Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
  • Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
  • The final cut of TMC Podcast #29 is available now with topics time-stamped. We discussed the Tesla Cybertruck's expected 1 MW Ultra-Fast Charging capability, the Tesla Semi Delivery Event, the coming Model 3 refresh (project "Highland"), and more. You can watch it now on YouTube.

How to report FSD bug

If it did something you didn't like, press this. That's it.

IMG_1084.jpeg
 
  • Like
Reactions: michaela0416
I expect it will be checked if you press the button but if you email them with a explanation with time and vin# it would ensure that they understood why it was pressed.
I would highly doubt there is any attention paid to the email trying to explain what happened. Same with humans looking at video clips from button presses.

Not to say it never happens, but the way this stuff gets trained (in very basic terms) is sampling good behaviors and bad behaviors, and having the system infer what it should do based on what it has seen before. What the driver did while in control of the vehicle is the truth is. There's no console at Tesla where they can manually improve the behavior for a single reported use case. Tesla is looking at thousands of video samples to put into a system to improve certain behaviors.

So for example 10.5, 165k videos have been taken in and automatically processed/labeled to improve the car's understanding of how road lines and curbs are interpreted. This is what the release notes are really saying when it says... "Improved static world predictions (road lines, edges, and lane connectivity) by up to 13% using a new static world auto-labeler and adding 165K auto-labeled videos."
 
I would highly doubt there is any attention paid to the email trying to explain what happened. Same with humans looking at video clips from button presses.

Not to say it never happens, but the way this stuff gets trained (in very basic terms) is sampling good behaviors and bad behaviors, and having the system infer what it should do based on what it has seen before. What the driver did while in control of the vehicle is the truth is. There's no console at Tesla where they can manually improve the behavior for a single reported use case. Tesla is looking at thousands of video samples to put into a system to improve certain behaviors.

So for example 10.5, 165k videos have been taken in and automatically processed/labeled to improve the car's understanding of how road lines and curbs are interpreted. This is what the release notes are really saying when it says... "Improved static world predictions (road lines, edges, and lane connectivity) by up to 13% using a new static world auto-labeler and adding 165K auto-labeled videos."

I'm not sure how well they are using machine learning for driving policy as opposed to perception. The perception side (interpreting images) seems to be getting better reliably and they have an infrastructure, automated and with people (they hire a bunch of them) to discern what's in the images. That's the meaning of the 'improved static world predictions'. BTW this is the side Karpathy worked on (who knows if he is returning).

The driving policy OTOH sounds to me (no first hand experience) like a major problem that they don't have a consistent methodology to improve sequentially.

I suspect that one reason is the bossman who insists on "no maps". To be blunt, I think MobileEye has it right here. Not super high resolution centimeter scale lidar maps like Waymo, but reasonably compressed crowdsourced maps obtained by driving around. There is an extraordinary opportunity to learn what people do by measuring how people drive off AP/FSD. And build in specific semantic information about intersections and use it to make suggestions to the computer. Yes it should be able to guess and do something OK in a mapless scenario but it should do better with maps. Humans use experience and they watch what other cars do to inform them about proper driving semantics---where should I drive to go there, what are the implied traffic patterns in this intersection? We drive similar routes frequently and we're better having already driven them and seen them before.

The computer should do so as well, except it could conceivably aggregate this information across the whole fleet.
 

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


SUPPORT TMC
Top