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.
Exactly - signal before you slow down so people are expecting it. If all cars were being driven by AI it wouldn’t matter as they’d see you slowing within milliseconds and react and brake just fine. Surprising the human driver behind me is a good way to get rear ended even though I (and FSD) wouldn’t be at fault it would be an avoidable accident.

Ultimately FSD needs to learn and be trained to be a defensive driver - anticipate what others might do and plan accordingly - be predictable and try to avoid hard braking. Fsd does stuff that would get our safety score massively dinged (hard braking last second to make a turn for example - that would incur a huge hit for hard braking and aggressive turning).
This might sound like I am disagreeing with you, but I'm not. For me it seems to try to be a defensive driver, but only for what is in front of the car, not what is behind it :oops:
 
  • Like
Reactions: RabidYak
The general definition of 'beta' is a product that is mostly complete and being subjected to wider testing by users for final debugging immediately prior to its final completion and release.

Simply because they may update a feature does not mean it is beta - any manufacturer is free to update and improve its products. The beta designation indicates that it is not complete or in its final, intended state in terms of functionality and bugs.

The problem is, they market and sell cars with features that are listed as beta for years.
 
Speaking as a (very) ex-Newtonite, and having learned how to drive up there, I completely agree with the "Masshole" stuff. It's not a new phenomena. Boston drivers have been terrorizing out-of-towners for a very long time. (And, now that I think of it, a California-designed FSD-b is pretty much the definition of an out-of-towner.)

As it happens, my Mother went to MIT back in the 1940's and used to drive around the area. (This is back when there was serious farmland all throughout Newton, for example. Which now raises a solid crop of suburban houses.)

She met my Dad in New York (that being where she and he were working, and where she was from). They then took a trip up to Boston and, as she tells it, she Warned Dad about the Boston drivers. He went, "As compared to the New Yorkers? Pshaw!" or something like that. Him being a Midwestern boy, where they lay streets out straight and at right angles.

When they got up there, she says he was like a soldier looking for a foxhole, ducking into side streets so he could hyperventilate and calm down a bit. And went, "Point. You were right." 😁

I've driven around NYC, and all I can say is that my Boston training paid off. I won't say which is more hostile; I encountered dumb-a$$ roads and aggressive drivers around NYC too.

But yeah, this whole region, including Jersey has mind-boggling idiocy when it comes to road design and logic. I have never seen its equal in California, my experience being limited to San Fran and bay area, and LA. LA has some weird stuff, and their traffic is insane, but it's way less intimidating to drive there first time than a Boston or NYC.

Some people consider San Francisco to be a challenge. I'll say it's way easier than a outlier city here in Boston metro, like Lowell, MA.

Basically the older the city, the crappier the road design is. too much legacy baggage to just redesign. Newer cities laid down their roads with some sort of big picture planning. Compared to the Northeast US, the south, midwest, and west/pac-nw are way friendlier to newcomers.
 
I'm of the opinion that if you really want to train FSDb on edge cases, the Northeastern US metro areas are where you need to focus. As far as sh*tty road designs and aggressive drivers go, the entire region has an abundance of both.

Chucks UPL is essential to get right, but that kind of maneuver is child's play compared to some of the sh*t we're forced to do in the Northeast.


During evening rush hour, most people entering this rotary are trying to cross the Boston University Bridge. Notice how 3 lanes (1 from Mem Drive, 2 from the rotary) merge together and immediately drops to a single exit lane. To make matters worse, that single exit lane to bridge has a stoplight, which logjams the entire rotary.

In summary, 3 lanes merging into 1, and that 1 lane is frequently stopped by the light. How do you think FSDb will handle that merge...........

Also, when the entire rotary is logjammed, and everyone in the rotary has the right of way, what's the likelihood of the Massholes in the rotary letting in a car willingly? FSD will sit there forever and never wedge itself in.
You can add some of our DMV drivers to your Massholes and Tesla in FSDb will scream and yell asking me to take control immediately with red hands of death.
 
But it's also possible to interpret Tesla's ubiquitous Beta labels as a commitment to continual improvement.
Definitely a unique take.

Beta means something hasn’t reached minimum viable product (MVP) yet. Iterative improvements are always possible after MVP, so the Beta designation doesn’t do much in that regard. In fact, it’s usually more profitable to get to MVP quickly and then eventually slap a “two dot o” (2.0) label on it.

We see this in practice as well. When was the last time Smart Summon got an update? An upcoming update was promised, sure, but that hasn’t happened yet. Even if it does, that’s far from “continual improvement.”

Leaving features in Beta is all about reducing liability, for the feature and its delivery timeline.
 
We see this in practice as well. When was the last time Smart Summon got an update? An upcoming update was promised, sure, but that hasn’t happened yet. Even if it does, that’s far from “continual improvement.”

I think Elon said they were aiming for the end of this month. So probably a few weeks until it makes it into a production branch.

But either way, Beta label or not, at least the continual improvement bit is baked into FSD. From the order page: "As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates."
 
  • Like
Reactions: FSDtester#1
Well after several days of flawless driving, the car went nuts yesterday. Turning right in a two lane right turn the car almost plowed into the car left of me. I had to disengage or hit the car next to me. The car acted bad all day, was the mother ship off line???

Today it is back to normal driving great.

One of our complaints has been entering a slower speed zone the car wants to drift, setting us up for a speeding ticket. I've been letting the car drive well into parking lots lately. I noticed the car enters at 25 MPH limit then drops to 15MPH. The drifting allows for a nice smooth slow down to your isle. I wonder if that was there intension.

Car is doing good making quick back to back turns (left then immediate right). It used to balk like crazy.
 
  • Like
Reactions: geekfam
I think Elon said they were aiming for the end of this month. So probably a few weeks until it makes it into a production branch.

But either way, Beta label or not, at least the continual improvement bit is baked into FSD. From the order page: "As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates."
Not really. How many features haven’t been updated in forever, beta or not? Yes, they are capable of being updated, but they aren’t being continuously updated. And there are non-beta features that get updated, too (eg blind spot camera).
 
  • Like
Reactions: sleepydoc
I don't take this route often but yanked the wheel back to the proper lane. A ticketable offence for sure if followed through on.

innovative_lane_change.jpg


Yesterday also had to yank the wheel after a quick FSDb move to STRADDLE the double yellow line on a frequently used road.

No audible disengagement sounds until I disengaged. No warnings. Has never done this before on this road.
I notice the blue line is starting to gray out but FSD still appeared to be engaged:

straddling_centerline.jpg


Along with increased phantom braking, which includes strange slowdowns on right turns that leave your car hanging in the lane you're turning from, I'd rather be driving 12.2.
 
  • Informative
Reactions: sleepydoc
I actually had FSDb 69.2.2 do something it's never done before - it chose the outside/wide lane when making a left turn. Prior to this it would always choose the inner most lane.
If only it would do this consistently. I have a left turn I make frequently onto a 2-lane road, followed by a right turn a block later, and the beta will alternate its lane choice despite fairly identical conditions. For me the main improvement of 10.69.x for such situations is it will travel a much shorter distance in the wrong lane before making a move, making it feel more natural.
 
Example of FSD deciding it should avoid the crossing car by turning towards the telephone pole. One wonders what would have happened if I did not disgengage? As FSDb testers know the feeling in the car is much jerkier then it appears on video. There was plenty of time to slow down.
 
Another example of FSD failing in 10.69.2.2. Until this build FSD did an excellent job of handling this 90 degree turn. Speed limit is 30 and FSD slows down enough to make the turn. Now it heads straight for the cones and if I don't take over will drive on the curb. 4 of the last 5 times it's failed the same way. The only time it was successful it was cloudy.

 
Another example of FSD failing in 10.69.2.2. Until this build FSD did an excellent job of handling this 90 degree turn. Speed limit is 30 and FSD slows down enough to make the turn. Now it heads straight for the cones and if I don't take over will drive on the curb. 4 of the last 5 times it's failed the same way. The only time it was successful it was cloudy.


looks like the shadow of a power line is making FSDb think it's a lane line.
 
looks like the shadow of a power line is making FSDb think it's a lane line.
I hadn't noticed that. Good observation. I've taken this curve over a hundred times in all light conditions and seems a bit of a coincidence it's only happened with 10.69. Maybe it's just the time of year and the shadow being cast.

I can see the police report now. "Driver (FSD) followed the shadow and crashed into the parked car":)
Kidding aside this is a pretty serious problem Tesla needs to solve if in fact it is the shadow.
 
Last edited:
If only it would do this consistently. I have a left turn I make frequently onto a 2-lane road, followed by a right turn a block later, and the beta will alternate its lane choice despite fairly identical conditions. For me the main improvement of 10.69.x for such situations is it will travel a much shorter distance in the wrong lane before making a move, making it feel more natural.
Yes, that’s been a consistent failure point for me, too. I’ve had it happen a couple more times, now, so it’s at least showing signs of improvement.
 
Another example of FSD failing in 10.69.2.2. Until this build FSD did an excellent job of handling this 90 degree turn. Speed limit is 30 and FSD slows down enough to make the turn. Now it heads straight for the cones and if I don't take over will drive on the curb. 4 of the last 5 times it's failed the same way. The only time it was successful it was cloudy.

Also the poor lane stripping doesnt help. It does look like it followed the shadow though as note above.
 
Another example of FSD failing in 10.69.2.2. Until this build FSD did an excellent job of handling this 90 degree turn. Speed limit is 30 and FSD slows down enough to make the turn. Now it heads straight for the cones and if I don't take over will drive on the curb. 4 of the last 5 times it's failed the same way. The only time it was successful it was cloudy.


Interesting. I don't think I've seen many black road lines so it's a curious thing to misinterpret. Maybe the NN generalizes lane colors.
 
Interesting. I don't think I've seen many black road lines so it's a curious thing to misinterpret. Maybe the NN generalizes lane colors.

I have seen FSDb use cracks in the asphalt in the middle of an unmarked road to represent a double yellow line in the visualization. Very reproducible on a particularly cracked road in my neighborhood.
 
Interesting. I don't think I've seen many black road lines so it's a curious thing to misinterpret. Maybe the NN generalizes lane colors.
There are some areas around Phoenix where they are trying black markings on white pavement to try and reflect more light and reduce solar heating.

Visual pattern recognition is something the human brain does remarkably well but is hard to train computers on. Color interpretation cal also be surprisingly difficult. (Remember the 'what color is this dress' meme?) If you teach the system to completely ignore all other colors besides yellow and white then you risk it missing things when the light changes or the stripes are dirty, etc. If you go solely on pattern recognition then you also miss a bunch. In reality, the brain probably uses color and pattern recognition combined with other context cues ('there's no way the line would go straight to the curb so that can't be a street marking') that the computer can't readily mimic.