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

FSD Beta 10.69

This site may earn commission on affiliate links.
Just a short drive on 10.69.25.2 this morning, but two things that stand out to me as different:

1. More gradual horizontal movement for lane changes. It felt a lot more human-like on changing lanes and entering turning lanes. Prior to this version it occasionally felt like it was aiming for the curb while entering a left-turn lane, and straightened out at the last second. Now it's a gentle drift across lanes. I imagine this is something they are working on for highway driving.

2. Greater confidence on the edges of my driveway. The edge-detection on my driveway has always been a little bit fuzzy, especially with some drifts of dead leaves. Today it displayed thinner, solid red lines.
 
Got the dot 2 last night too. Did my initial short loop run this morning. I had one crawl episode where it felt like they may indicate an effort to speed up the crawl with short awkward throttle stomp with a hard brake. Ugh. But overall a couple of smoother lane changes. One right lane change with excess angle to the curb but possibly not as bad as before. The tight 90 degree right turns still generate too much lateral acceleration for comfort. One UPL resulted in excess braking and blipping throttle as it negotiated traffic openings. Braking still feels robotic and with a sense of desperate urgency which is not shared or appreciated by occupants. At least there's some areas of improvement.
 
  • Like
Reactions: willow_hiller
Crossing a single white line is discouraged, but legal. Crossing a double white line is not legal. Changing lanes close to, and even within an intersection is also legal.
Good input. I had to look it and it's a frequently asked question. But as others say on the national/freeway level crossing white seem to legal. I guess I don't need to be so anal waiting for the solid white line to end before safely merging onto freeways.

Things aren't so clear on the local/city level where solid white lines are frequently used to control traffic/limit lane changes near crosswalks, busy intersections, bike lanes, ...
 
I've seen a big increase in phantom blinkers - to the point that I frequently don't use FSD because it's so confusing to other drivers around me. Currently (v2022.44.30.5) FSDb turns on a blinker every time there's a turn lane, whether it takes the turn or not.
Now that you mentioned it, I had one in my short initial drive too. I had an upcoming distant lane change so it was too early for that but fortunately it was disabled shortly after initiated.
 
As FSDb was chauffeuring me home tonight from the brewery after a delicious beer, plus bean and cheese burrito, I watched carefully the behavior over this hill to this light, going eastbound. I had nothing better to do as I was enjoying the ride, the beer, and my general satiety.


The light was red and there was a lead car already stopped at the light. The light stayed red. I was traveling about 55mph (don’t know exactly).

The car did its typical massive slowdown as soon as it decided to slow, after cresting the hill, about three seconds after I would have eased off. The point at which it started slowing was fine. Later than I would, but sufficient for near full regen stop with no brakes.

It of course used regen and brakes at maximum stopping power, maxing out the bar immediately, and then it quickly followed it with *application of power* then of course quickly back to regen, all the way to a halt.

So I don’t know how a processing latency would explain that *application of power* towards a red light and a stopped car (immediately after massive braking). But a simple oscillation WOULD cause that.

Now “unstable control system” (part of the planner I guess - it may well be separate, though) is top of my list as most likely problem. It’s not clear to me that replacing the planner with NNs will help. Depends on the partitioning.

Sadly, I was not running the GoPro so I did not capture this.

Super interesting; first time I have noticed it. Have to crest a hill (note the massive downhill - this is how I knew to watch carefully!) with some downhill to a red light at high speed (55mph or so) to get that. The speed increases the oscillation amplitude of course.

10.69.25.1 (2022.44.30.5)
Cresting hill PB has been a huge problem for me with 10.69.25 (just got .2 last night and it's still an issue). My daily drive is pretty simple and therefore I see less of the navigation/blinker issues, but this one makes FSDb practically unusable in my area. I literally have to hover my foot over the accelerator at each hill because it brakes HARD, almost every time (and literally every time in some areas). This was new behavior in 69.25 so hopefully it'll get smoothed out in the next version.
 
I've seen a big increase in phantom blinkers - to the point that I frequently don't use FSD because it's so confusing to other drivers around me. Currently (v2022.44.30.5) FSDb turns on a blinker every time there's a turn lane, whether it takes the turn or not.
I have seen this in one particular lane location, it always blinks no matter what. What drives me crazy is the random blinking and then canceling on a straight road, with no obstruction, the right turn is coming up in a mile+, but it never shifts to the next lane, others behind me must be asking "make up your mind!". The other bad thing is when the leading vehicle turns into a shopping plaza, it follows closely, and slows down, I expect it will wait for it to complete the turn and then speed up, but it always jerks the wheel and tries to drive around it at the very last moment, making it unsafe as there may be traffic from the next lane, it just has to ease of throttle and be patience
 
Its valuable because it tells you how many times someone had to take over because the car couldn't complete the maneuver properly. You can compare that with actual known human accident rates of about 1 in 10,000 miles. For eg. Waymo is at about 1 in 9,000 miles - so about human level.

Ofcourse there is a lot of fuzziness in the data - just like with many real-world stats.
Looking forward to seeing the real world stats for 2022 in California. I'm guessing the accident rate for driverless Waymos is significantly less than 1 per 9,000 miles. haha.
Their published data for Chandler says they were hit once during 65,000 miles of driverless operation.
 
Cresting hill PB has been a huge problem for me with 10.69.25 (just got .2 last night and it's still an issue).

Note that this was not phantom braking. It was just slowing down for the stopped traffic in front, and the red light. It was after the crest of the hill that the slowing started taking place (and it really is not much of a crest anyway).

The remarkable thing was the application of power.
 
Last edited:
  • Informative
Reactions: VanFriscia
Good input. I had to look it and it's a frequently asked question. But as others say on the national/freeway level crossing white seem to legal. I guess I don't need to be so anal waiting for the solid white line to end before safely merging onto freeways.

Things aren't so clear on the local/city level where solid white lines are frequently used to control traffic/limit lane changes near crosswalks, busy intersections, bike lanes, ...
The single white line is used as a caution and to discourage lane changing in such situations, but not to make it illegal. If the traffic engineers wanted to prohibit lane changes in these situations, they would have specified a double white line.
 
Has anyone determined how we can manually edit map data with regards to lane selection? Is it open street map? Tom Tom looks like it is being used for speed limit data, as all the errors my car makes are also represented on Tom Tom’s speed limit data. But I don’t know how to edit Tom Tom’s lane data, or if you even can…
 
I sure hope FSD maps isn't editable by the public.. that would be the dream of hackers
You touched on a much larger problem in the auto industry - hacking. Every element of modern cars is electronic and computer controlled. As more and more cars get WiFi and NFC, with apps to control the car, hackers will find a way to exploit the systems.

From a Car and Driver article:

Ransomware for Cars Is Coming​

All of these attacks mean automakers have to take a proactive stance in this fight. Part of the automakers' defense strategy is to ask "white hat" ethical hackers to show them where the cars are vulnerable in exchange for monetary rewards or, in some cases, jobs. The famous hacker duo who took control of a Jeep Cherokee back in 2015 now work for Cruise, the autonomous-vehicle subsidiary of General Motors.

Michael Dick, CEO of C2A Security, an Israel-based automotive cybersecurity company, told the Free Press he expects the current trend of hackers holding digital data on computers for ransom to move to cars at some point. When this happens, driver will not be able to start their vehicle until they pay off the hacker or suffer the consequences. "There's no way around it," he said. "You'll have to get it towed and get all new software to start it."
 
The lane selection certainly is terrible, but I am still having a hard time with understanding why people have a huge problem with it. Certainly spurious blinking is embarrassing and annoying to other drivers, but I’ve found that I can effectively just do the lane positioning myself. And just cancel any unwanted changes (over and over again - and if I am alert, after just one or two blinks - it gives me something to focus on intently to stay alert, which is good). I’ve been generally running in Chill mode but not sure how much difference that actually makes to lane changes.

Certainly there is a LOT of unnecessary blinking and I would disengage if there were any significant traffic around, where it would be confusing or a hazard to other road users. But no one is going to be using FSDb when there is significant traffic, anyway. So not really a big deal.

Anyway, hopefully V11 will allow it to be disabled completely! That is the dream.

Turn lanes are generally pretty poorly handled so I can understand complaints there, and it’s not clear to me that manual signaling really addresses those issues of actually getting in the turn lane.
 
Last edited:
  • Like
Reactions: Dewg
The lane selection certainly is terrible, but I am still having a hard time with understanding why people have a huge problem with it. Certainly spurious blinking is embarrassing and annoying to other drivers, but I’ve found that I can effectively just do the lane positioning myself. And just cancel any unwanted changes (over and over again). I’ve been generally running in Chill mode but not sure how much difference that actually makes to lane changes.

Certainly there is a LOT of unnecessary blinking and I would disengage if there were any significant traffic around, where it would be confusing or a hazard to other road users. But no one is going to be using FSDb when there is significant traffic, anyway. So not really a big deal.

Anyway, hopefully V11 will allow it to be disabled completely! That is the dream.
The issue as I see it is that there are two camps. People who use the ADAS functions as an L2 assist, and people who think the software should be L4. If people treat the ADAS functions as L2, then you're totally right - it's easy to use and control as you're the one driving and the car is assisting you with mundane tasks (just like old-school cruise control on freeways). If people are treating it as L4, then it's a terrible experience and frustrating for everyone involved in the drive.
 
  • Like
Reactions: old pilot
The lane selection certainly is terrible, but I am still having a hard time with understanding why people have a huge problem with it. Certainly spurious blinking is embarrassing and annoying to other drivers, but I’ve found that I can effectively just do the lane positioning myself. And just cancel any unwanted changes (over and over again - and if I am alert, after just one or two blinks - it gives me something to focus on intently to stay alert, which is good). I’ve been generally running in Chill mode but not sure how much difference that actually makes to lane changes.

Certainly there is a LOT of unnecessary blinking and I would disengage if there were any significant traffic around, where it would be confusing or a hazard to other road users. But no one is going to be using FSDb when there is significant traffic, anyway. So not really a big deal.

Anyway, hopefully V11 will allow it to be disabled completely! That is the dream.

Turn lanes are generally pretty poorly handled so I can understand complaints there, and it’s not clear to me that manual signaling really addresses those issues of actually getting in the turn lane.

Probably different road conditions. Here, the lane changes are immediate, and you have to swerve it back to prevent an erroneous lane change. Couple that with the blinker, and people around you quickly get annoyed at the non-sensical behavior.

I also don't think "cancelling any unwanted changes over and over again" is something that I want to be doing, when the whole point of FSD is to reduce/eliminate driver input. It would be like saying "I don't know why ppl are so annoyed by phantom braking; it's so easy to override it over and over again with the accel pedal every time it happens."

On the positive side, phantom braking is no longer my top gripe for FSDb. It definitely was at one point, but it's improved a lot.
 
it's easy to use and control as you're the one driving and the car is assisting you with mundane tasks (just like old-school cruise control on freeways). If people are treating it as L4, then it's a terrible experience and frustrating for everyone involved in the drive

Just to be clear, it is still a terrible experience. I am just saying it is workable with interventions. Who wants to be flashing blinkers all the time?

I also don't think "cancelling any unwanted changes over and over again" is something that I want to be doing, when the whole point of FSD is to reduce/eliminate driver input.

No, absolutely not. It’s terrible.

I am just saying that for me at least, this seems controllable and unlike phantom braking I can actually mostly eliminate the behaviors (you cannot get rid of a phantom brake after it happens - analogously you can’t get rid of a spurious signal but that seems like much less disturbing). Unless there is a critical turn coming up and I am in the wrong lane, it won’t signal over and over again. If I take control and insist on the lane change at an appropriate time and place the vehicle in the appropriate lane, it doesn’t do tons of unnecessary signaling. And it’s nearly always in the correct lane with this approach. It still does do some signaling though, and that is super annoying for sure. And I think turn lanes aren’t great.

Just seems like less of an issue than a bunch of the other issues (surging forward at red lights, slamming on the brakes, not being able to handle stop signs, inability to handle road markings, etc.).

Again, hopefully this can be disabled completely. I can’t really think of another workable solution long term (next few years at least - this is an extremely difficult problem, as it goes way beyond perception, and requires planning up to several minutes in advance - immense sophistication and knowledge of history of traffic patterns is needed).
 
Last edited:
  • Like
Reactions: nvx1977
Yep, I experience the side approaching vehicle issue occasionally. For the worst instance I was driving home from Costco surprised with a very hard brake and items flew off the seat. The braking force applied was as if an accident was certain which was far from reality. Fortunately no one was following.

Tight roundabouts are tough in my area. Not sure if it's another challenged field of view issue but for some reason 10.69.25 doesn't have a handle on them and it's even worse for tight dual lane roundabouts where it stops plus adds confused lane changes. At times FSDb seems to punt and tries sharing a lane with another car already in the roundabout.
We have a lot of median separated four lane roads in our neighborhood. There are lots of instances where cars coming into the median area to prepare for a merge are assumed to be existential threats with very hard braking. Interesting problem, when you think about it. A car moving quickly into the median from one side of traffic COULD just continue and ram into you, but we humans assume other drivers will do the right thing. FSDb just sees it as a car moving quickly at an angle and speed that could create a crash if it didn't stop in the median area.