Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
How did version 11 handle round abouts
As a matter of fact, V11 just did splendidly on a 2-lane rotary in Boston last week, but I would only try it at night with no significant traffic. Despite no challenging traffic, it is by no means obvious how to navigate this particular rotary (it's the one on I95 in Reading, MA) when you hit it for the first time, so just to have done the correct lane positioning was pretty impressive in my book. I'll never find out how V11 does with traffic in it.
 
  • Like
Reactions: aronth5
I went to supercharger station in Mira Mesa today to see how many Ford Mustang E were there. None. On the way out of the parking I disengaged FSD because v12.2.1 didn't move to make right turn when a car on the left yielded the way. Ironically v12.2.1 stopped to yield to another car moving from a parking to the street and the yield was not necessary because that car only wanted to be on the rightmost lane (my car was in the middle lane).
 
Last edited:
I’ve had too many problems with 2 and 3 lane roundabouts on v11. If there’s no traffic I’ll let it try but it has never once done it once and it’s done some quite dangerous things a number of times. I usually with try a few times on each new release but it’s never been trustworthy even with no traffic.
 
How did version 11 handle round abouts
Not well especially large roundabouts where I live. FSD usually comes to a full stop or at best slows down to 2mph when the average entrance speed is about 20mph. Needless to say if there is anyone behind me I use the accelerator pedal to avoid getting honked at. Smaller roundabouts are better but still a crapshoot.
 
Nice job driving roundabouts including one with 2 lanes taken several times.

🔥🔥 🔥

Good to see potential improvement here. Looks like it understands what needs to be done. Would be interesting to see performance at smaller less well-marked instances. The question is how general the behavior is…I would also like to see v11 here.

Erratic signaling; doesn’t seem to know all the rules, but neither do humans! Signal left if you’re going left, right when you’re going right, and right to exit the roundabout when appropriate. Basically let people know what to expect especially if there is a possibility of crossing paths. But did do some signaling, which is good!

Had a failure at 3:41 to go when clear. Yield problem.

There was also a sharp slowing from 20mph to 15mph at 2:25. Not cool. That would have passengers up in arms.

Definitely not A+ as described. Would give it a B. What is up with the lack of standards? These YouTubers need to be super critical and call things as they see them, for credibility.

If people can just avoid getting into serious accidents for the next couple days maybe more people can enjoy v12. It seems like on average it is better so maybe they can expand it.
 
Last edited:
  • Like
Reactions: FSDtester#1
The great thing about V12.2.1 is that it's a surprise with every drive, since its behavior is a not based on some predetermined code

Perhaps Tesla's goal is to have it be more reliable, but it may be a while before that happens

You'll be much more excited to use V12 than V11 because V12 works in so many more situations

With V11, when I saw a situation that seemed too difficult, I'd rather disengage and do it myself to save myself the discomfort and anxiety
 
The great thing about V12.2.1 is that it's a surprise with every drive, since its behavior is a not based on some predetermined code

Perhaps Tesla's goal is to have it be more reliable, but it may be a while before that happens

You'll be much more excited to use V12 than V11 because V12 works in so many more situations

With V11, when I saw a situation that seemed too difficult, I'd rather disengage and do it myself to save myself the discomfort and anxiety
I’m actually getting excited to try V12. One thing with V11 is it’s predictable, like you said. At this point I know exactly where it’s going to fail and I usually preemptively disengage. Why should wait for it to enter the wrong lane and then veer back?
 
I’m actually getting excited to try V12. One thing with V11 is it’s predictable, like you said. At this point I know exactly where it’s going to fail and I usually preemptively disengage. Why should wait for it to enter the wrong lane and then veer back?
Given propensity for people to use their phones and push the nags while using FSD, this unpredictability could lead to higher accident rates, because it is more perhaps likely than v11 to do something incorrect when the driver is not expecting it.

Will be an interesting experiment!
 
I don't know how Tesla is gonna train V12 for all the thousands of different situations. So far, NNs don't seem to "get it" yet.

I had to disengage today, optionally because 12.2.1 drove into a backed up left turn lane, it wasn't blocking the intersection, so I guess it's fine, but it wasn't ideal

V11.4.9 usually stayed back if it saw the lane was backed up
An issue I’ve seen for years (we can say that now) is that FSD follows the turn lane rules a little too closely. There are many instances with lane situations that FSD will only get into the turn lane when it’s “supposed” to. Around here, people pile into said turn lanes as soon as they can, often backing up onto otherwise “undriveable” space. This is how people drive around here.
 
Around here, people pile into said turn lanes as soon as they can, often backing up onto otherwise “undriveable” space. This is how people drive around here.

That's one of the surprising decisions that 12.2.1 made yesterday. It was so delightful when it acts human in this way.

The lead car was creeping over to the right of a 2-car-wide-lane to get in line for a backed-up straight and right turn lane. 12.2.1 was smart enough to understand that a line was forming for the turn lane to get on the highway, so it followed the lead car.

AND today, it was again, VERY surprising but 12.2.1 made a different decision. In today's situation, the lead car was doing something very similar to yesterday, BUT there wasn't as much of a line, so 12.2.1 didn't follow behind the lead car, so I was like "oh no! it's not gonna get in line!" BUT BUT it turns out the lead car was slightly slowing down and ended up turning into a driveway, so THAT's why 12.2.1 didn't follow directly behind it but to the left of it! I was so surprised by this because I wasn't predicting that the lead car would turn into a driveway. That's why I said that 12.2.1 makes decisions based on very subtle cues.

I'm sorta shaking posting this because it blew my mind :)
 
Last edited:
Watch out for odd situations in parking lots like caution tape:

12.2.1 caution disengagement.jpg


There's a parking lot here that has various entrances blocked by chains that's even harder to see than the yellow here. I would expect needing to disengage current 12.x for these for a while.
 
Watch out for odd situations in parking lots like caution tape:

View attachment 1023462

There's a parking lot here that has various entrances blocked by chains that's even harder to see than the yellow here. I would expect needing to disengage current 12.x for these for a while.

Geez, that's a poorly set up barrier, especially with low visibility at night

Talk about an edge case
 
Watch out for odd situations in parking lots like caution tape:

View attachment 1023462

There's a parking lot here that has various entrances blocked by chains that's even harder to see than the yellow here. I would expect needing to disengage current 12.x for these for a while.

I’ve had to disengage FSD on multiple attempts as it tried to enter the wrong and gated entrance to the local Tesla service location. Apparently it couldn’t or can’t see steel box gates. Too much air and too little surface.
 
Watch out for odd situations in parking lots like caution tape:

View attachment 1023462

There's a parking lot here that has various entrances blocked by chains that's even harder to see than the yellow here. I would expect needing to disengage current 12.x for these for a while.
Considering Cruise did the same thing (completely ignored caution tape and ran right into downed power lines).
Cruise Robotaxis Spotted Tangled In Caution Tape And Downed Wires After SF Storm

Waymo also had their expansion suspended for review with CPUC, with article citing an example with them also: "one drove through the yellow tape at a crime scene"
Waymo's robotaxi expansion beyond SF temporarily suspended as part of review

I think this corner case will not be handled for a long time. I'm guessing the NNs have a hard time distinguishing them from other soft plastic materials they are expected to drive right through and not stop for (for example a plastic bag in the wind).