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.
I tried FSD Beta 10.12.2 on a unmarked back road yesterday and it did not do well. It ping ponged quite a biit, trying to center itself in the middle of the road. And since it was driving in the middle of the road, it had to move over when incoming vehicles were coming. So I would say 12.12.2 is rather unreliable on roads with no lane lines. However, on roads with clear lane markings, it is quite good at lane keeping IMO. I was cruising on a major state road and FSD Beta handled it great, except for a couple phantom braking moments. As a driver assist, I find FSD beta is quite good when the route does not have any turns in it, when it is basically just lane keeping, maintaining a safe distance from a car in front and stopping and going at stop signs and red lights.
 
I tried FSD Beta 10.12.2 on a unmarked back road yesterday and it did not do well. It ping ponged quite a biit, trying to center itself in the middle of the road. And since it was driving in the middle of the road, it had to move over when incoming vehicles were coming. So I would say 12.12.2 is rather unreliable on roads with no lane lines. However, on roads with clear lane markings, it is quite good at lane keeping IMO. I was cruising on a major state road and FSD Beta handled it great, except for a couple phantom braking moments. As a driver assist, I find FSD beta is quite good when the route does not have any turns in it, when it is basically just lane keeping, maintaining a safe distance from a car in front and stopping and going at stop signs and red lights.
The drive I took was this one. This section is 66 miles, I had nine slowdowns (what some call phantom braking) from 75mph to 60mph, followed by an incredibly slow resumption of previous speed. Speed limits on this road vary from 30 or 40 in the small towns to 70/75mph for the majority of it.
 
  • Helpful
Reactions: diplomat33
I tried FSD Beta 10.12.2 on a unmarked back road yesterday and it did not do well. It ping ponged quite a biit, trying to center itself in the middle of the road. And since it was driving in the middle of the road, it had to move over when incoming vehicles were coming. So I would say 12.12.2 is rather unreliable on roads with no lane lines. However, on roads with clear lane markings, it is quite good at lane keeping IMO. I was cruising on a major state road and FSD Beta handled it great, except for a couple phantom braking moments. As a driver assist, I find FSD beta is quite good when the route does not have any turns in it, when it is basically just lane keeping, maintaining a safe distance from a car in front and stopping and going at stop signs and red lights.
I have been saying that since day one. I don't think NN will be able to correct the center bias on unmarked roads. Driving on the right side of unmarked roads has to be hard coded in at the beginning!!! So fundamental in driving 101. There is no mechanism on the current FSDbeta allow for user lane bias input. i.e. there is no training!! I have also said current FSD Beta will fail a DMV driving test right out of the DMV parking lot if the first road is an unmarked road,..which happened to be on my city's DMV.
 
I have been saying that since day one. I don't think NN will be able to correct the center bias on unmarked roads. Driving on the right side of unmarked roads has to be hard coded in at the beginning!!! So fundamental in driving 101. There is no mechanism on the current FSDbeta allow for user lane bias input. i.e. there is no training!! I have also said current FSD Beta will fail a DMV driving test right out of the DMV parking lot if the first road is an unmarked road,..which happened to be on my city's DMV.
And yet there is a ½ mile stretch to my home with unmarked roads (and a couple of turns) that the car handles well for me (including staying on the RHS). I'm not sure why I'm seeing such different results to @diplomat33 ? Map data?
 
  • Like
Reactions: diplomat33
And yet there is a ½ mile stretch to my home with unmarked roads (and a couple of turns) that the car handles well for me (including staying on the RHS). I'm not sure why I'm seeing such different results to @diplomat33 ? Map data?

It could be a map data issue. Perhaps the county road I was on is not mapped so the car fell back on vision to try to drive it.
 
  • Like
Reactions: jabloomf1230
Gents, I think we need to talk about the two different types of nags.

1. Steering wheel induced nags. This type of nag is triggered by the car not sensing your hand on the steering wheel, and produces the "Apply Slight Turning Force to Wheel" nag, which is corrected by doing exactly what the warning says to do.

2. Cabin Camera induced nags. This type of nag is triggered by the cabin camera sensing that your eyes aren't on the road, and produces the "Please Pay Attention" nag, which is corrected by looking straight ahead at the road. Applying turning force to the wheel is NOT the corrective action for this nag. If the system is not able to correctly determine where you are looking, the nag will not clear itself, and will eventually end up with the red hands of death and a strike.

Also note that (obviously :) ) the 2nd type of nag requires the car to have a cabin camera installed and working. Non cabin camera cars are only capable of producing the steering wheel nag.

Please note the difference in wording of the nags, as that is the way to determine what type of corrective action is needed to remove the nag.
Good points, but this presumes the car isn't misbehaving. The cabin camera error seems to be coincidental with the strikes occurring. Just recently on my Model 3 after getting the cabin camera not availble error, I got repeated apply turning force messages, which would recur less tha 10 seconds after applying force. After 3-4 consecutive errors, I got a strike, and it was number 5. Thinking back, strikes 4 adn 5 happened after getting the cabin camera error at the beginning of the journey.
 
So, there's this one road that I use pretty frequently going back and forth to work. No markings. Three or so hills where one can't see what's coming over the top of the hill. People going about 35-40 mph, closing speeds 70 to 80 mph. Fairly straight road, no curves until near the end of this segment, and those curves are minor.

Sane drivers hug the right side of the road and keep an eye out.

FSD-b, if there's no traffic oncoming, drifts to the center of the road. On the uphill, that's frightening, for obvious reasons.

If there is traffic oncoming, FSD-b will drift to the right, leaving plenty of room. In the narrower portions of the road, when this happens, it occasionally freaks out and starts acting like it's avoiding a collision (driving sharply to the right, hitting the brakes hard) before recovering.

If there's a line of cars whom one is following, it sort of follows the line of cars.. but still drifts further to the left than the bunch of them, but doesn't go over quite as far.

So: In heavy traffic, both ways, it's relatively safe because it sees the oncoming cars and stays in lane. In very light traffic it'll give one a heart attack.

In the one area with curves, there's bushes and such on the right. Sane drivers purposely miss those bushes with a most a foot to spare because approaching traffic can't see one and one doesn't know if one is being approached by a Mini or a Hummer. I never let FSD-b drive there, I want to live.

Wish there was a way beyond hitting the "Record" button to let Tesla know about these particular, repeatable, bad spots. And if one literally can't drive through there on FSD-b, there's no way of providing feedback to Tesla to let them know They've Got A Problem.
 
i dont live in a city area but honestly the updates have been ever so slight incremental improvements at best. Compare today vs when i first got the beta, obviously drastic improvements. But the 3+ month delay in getting any kind of vehicle system updates is getting dry. Been dealing with the dashcam bug ever since 10.12, and unplugging the USB every other drive is starting to wear on me.
 
I had a weird situation today. FSD Beta just decided to take a small detour for apparently no reason.

See the red path I drew on the map. Instead of continuing straight on Chamberlain St, FSD Beta took a right on E Chandler Ave then a left on N Sharon St, then another left on E Milner Ave and then a right to get back on Chamberlain St. It is weird because there is no construction or obstruction on Chamberlain St. I don't see why the car would take a detour, especially for such a short distance. The only thing I can think of is that there is a map error where the car thinks that the road is closed at that one spot on Chamberlain St. Or maybe because of my setting in nav to take detours to save time, the car thinks that the roads marked as avenues have a higher speed limit and therefore the car thinks that path is faster.

cK3I4ql.png
 
  • Informative
Reactions: VanFriscia
I had a weird situation today. FSD Beta just decided to take a small detour for apparently no reason.

See the red path I drew on the map. Instead of continuing straight on Chamberlain St, FSD Beta took a right on E Chandler Ave then a left on N Sharon St, then another left on E Milner Ave and then a right to get back on Chamberlain St. It is weird because there is no construction or obstruction on Chamberlain St. I don't see why the car would take a detour, especially for such a short distance. The only thing I can think of is that there is a map error where the car thinks that the road is closed at that one spot on Chamberlain St.

cK3I4ql.png
Hmm. Collecting data, Google Survey Car style?

My way back and forth to work can, efficiently, be on interstates (longer but faster) or local roads (shorter but stoplights and more turns). On Any Given Day, it's a rough flip of the coin as to the direction that Nav, be it the one in the car, Waze, Apple Maps, or Google Maps will think is the best way to get from A to B. Especially when traveling on the local roads.

In my work history, I've had the fun and delight of designing (with a team) large ASICs; I was the guy who drove Synopsys most of the time, a VHDL-in, gates and layout-out package. When one is talking millions of gates and utterly ridiculous numbers of design choices built into the silicon libraries, throwing random numbers at the design and its constraints, then repetitively optimizing the design has been, for decades now, the way to go. But, when running the optimization software, it's virtually impossible to get the same results twice. Yeah, one will get working chips, but the design will vary at the small scale. They'll probably be close for two different runs, but highly unlikely to be the same.

Routing algorithms that send cars across the landscape have similarities to the above optimization software. The tricky bit is that if a road never has a car on it at a particular day and time, then the speed and distance of that road isn't going to end up in the routing database. So.. I strongly suspect that Google/Tesla/Apple/Waze/etc. will send $RANDOM1 cars down $RANDOM2 paths from time to time to Collect Data. Yeah, the car experimented with (!) might have a longer trip; but across the zillions of cars that have to travel the landscape, doing this might reduce overall the time spent traveling.

Although, one has to wonder. Suppose that a punter happens to travel across a road that, randomly, has no other traffic on it? Or that road has Every Truck In The Universe doing U-turns for some reason. Hence, the data could be really noisy. Wonder if the whole thing might get.. well... unstable.

Aren't feedback loops fun?
 
I had a weird situation today. FSD Beta just decided to take a small detour for apparently no reason.

See the red path I drew on the map. Instead of continuing straight on Chamberlain St, FSD Beta took a right on E Chandler Ave then a left on N Sharon St, then another left on E Milner Ave and then a right to get back on Chamberlain St. It is weird because there is no construction or obstruction on Chamberlain St. I don't see why the car would take a detour, especially for such a short distance. The only thing I can think of is that there is a map error where the car thinks that the road is closed at that one spot on Chamberlain St. Or maybe because of my setting in nav to take detours to save time, the car thinks that the roads marked as avenues have a higher speed limit and therefore the car thinks that path is faster.

cK3I4ql.png
If the car thinks that Sharon St and Chamberlain St have the same speed limit, then no speed on the Avenues can make the detour faster than straight through unless the car thought there was a slowdown on the straight path.

However, if there is a map error, such that the Chamberlain St. segment between Chandler and Milner was not contiguous, then the nav system would have no choice but to route around the disconnect. I had a desktop mapping program once that had issues like that. Some road endpoints were off by a few feet, making for some strange routes. Fortunately, that software allowed you to add road segments to fix problems like that. No such luck here.
 
If the car thinks that Sharon St and Chamberlain St have the same speed limit, then no speed on the Avenues can make the detour faster than straight through unless the car thought there was a slowdown on the straight path.

However, if there is a map error, such that the Chamberlain St. segment between Chandler and Milner was not contiguous, then the nav system would have no choice but to route around the disconnect. I had a desktop mapping program once that had issues like that. Some road endpoints were off by a few feet, making for some strange routes. Fortunately, that software allowed you to add road segments to fix problems like that. No such luck here.

I suspect there is some kind of map error or speed limit error. The detour is definitely not faster. E Chandler Ave is a residential street with no lane markings. The speed limit is like 15 mph. Chamberlain St has a speed limit of 25 mph, in some places it is 40 mph. Taking the detour through a residential area at 15 mph is not going to be faster.
 
I suspect there is some kind of map error or speed limit error. The detour is definitely not faster. E Chandler Ave is a residential street with no lane markings. The speed limit is like 15 mph. Chamberlain St has a speed limit of 25 mph, in some places it is 40 mph. Taking the detour through a residential area at 15 mph is not going to be faster.
You stated that FSD took this route. Did the nav system plan it that way? If so, and there was no traffic issue shown on the map at the time, then it certainly looks like a mapping issue.
 
You stated that FSD took this route. Did the nav system plan it that way? If so, and there was no traffic issue shown on the map at the time, then it certainly looks like a mapping issue.

I am not sure if the nav planned it that way. I just noticed when FSD Beta suddenly slowed down and put the turn signal on, that it wanted to turn right on Chandler. There was no traffic so I don't think it rerouted to avoid traffic. Normally, I would cancel FSD Beta to go straight but I figured I would let it try and see what happens. I was curious how FSD Beta would handle those residential streets since I've never ben on those roads before. It ping ponged a bit since there are no lane lines but it handled the turns at the intersections pretty well. It crept forward and made the turns pretty assertively.
 
I had a weird situation today. FSD Beta just decided to take a small detour for apparently no reason.

See the red path I drew on the map. Instead of continuing straight on Chamberlain St, FSD Beta took a right on E Chandler Ave then a left on N Sharon St, then another left on E Milner Ave and then a right to get back on Chamberlain St. It is weird because there is no construction or obstruction on Chamberlain St. I don't see why the car would take a detour, especially for such a short distance. The only thing I can think of is that there is a map error where the car thinks that the road is closed at that one spot on Chamberlain St. Or maybe because of my setting in nav to take detours to save time, the car thinks that the roads marked as avenues have a higher speed limit and therefore the car thinks that path is faster.

cK3I4ql.png
OSM shows Chamberlain St as contiguous but TomTom shows a gap.
 
OSM shows Chamberlain St as contiguous but TomTom shows a gap.

Thanks. That explains it.

I have no idea why TomTom would show a gap there. I've driven on that road many times for years and I don't remember there being a gap. How old are the TomTom maps? LOL.

Also, FSD Beta does not seem to recognize the left turn lane on E Indiana 42 when turning left onto Chamberlain St. FSD Beta stays in the right lane and makes the left turn from the right lane instead of the turn lane as you can see from the path I've drawn. IMO, another example of obsolete maps.

vnezW3z.png


I wish Tesla would use crowdsourcing from the Tesla fleet to update maps like Mobileye does. Teslas drive on these roads all the time. That would be a great way to keep these maps up-to-date and fix these issues.
 
:eek:
Plaid did poorly. Personally I would not show myself traveling at 100mph in a video posted on the interwebs. But arguably not the greatest risk in this video. Hopefully 10.69 will fix everything. I am most curious about whether the car will understand when its vision is obscured (the closest call here).

Started off well enough but took a hard nosedive after the first few attempts. Also that one instance where they reset it too close to the stop sign and it just goes for the turn without even stopping is pretty insane behavior. Stop sign or not it should know better o_O