FSD Beta 10.12

[boonedocks]

All these recent posts show is that no one outside of Tesla knows what source data are used for the navigation.

The routing is good enough for getting from Chicago to Des Moines, but it is a hot mess locally. The routing algorithm appears to try to shave a minute or two off each trip by taking back roads and side streets. At least for where I live, it rarely picks the route that I would have chosen. And Tesla's implementation of waypoints is terrible.

But all that could be a criticism of any of the existing consumer GPS navigation systems. They are better than nothing, but far from perfect. IMO, navigation data and routing algorithms combined are the Achilles heel of autonomous driving. They don't pose a safety risk per se, but they are a barrier to robotaxi adoption.

and absolutely nothing to do with FSD Beta 10.12

 

[Reddy Kilowatt]

I have a mysterious section in a city where the speed limit increases 5mph in one direction but is the correct 25mph traveling the opposite direction. I carefully watched to see if it was corrected with the map update but was not.

And recently, it changed to the incorrect 30mph in both directions! (clear signs it is 25mph previously in the good section; the bad stretch has none). So something new happened. I will have to see if the sign visualizations have disappeared on the previously good section.

We have similar issue, In front of my office the speed limit is 35 MPH West bound but 45 MPH East bound. I go back and forth looking for the missing 35MPH sign, and have even reported it to VDOT when they were replacing a 35MPH sign. They never replied. But I'm SURE I would get a ticket for doing 45 east bond. I kick the speed down as soon as I hit that stretch of road. The cops camp out in my driveway passing out speeding tickets. I know it is a missing sign coming off the highway. They like the revenue. The cops can simply point to the west bound speed limit sign and trick the victim.

 

[VanFriscia]

All these recent posts show is that no one outside of Tesla knows what source data are used for the navigation.

The routing is good enough for getting from Chicago to Des Moines, but it is a hot mess locally. The routing algorithm appears to try to shave a minute or two off each trip by taking back roads and side streets. At least for where I live, it rarely picks the route that I would have chosen. And Tesla's implementation of waypoints is terrible.

But all that could be a criticism of any of the existing consumer GPS navigation systems. They are better than nothing, but far from perfect. IMO, navigation data and routing algorithms combined are the Achilles heel of autonomous driving. They don't pose a safety risk per se, but they are a barrier to robotaxi adoption.

Another thing I've noticed specific to FSD routing (at least, I assume it's FSD-specific - it didn't do this before I got into the beta) is that it will sometimes make very strange routing choices in order to make the drive easier for FSD to handle. One example, I was leaving my local Subway the other day and started navigation that required me to get on the highway about a half a mile down the road. To do this, however, you had to turn left out of the parking lot, across double-lane traffic going in both directions - a crappy driving situation for sure, but it's clearly the fastest and most efficient path forward. However, FSD instead wanted me to turn right out of the lot and turn around far down the road instead - adding several minutes of time to get to the same spot, but in a way that was infinitely easier to accomplish.

I just made the difficult turn myself, at which point the navigation rerouted and I engaged FSD, so it wasn't a big deal...just interesting. It definitely threw me off at first though - I couldn't figure out where on Earth it was trying to send me! I have to assume that FSD didn't want to attempt making that difficult left turn.

 

[sleepydoc]

Another thing I've noticed specific to FSD routing (at least, I assume it's FSD-specific - it didn't do this before I got into the beta) is that it will sometimes make very strange routing choices in order to make the drive easier for FSD to handle. One example, I was leaving my local Subway the other day and started navigation that required me to get on the highway about a half a mile down the road. To do this, however, you had to turn left out of the parking lot, across double-lane traffic going in both directions - a crappy driving situation for sure, but it's clearly the fastest and most efficient path forward. However, FSD instead wanted me to turn right out of the lot and turn around far down the road instead - adding several minutes of time to get to the same spot, but in a way that was infinitely easier to accomplish.

I just made the difficult turn myself, at which point the navigation rerouted and I engaged FSD, so it wasn't a big deal...just interesting. It definitely threw me off at first though - I couldn't figure out where on Earth it was trying to send me! I have to assume that FSD didn't want to attempt making that difficult left turn.

I've noticed the same thing on occasion and had the same assumption. It's actually a pretty rational approach; it all has to do with the relative value of safety and time. FSD is putting a higher value on safety and ease of navigation and a lower value on time and distance. If you think about FSD like driving with a new teenage driver then you'd probably make the same decision!

 

[ammulder]

Not sure if eveyone's aware but you can go to safety.sfnet.app and enter your past 30 days Safety Score and it will tell you how many miles until you get to 100. It's reported to be accurate within 1 mile. I'm at a solid 99 with mostly 100's. A few 95, 97 and 98's. But after I entered one 88 and one 89... it has me at 1236 miles from 99 to 100, yikes!
What's interesting is if I delete my 88 and 89 I go to a 100 in 14 miles. So I presume I can drive on auto pilot a lot to keep my score good... and then on May 29'th they'll fall of the 30 day rotation and I'll hop to 100. hummm?

Hey, that's nothing. The first day I got a safety score on our new X, I was driving past cars parked on the street in my neighborhood. It gave me a Forward Collision Warning for the middle car in a row of three. Really? How would I hit any parked cars, much less the middle one in a row of three?!?

So I started off with a completely bogus FCW and a safety score of 37.

A week later I'm now up to a 98, but it'll be three more weeks before that 37 falls off, so I ain't getting to 100. I am assuming a 98 will be good enough if Elon is talking about letting 95s in the beta. Well, I'm hoping, anyway.

Really eager to see someone reporting 10.12 release notes, which seems to happen as soon as the initial rollout starts... I can't drive like a grandma forever! Come on, June!

 

[EVNow]

All these recent posts show is that no one outside of Tesla knows what source data are used for the navigation.

Our best guess is that Tesla takes data from various sources and integrates the data. See my old thread on this where I found several issues near my home, corrected some of them in Open Street and some in TomTom and figured out that neither of those updates always made it into Tesla mps.

FD Beta : TomTom Map, Roundabouts and the Speedlimit

TL;DR : Let us use this thread to talk about map changes that could potentially make FSD Beta behave better Thanks to the post below by @MP3Mike I discovered an interesting little fact. Tesla probably does use TomTom maps rather than Google or OpenStreet for city FSD Beta driving. I've two...

teslamotorsclub.com

 

[Twiglett]

I'm guessing some of this degradation has to be coming from mapping screwups.
For the last few months I've been making identical weekend trips. About half way through, the nav started to take us on back roads for the last few miles and gave the final destination in a field far from the actual address.
Then a week or two ago, it just started getting it right again.
Now taking many more journeys with the nav turned off, but the downside is that for some roads 10.11 seems much happier with a route.
Other times it almost seems to zigzag down the road changing lanes.

All through this, Omar is posting perfect, zero disengagement, smooth, perfect journeys.

 

[apsen]

Hey, that's nothing. The first day I got a safety score on our new X, I was driving past cars parked on the street in my neighborhood. It gave me a Forward Collision Warning for the middle car in a row of three. Really? How would I hit any parked cars, much less the middle one in a row of three?!?

FCW They really need to improve that detection if they are using it in safety score... It is bad enough when the car starts screaming at you for no reason but then it is even worse when it also affects your safety score...

There's a stretch of the road where it would often freak out - divided highway turning slightly right and elevation changing in such a way that you have easier time seeing over the divider than on other stretches of this road. Judging by all those different instances it seems that one thing in common is the oncoming car on the other side of the divider. At least it doesn't do it anymore there.

I also seem to notice/remember more FCWs after that snafu with FSD beta when they had to scramble to fix it but I thought it may be observer bias or just specific to beta. Now since I have new car and going trough safety score exercise again I could definitely say it is worse than the first time around.

Also on a related note: when the hell will they fix the following distance setting on a vision cars? It so bad that it will ding your safety score for following to close on setting 7. I just did an experiment by switching on cruise control instead of autopilot on highway and got hit with around 13% on the following distance metric - if you are gaining on the car in front of you tesla will get way too close before eventually backing off a little. But still nowhere as far behind on setting 7 as it used to do on setting 5 with radar

 

[jabloomf1230]

and absolutely nothing to do with FSD Beta 10.12

Well, since you mentioned that, I disagree. When you drive with FSD beta, you need to use the navigation. If you don't, the car will just drive straight, of course until you reach a tee or a dead end. With both FSD and navigation enabled, it's a constant struggle when you want to deviate from the preordained route. You must disengage and let the navigation reroute. Hence there are intersections that are not tested with FSD beta.

Okay, you say, why not just follow the original routing? In my experience that leads you down roads that no driver would ever purposely pick. And it's not done to avoid either poor visibility ULTs or some other edge case.

Tesla should immediately fix the waypoint function. That would prevent the above issue.

 

[sleepydoc]

I'm guessing some of this degradation has to be coming from mapping screwups.
For the last few months I've been making identical weekend trips. About half way through, the nav started to take us on back roads for the last few miles and gave the final destination in a field far from the actual address.
Then a week or two ago, it just started getting it right again.
Now taking many more journeys with the nav turned off, but the downside is that for some roads 10.11 seems much happier with a route.
Other times it almost seems to zigzag down the road changing lanes.

All through this, Omar is posting perfect, zero disengagement, smooth, perfect journeys.

I’ve been wondering if the navigation data is the wildcard that we don’t know about. Is there any way to see when it’s been updated? It does it automatically in the background so you don’t know when it happens unless you happen to check the software screen.

I know my navigation data updates with my last software update a few months ago. When I checked it was 2021.44.xxx so I don’t think it’s been updated since. But I can’t be sure.

As an aside, if my map data really is from October 2021 that would explain why it takes so long for things like speed limits to get corrected!

 

[azred]

Not sure how the topic became mapping but since 10.12 seems to be frozen in time, I’ll comment on mapping, too. I believe mapping will never fully satisfy me until we can drag/adjust the route on the screen. Adding waypoints helps a little but is no substitute for dragging the map route to alternative streets. But that issue doesn’t stop me from engaging maps.

What does limit my use of maps on FSD city streets are the ridiculous lane changes. For example, if I am driving on a long straight stretch on a street and engage FSD without a destination, the car mostly stays in the (hopefully) logical lane that I have chosen. But if I activate mapping to drive to a location on that street, the car aimlessly bounces from lane to lane and frequently tries to make a last minute lane change that makes turning into my destination nearly impossible.

 

[Matias]

We’ll know as soon as Omar posts the obligatory “OMG this version is so great! ZERO disengagements!” video.

Since every version is perfect with zero disengagements it’s unclear to me how the next version can be any better.

 

[MP3Mike]

Tesla should immediately fix the waypoint function

What is broken in the waypoint functionality?

 

[tmoz]

This Thread is mislabeled. It should have been called "Waiting for Santa 10.12".

 

[n.one.one]

This Thread is mislabeled. It should have been called "Waiting for Santa 10.12".

Tesla barada nikto

 

[PACEMD]

This Thread is mislabeled. It should have been called "Waiting for Santa 10.12".

Well, you better watch out.........you better not pout..........

 

[PACEMD]

Hey, where in the ham sandwich is my update? It's Tuesday already, hello, Tesla! 30 minutes or the update's free......

 

[Iain]

I'm guessing some of this degradation has to be coming from mapping screwups.
For the last few months I've been making identical weekend trips. About half way through, the nav started to take us on back roads for the last few miles and gave the final destination in a field far from the actual address.
Then a week or two ago, it just started getting it right again.
Now taking many more journeys with the nav turned off, but the downside is that for some roads 10.11 seems much happier with a route.
Other times it almost seems to zigzag down the road changing lanes.

All through this, Omar is posting perfect, zero disengagement, smooth, perfect journeys.

I'm also in Austin and did a map update a month or so ago that messed up speed signs and certain nav routes. The car seems to drive a lot better with this last update, but then I experience constant lane changes and odd behavior out of nowhere and then it's back to driving well again (BTW I don't ever experience anything like Omars perfect drives)

 

[Ramphex]

Did we have another failure to launch?

 

[jabloomf1230]

Electrile Dysfunction?