety. Thanks for that, ADAS :)
 
Had a airport run today. On the way back, FSD Beta portion of some 7 miles was uneventful - infact zero disengagements and interventions. Doesn’t happen often for me because of unmapped roundabouts and school zones.

If they handle school zones better i.e. recognize school zone speed limits, I can start having zero disengagement drives regularly.

p.s. And on freeways with bright sunlight and strong shadows - no hint of phantom braking too.
 
If they handle school zones better i.e. recognize school zone speed limits
There was a time, several versions back where it did seem to work. I had a passenger who asked me, while driving through a school zone, "What does that symbol mean?". It was near the speed limit sign, and had an exclamation point. I couldn't tell what it was referring to, but later when looking it up online saw that it was for school zones. Never saw it again. 😄
 
  • Informative
Reactions: EVNow
With getting closer to 1 stack, do they have enough room to run two FSDs like they originally planned? If so, maybe the chimes of doom are some kind of miscompare between the two instances. Never heard if they were going to do any kind of comparison, so probably not

It never actually did that.

For a good while the second node wasn't used for anything.

Then for a bit it was running a clone of the first node, but passively- like a failover. That didn't last super long because they ran out of compute on the first node, and began having to split the single FSD code between both nodes.

To the point where they're using all compute on both nodes, and still don't really have enough.

Some folks are convinced they'll magically "optimize" their way back to not only a ton more functionality they don't have at all now AND get it to fit in a single node.

Others are convinced there's no way to get >L2 without at least HW4 (possibly more, but we have no way to know that until they get it working at all).
 
  • Informative
Reactions: VanFriscia
Some folks are convinced they'll magically "optimize" their way back to not only a ton more functionality they don't have at all now AND get it to fit in a single node.
They can definitely optimize - and probably don't need a ton more functionality. I think moving things from C++ code to NN, like they are doing, is a way to optimize.

I've no way to tell if they can optimize enough to fit in one node or not - but I do know when still in dev, code is quite sub-optimal.
 
They can definitely optimize - and probably don't need a ton more functionality. I think moving things from C++ code to NN, like they are doing, is a way to optimize.

I've no way to tell if they can optimize enough to fit in one node or not - but I do know when still in dev, code is quite sub-optimal.


They absolutely need a TON more for >L2.

Read the CA DMV docs. They openly admit they're missing major features required for >L2 operation, like they don't exist in the code at all. The ODER stuff especially.
 
They absolutely need a TON more for >L2.

Read the CA DMV docs. They openly admit they're missing major features required for >L2 operation, like they don't exist in the code at all. The ODER stuff especially.
CA DMV docs talk about the stupid SAE levels ?! Link .. ?

Here is the list NHTSA suggests ...

 
CA DMV docs talk about the stupid SAE levels ?! Link .. ?

Here is the list NHTSA suggests ...



CA DMV docs specifically have Tesla explaining they don't need to follow the CA rules for L3+ systems because the beta is incapable of L3 or better operations, because it's missing numerous major features required, including an OEDR capable of such autonomy.

They even state that the "finished" wide-release version of the beta (also called city streets) will ALSO not have these features and not be capable of >L2.

Some FUTURE software, that doesn't exist and is not being tested in the beta, will eventually add those, per the docs.


This was pretty widely reported at the time (TESLA ADMITS THEY HAVE NO SELF DRIVING) and such, surprised you missed the news.

For example:


Tesla said:
we do not expect significant enhancements in OEDR or other changes to the feature that would shift the responsibility for the entire DDT to the system. As such, a final release of City Streets will continue to be an SAE Level 2, advanced driver-assistance feature

Story has other quotes and a link to the full set of docs.
 
So far in my case it seems 10.10.2 is a major step backwards.

Just on my 15 min drive over to the supercharger it tried to run an obvious four way stop with clear and visible signs and markings.

It also tried to swerve into last min right turns (one into oncoming traffic) when in both cases it was supposed to go straight and the navigation route showed as much.

It to mention it’s also not handing long night driving or light sprinkles as well as it used to. Keeps throwing up error messages that conditions aren’t good and sometimes disengaging.
I half agree with this. In the daytime with clear visibility, 10.10 seems a bit better with the exception of merging. I do agree that night driving has taken a step back and is now as bad as some of the pre 10.8 versions.
 
No SAE J3016 defines each level. Many states (including CA) use some SAE definitions in their regulations though.

Tesla is making clear city streets (FSDBeta) is incapable of meeting L3 or greater requirements, and the finished/wide release version will not change that.

Anything they do that us L3 or greater will be some future version that adds elements city streets simply does not and is not intended to have.

They especially call out the object and event detection and response (OEDR) sub-task as being too limited to operate beyond L2, with a list of circumstances that could cause failures because the system may not even be capable of recognizing (let alone properly responding to) them- adverse weather, complicated or adversarial vehicles in the driving path, construction zones, emergency vehicles, large uncontrolled intersections with multiple incoming ways, occlusions, road debris, static objects, and unmapped roads

As such the inherent design of the system may result in situations the human driver MUST respond to without the system being capable of alerting them to do so.... (which means it can't be L3, let alone higher).
 
Tesla is making clear city streets (FSDBeta) is incapable of meeting L3 or greater requirements, and the finished/wide release version will not change that.
I think we all know why Tesla is claiming that.

Yes, wider release of FSD Beta will still be called "ADAS" by Tesla. Tesla will not say you can take your mind/eyes off the road until they satisfy their actuaries that the risk is worth it. That would only happen when the disengagement rates are 5x to 10x better than humans and there is a compelling reason in the market. We are far from that ...
 
  • Like
Reactions: RabidYak
I think we all know why Tesla is claiming that.

Of course we do. Tesla tells us, in writing, to a government agency.

The OEDR is not designed for and not capable of operating safely without constant, 100% of the time, human supervision.

This has nothing to do with actuaries, it has to do with lacking specific capabilities required to drive safely without a human.

FSDBeta getting better at the stuff is can do won't change that, because there's stuff it flat out can not do that is required for any actual autonomous driving (ie without a human).

Tesla calls this out, and also calls out they'll be developing some FUTURE system that DOES do these things- but FSDBeta, even in the wide release version is not that

City streets is L2. Period full stop.

Which is exactly what Tesla has told you about it on the sales page since March 2019, so it shouldn't be surprising.