Maybe we should starting counter steering for this to disengage when the car is veering too much. Would be great feedback for Tesla's learning machine. Seems like they could dial that veering down while still preventing curbing the wheels.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Wiki MASTER THREAD: Actual FSD Beta downloads and experiences
- Thread starter jebinc
- Start date
old pilot
Member
I have driven on back roads that are well marked with center and edge lines where there are curves from just a few degrees to 90%, the car has consistently come very close to or touched the yellow line on right turn curves. It is as if the care does not understand its own geometry in a turn. We all understand that the rear wheels will track to the inside of the turn and can judge how close to come to the center line with the left front wheel to avoid moving the right rear beyond the paved surface, keeping in mind that going over the yellow line, that divides opposing traffic, is a less safe option than having the rear wheels leave the paved road. The cars two computers do not seem to understand this.
KArnold
Active Member
Yes, works fine. Steps:
- Opt out,
- wait 1 to 3 hours until the SS in the app disappears,
- opt back in,
- after another hour or so your SS will be ready to start anew.
Cleaned up forward camera module glass. Confirmed failure. Pulled data logs. Cameras look good but will recalibrate cameras over the next day and see if it is resolved. If not resolved, it will get escalated. Happy with Tesla‘s response.
I've done this as well as many in this forum with good results (we have FSD beta).
When I want to "save my results" I usually take what I call a resetting drive which is not just a drive more than 0.1 miles (usually about half a mile), but ALSO goes at least 35 miles an hour.
Once I did a resetting drive, I would park the car (where there was wifi), and reset my phone.
#2 and #4 were each 10 minute events for me instead of hours. YMMV of course since all of this is "undocumented functionality"
Doreheel
Member
Question - does an individual car learn/modify its behaviors in FSD, or do all improvements come during software releases? I went smoothly through an intersection today that two days ago required intervention - may have been different lighting or traffic conditions, but it was a pleasant surprise. Overall, good first impression of FSD and can see how, with future patches, it can improve on some of the behaviors noted in this thread.
The "neural network" is supposedly learning all the time. There are indications from many TMC posters that FSD appears to be learning its way through local routes after 1,2, 3 or more times through. Which might be many times more quickly than it takes me to learn...
I saw the same where the car stopped at my street until I pressed the accelerator. After 3 times it made the turn itself on 10.3 but then had to be re-trained on 10.4
Unenrolled from 10.4 last Friday, yet the beta firmware remains on my Plaid. What does it take to replace this firmware with the current production build? An act of Congress? Divine intervention? Driving on FSD while not paying attention? What?! 
You could tray taking a turn on Beta and running into a car in an adjacent lane. Then file the following complaint to NHTSA:Unenrolled from 10.4 last Friday, yet the beta firmware remains on my Plaid. What does it take to replace this firmware with the current production build? An act of Congress? Divine intervention? Driving on FSD while not paying attention? What?!
The Vehicle was in FSD Beta mode and while taking a left turn the car went into the wrong lane and I was hit by another driver in the lane next to my lane. The car gave an alert 1/2 way through the turn so I tried to turn the wheel to avoid it from going into the wrong lane but the car by itself chopped the top ½ of my steering wheel off
and forced itself into the incorrect lane creating an unsafe maneuver putting everyone involved at risk.A little crazy Lane changing. Sorry for the LAZY iPhone video.
Crossing intersection onto a 2 lane + BUS lane. Beta hunting and weaving and hading for (correct) center lane and then goes into the BUS lane. Turns on left signal and moves over to center lane and turns signal off. Then heads over into the left turn lane with no signal as it is going through intersection. Need to make a left so OF COURSE Beta immediately moves back into the center lane (through another intersection) only to immediately move back into the left lane. If there is low traffic it will almost always repeat this same pattern.
Crossing intersection onto a 2 lane + BUS lane. Beta hunting and weaving and hading for (correct) center lane and then goes into the BUS lane. Turns on left signal and moves over to center lane and turns signal off. Then heads over into the left turn lane with no signal as it is going through intersection. Need to make a left so OF COURSE Beta immediately moves back into the center lane (through another intersection) only to immediately move back into the left lane. If there is low traffic it will almost always repeat this same pattern.
Marcus Adolfsson
Member
At AI Day they said they weren't doing continuous learning, wonder how that reconciles with poster's experience.
Most likely attributed to random variation and map enhancements which come outside of map version updates
Just to add what others have said. Your individual car doesn't "learn". If so you would have some cars that are very smart and some that are dumb. Also EVERY time your car comes to the same intersection it is seeing it "for the first time" and makes the decisions on the fly. Sometimes the outcomes are different and also the stating input is often different too.
So tomorrow you may come to that same intersection and Beta may fail it miserably or act differently in some other way.
willow_hiller
Well-Known Member
At around 1:09:10 Karpathy also talked about using the bird's eye view output for the vehicles to collectively build maps of intersections. So if they're utilizing that already, the vehicles could be creating their own local maps as they drive:
Yeah, probably uploaded post-drive and then merged into the map data for the area. Surprised they didn’t just start silently collecting this data sooner
Good to know that FSD Beta isn't geofenced in Canada. I don't know the official legality which may vary by federal/provincial regulations. You might want to verify before using it, may not improve cross-border relations to crash here.
We did hear that Smart Summon was not legal in BC, but I think that had more to do with nobody being in the car.
I think its more like adaptive filtering and building more accurate estimates/probabilities.
Situation: Lets say you get your new SW build and you get phantom braking due to stuff it sees on the road like patchwork alphalt and concrete.
- NN(s) calculates 80% chance of road debris and brakes. Driver overrides with the gas pedal. NN(s) understand it's estimation was incorrect and adjusts to get better estimates.
- Next time on same stretch it calculates 30% chance of road debris and doesn't brake. Driver doesn't override with brake. NN(s) understands its calculation was correct and adjusts to get better estimates.
I don't think the NNs are reset every time you get in the car and start driving as if you power cycled your computer. I think it retains it's at least some states and continuously adjusts.
Situation: Lets say you get your new SW build and you get phantom braking due to stuff it sees on the road like patchwork alphalt and concrete.
- NN(s) calculates 80% chance of road debris and brakes. Driver overrides with the gas pedal. NN(s) understand it's estimation was incorrect and adjusts to get better estimates.
- Next time on same stretch it calculates 30% chance of road debris and doesn't brake. Driver doesn't override with brake. NN(s) understands its calculation was correct and adjusts to get better estimates.
I don't think the NNs are reset every time you get in the car and start driving as if you power cycled your computer. I think it retains it's at least some states and continuously adjusts.
EVNow
Well-Known Member
Not how it works. Training happens only at Tesla - and in the car it is purely execution mode.
There has been some speculation that there is some configuration or something like that in the car that gets used. But it hasn't been confirmed ...
ps : To explain what I meant above better - we have all seen the car seems to perform better on the same routes after a while. Is that because we just get used to how it works - or is it because Tesla stores some secret configuration in each car that keeps track of some difficult situations and what worked, which can be used by the planner (not NN) later on ? There were reports on twitter a couple of years ago of someone testing this theory by using different cars with the same firmware on some roads. One car was "used to" the roads and performed better than a car that drove on those for the first time. I've to dig up those tweets - also I mentioned those tweets here.
Last edited:
Similar threads
