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

My Tesla Learned to Autopark in my Driveway (Video)

This site may earn commission on affiliate links.
So for the back story, every night I back into my driveway in the exact same manner. After a few weeks of doing this, and seemingly after an update to 2019.12, the car began offering to auto park into my driveway at times. I didnt let it do it out of fear it would hit the curb but tonight I let it and it was FLAWLESS.

The car isn't even supposed to offer auto park if it doesnt have 2 other vehicles to park between, correct? I believe my car has learned my action.

It is HW3 2019.16.3 FYI. Unfortunately I didnt get the entire process on film because it didnt hit me to record at first.

 
So for the back story, every night I back into my driveway in the exact same manner. After a few weeks of doing this, and seemingly after an update to 2019.12, the car began offering to auto park into my driveway at times. I didnt let it do it out of fear it would hit the curb but tonight I let it and it was FLAWLESS.

The car isn't even supposed to offer auto park if it doesnt have 2 other vehicles to park between, correct? I believe my car has learned my action.

It is HW3 2019.16.3 FYI. Unfortunately I didnt get the entire process on film because it didnt hit me to record at first.


Pretty cool. Do you have EAP? or AP + FSD?
 
Not true, they are learning. How Are Tesla Vehicles "Learning" To Drive By Themselves?
Software updates enable new abilities and change some behaviors but the learning is going on all the time.


This I suppose gets into a pedantic debate about what "learn" means...


The cars don't learn in the sense of "gain info and gaining that info changes their behavior".


If they did their behavior would change without a software update. But it doesn't.


The cars gather data to feel back to the learning AI running at Tesla- which then uses that data to train the NN back at HQ, and then they push out an updated NN (or other code changes) to the fleet.



Back to the OPs post-

After a few weeks of doing this, and seemingly after an update to 2019.12


If the cars behavior genuinely changes, the "weeks of doing this" did not change the cars capabilities or behavior. The SW update did.
 
  • Like
Reactions: Common_Loon
Someone needs to come up with the right answer here, does the Tesla learn or what? If it’s like Sky-Net we are doomed!

Fred


I just gave you the right answer. The car gathers info. It does not change behavior without a software update.

Doing otherwise would be a completel nightmare for troubleshooting since you'd have 500,000 cars all acting differently based on their personal experiences.


Elon has confirmed this on twitter previously, and I expect Karpethy or one of those guys has in one of the various calls/presentations but too lazy to find it right now.
 
  • Like
Reactions: davewill
The car isn't even supposed to offer auto park if it doesnt have 2 other vehicles to park between, correct? I believe my car has learned my action
I think the vehicle cameras are detecting the edges of your driveway to define the parking area. Apparently the recent firmware update you received included improvements to Auto Park.
 
Can fleet behavior improvements/changes be pushed without an official firmware update? Or is the car truly stagnant between updates? I’ve certainly noticed subtle improvements with my 3 that occur between updates.

Seems people notice various activities on their WiFi router associated with their 3 even when no official firmware sent.
 
the FSD investor day talk cleared up a lot of misconception about how tesla's fleet learns, and to what extent they're employing the neural net accelerator in the car. as expected, they're employing them for the purposes that they're currently good at: image recognition and classification.

the hardware on the car on performs inference, no training. this means it doesn't 'learn' anything on its own. it can feed samples back to a server that are then trained on. once lots of this training has occurred and a new model has been tested and validated it is deployed via software update. the car does not 'learn'. it can provide samples for training offline.

regardless, the car will never 'learn' to plan paths better because path planning has nothing to do with the neural nets. unless they were intentionally misleading, it would seem that all path planning (and literally everything that isn't object recognition or classification) is done by another 'layer'. another way to think of it is that the neural network helps to identify things and where they are in space... it's an entirely different system that handles how to respond to these things. thus, 'learning' will make it better at recognizing and classifying things. it will do nothing at all with that information.. that's still hard-coded in the above layer so to speak.

elon did mention that over time they want to bring more functionality out of the higher layers and down into the neural nets. this is true of basically everyone working on machine learning today... more generalization is something we all want to accomplish, no surprise there.