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.
Doing lots of good road drives, not highway, therefore using the v12 stack. Guess my issue now is, I’m in large mall and multi store parking lots and there not always good drives. Need the end to end to have learned data driving in these areas to eliminate my edge cases.
 
Uh, not exactly. California is, uh, different. In many ways, including speed limits. Some would call California "special" in the SNL sense.

California law calls posted speed limits "pima facia". But Calif's "basic speed" says you can't drive faster than is "reasonable and prudent". I call it fitting' and proper. Under case law precedents, posted speed limits can not be less that the 85th percentile speed as measured with an traffic study, unless accident history suggests this is unsafe. Exceptions exist for school zones and such. (So when we see those two rubber tubes stretched across a road, we drive extra fast so as to raise the 85th percentile.)

So, when California Alan says that particular "speed limit" is not enforceable, what he actually means is that a ticket for violating the posted limit may be successfully contested. So police instead cite for reckless driving, which includes driving 15 MPH over statutory maximum speeds, such as 65 on freeways or 55 on undivided two lane highways, unless posted higher in either situation.
Yep, there's a law firm's article on this.
Speed Traps & Engineering and Traffic Surveys :: Los Angeles County Crime Defense Lawyers Greg Hill & Associates
Basically in California, if the post limit is lower than the statutory limit, a valid engineering study is required to justify that limit. If there is no such study or the study has expired, the limit is not enforceable, meaning if you contest the ticket in court, you can win. That said the officer can always still write you a ticket, but they can always do that even if it is not justified, which means the fact you got a ticket doesn't necessarily mean you violated the law (which is the whole point of traffic court).

As it relates to FSD however, I don't ever expect it to deal with such nuances.
 
So based on your feedback it’s doing exactly as it should. At speed or 2-3 over. People need to start excepting that the target is a law abiding autonomous transport. Not a “I drive illegal it should do as I would”. IF there is Any goal of liability transfer we need to start accepting that it will Never disobey the primary law even if it’s not what others are doing or us.
I agree with this. One problem though is that speed limits are completely arbitrary in some parts of North America compared to Europe or Asia, and so is enforcement. There are 30 zones here where you really should not drive much above 30, but if you do, probably nothing will happen unless it’s next to a school and a cop happens to be there, so unfortunately, many still do speed on these roads. OTOH you have highways posted unreasonably at 70 where even trucks will drive over 100 and speeders at 120. Abide by the law and you’re going dangerously slow. North America relies too heavily on arbitrary enforcement along with unwritten norms of how much speeding is acceptable.

Elsewhere in the developed world, many urban roads are slow and strictly enforced, highways tend to be fast (more so in Europe rather than Asia), and most enforcement is automated, also with a lot less reliance on stop signs and other signage, and more regularized speed limits.

This mess will have to be cleaned up here for robotaxis to more safely mix with other traffic.
 
For clear views there isn't a purpose to creep. It's when there isn't a clear view FSD has to creep. How else can FSD make a safe decision?

Good example. FSD actually had to creep past the center line on the road before FSD would turn left. If there were any cars from either direction they have to stop. I tried to keep the camera at the drivers position so the B-pillar camera actually saw less. So even with cameras up front how could I supervise FSD if the car didn't creep far enough for the driver to see? Of course FSD wouldn't need to creep with front cameras to make a decision. Interesting question.

That’s a tough intersection for sure. It doesn’t seem so common here, but in other places I’ve seen round mirrors put up to look around obstacles like that. I wonder if FSD can be trained to use those?
 
  • Like
Reactions: zoomer0056
This mess will have to be cleaned up here for robotaxis to more safely mix with other traffic.

Agreed it's a mess (it's a bit of a ridiculous case of "you can break this law this much but not more", and it's highly contextual. There's a 55 zone on I70 in Denver that's that is 100% ignored, lol).

That said I've been surprisingly happy with how 12.3.3 has been doing with auto-speed. If every other car on the road followed V12 auto-speed I wouldn't find that annoying at all.
 
Auto speed works great for me. I’ve only had to override it maybe a couple of times thus far, and it is generally spot on with how fast I would go.

I am irritated that the car decides so late to change lanes in preparation for a turn. It has done the lane changes well, but I’m just waiting for a case where the traffic is heavy and it can’t find a gap, where there was one half a mile earlier. They need to fix it to start looking for an opportunity much sooner.
 
So based on your feedback it’s doing exactly as it should. At speed or 2-3 over. People need to start excepting that the target is a law abiding autonomous transport. Not a “I drive illegal it should do as I would”. IF there is Any goal of liability transfer we need to start accepting that it will Never disobey the primary law even if it’s not what others are doing or us.
That's not my experience. It was going 12-13 miles over the limit on a 45mph surface road, even when it was not in traffic. I finally disengaged automatic max. It was making me nervous when I had seen a trooper on that stretch a little while earlier.
 
They need to fix it to start looking for an opportunity much sooner.
Similarly with lane blockages ahead (for construction, for instance, or a bus stopped). It doesn't plan ahead as far as I would and position itself relative to other cars for a merge, or actually move over sooner (I know, zipper merge blah blah merge as late as possible for lane closed).

Overall it feels like it only plans for the next 10-15 seconds and isn't able to plan for an upcoming maneuver as well when it falls beyond that time horizon.
 
  • Like
Reactions: primedive
So I can essentially navigate by using just "chill mode" and turn signals and just "bop around the neighborhood", then "hit nav to home"
we have 400 miles of canals and lots of dead ends
thanks for response
Be careful with that. I was doing the same, at a T intersection I signaled left, the preview showed going left, and then it just suddenly turned right for no reason, and with no signal. No traffic around, luckily.
 
Using the turn signal to influence a turn at an intersection has never been a feature. It sounds like you just had random turns that sometimes correlated to your turn signal use.
Yesterday while meandering around accidentally deleted my Destination trying to delete a Way Point. So I just left on and tried the signals. Didn't work and would only turn if it came to a junction or a split in the street.
 
I’ve posted before how my car creeps even when there’s a clear field of view so I’m not sure what the purpose of creep is.
Yeah I have a street in a neighborhood that is super wide and completely wide open view in both directions 50' before the Stop line because of a large traffic island and a super wide street (harks back to early street car). It is 100% of a roller and no one can stop completely if there are no cars coming. But after stoping it then starts slowly creeping forward even though you can see to China in both directions. I must use accelerator if someone is behind me since I look stupid and get honked at.
 
  • Like
Reactions: sleepydoc
Using the turn signal to influence a turn at an intersection has never been a feature. It sounds like you just had random turns that sometimes correlated to your turn signal use.
Is this correct? => Using the turn signal to influence a turn at an intersection has never been a feature

I have read several times the last few pages and it seems some folks can essentially "steer" using turn signals alone
I had mixed results with this just cruising about the neighborhood that has a lot of dead ends (400 miles canals)

Is there a definitive "yes/no/maybe"
or
does the vehicle randomly
"take the scenic route since there's no destination in the "nav to"'
or
It's a bug not a feature
 
Last edited:
It’s been quiet on TeslaFi.

All the Tesla engineers high fived each other and went to MCFC2024 or something.

They gone.

I want vision based summon :(
I suspect they are working on getting a 23.8.x stable version of 12.3.x to satisfy the "you get a trial and you get a trial" "this week". I bet the current unpredictable behavior of FSDS in parking lots foreshadows the current state of ASS since it is FSDS based.
 
Is there a definitive "yes/no/maybe"
or
does the vehicle randomly
"take the scenic route since there's no destination in the "nav to"'
or
It's a bug not a feature
In my experience:

1. You can use turn signals to direct the car to turn, but only at stops. This works both with and without a destination.
2. When the car doesn't have a destination, it will only take right turns on its own.
3. If you change lanes into a dedicated turn lane, the car will make that turn.
4. The car won't always execute a lane change when requested (which is clearly a bug)

So you cannot simply turn signal your way around an area. In fact, you can't even rely on the car going straight down a residential road. This may be in some way related to cars spontaneously entering exit lanes off multilane roads.

I assume that this is just an artifact of the current implementation. That is, it comes to a given intersection and most of the training scenarios that it saw that relate to that intersection involve a right turn. I wonder if the same is true of taking exit lanes - do they have too much training data on doing that, instead of proceeding straight ahead, resulting in a bias to exiting.

My "grand day out" with this at one point involved a spontaneous right turn off a residential street for no apparent reason. The car went onto the side road, ran into a cul-de-sac, turned around, came back out, made and right and continued on the original street.

Bottom line: While using turn signals at stops can be useful, the rest of it is just a glitch that's fun to play with a couple times for the novelty of it.
 
  • Informative
Reactions: primedive
For clear views there isn't a purpose to creep. It's when there isn't a clear view FSD has to creep. How else can FSD make a safe decision?

Good example. FSD actually had to creep past the center line on the road before FSD would turn left. If there were any cars from either direction they have to stop. I tried to keep the camera at the drivers position so the B-pillar camera actually saw less. So even with cameras up front how could I supervise FSD if the car didn't creep far enough for the driver to see? Of course FSD wouldn't need to creep with front cameras to make a decision. Interesting question.

And this is a perfect example of why the B-pillar camera is inadequate at low visibility intersections and why Tesla should just reroute you since the hardware cannot make a safe decision. I had someone checking for vehicles.

Has there been definitive conclusion why a car that relies solely on vision does not have camera(s) at the front? Other than not to piss off the HW3 guys? Is it somehow easier to make the vision technology more portable (and so easier to license out) if the cameras are relatively close to each other?