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

How to be a Good Steward of Your FSDBeta Access

What is the Tesla FSDBeta team looking for from us? What feedback is automatic (i.e. Tesla retrieves it to further refine or develop features), and what can we purposely do to support the software team's progress (in terms of feedback - i.e. "Video Record" button, email feedback, etc). I know that the pool is currently relatively small (but growing), so I'm curious what is most beneficial for us early access folks to aid in the long-term endeavor and success of this pioneering effort?

Post your ideas or informed responses here.
 
What is the Tesla FSDBeta team looking for from us? What feedback is automatic (i.e. Tesla retrieves it to further refine or develop features), and what can we purposely do to support the software team's progress (in terms of feedback - i.e. "Video Record" button, email feedback, etc). I know that the pool is currently relatively small (but growing), so I'm curious what is most beneficial for us early access folks to aid in the long-term endeavor and success of this pioneering effort?

Post your ideas or informed responses here.
It would first be helpful if we don't have to guess what's important to the Tesla FSD developers. Better comunication on their part would be an important first step.

We need to know how are they dealing with the data they are currently getting from the beta testers? Are they equipped to handle the data and feedback they are currently getting from the beta testers? Can they handle even more as they add more and more testers to the current pool?

Or are they already backed up with data? If so how, how is the data being currently handled? Is too much of it being sorted and dealt with by hand? Are computers ready now to do most of this work? Otherwise, anything more we send them may just sit, like all the ships carrying goods, currently off shore in California, that can't be unloaded.
 
  • Like
Reactions: SalisburySam

Mardak

Active Member
Oct 13, 2018
1,485
3,234
USA
Or are they already backed up with data?
Drew Baglino responded at 2021 Q3 earnings call: "At this point, it's not so much about how much data can we collect but how quickly can we process the data we've collected."

Not that they have "too much" data as they can easily ignore the data sent from vehicles if they want to focus on specific triggered data that they specifically asked vehicles to send back.

More communication would be nice, but overall it seems similar to production Autopilot in that Tesla doesn't need to specifically ask users to do anything special, and they have their ways to get the data they need. If at this point it's collecting "miles driven with FSD Beta," then expanding the fleet size of FSD Beta is a natural way to do it.
 
Drew Baglino responded at 2021 Q3 earnings call: "At this point, it's not so much about how much data can we collect but how quickly can we process the data we've collected."

Not that they have "too much" data as they can easily ignore the data sent from vehicles if they want to focus on specific triggered data that they specifically asked vehicles to send back.

More communication would be nice, but overall it seems similar to production Autopilot in that Tesla doesn't need to specifically ask users to do anything special, and they have their ways to get the data they need. If at this point it's collecting "miles driven with FSD Beta," then expanding the fleet size of FSD Beta is a natural way to do it.
Other than seeing no accidents in so many miles driven, the question remains are they already overloaded with data that has not been processed? Sending more data is worthless until it can' be unloaded and processed, even if they can be selective with what they unload. On the other hand, if computers are doing all of this, now, this might not be true. It's my understanding that the plan to have all this done by a mega computer is not yet been implemented.
 

Mardak

Active Member
Oct 13, 2018
1,485
3,234
USA
Sending more data is worthless until it can' be unloaded and processed
It's even worse than worthless right now as Tesla pays the cost to store the data. But that's probably an upfront payment for getting the benefits later when they can process a backlog of data.

My guess is that Autopilot team primarily focuses on their triggered data collection to train the specific focus areas. For example, they focused heavily on removing radar earlier this year, so they needed a lot of video to have the network learn how to accurately predict velocity.

As presented at AI Day, their compute clusters include FSD hardware so that they can recreate situations collected from the fleet. This is useful in determining how a neural network iteration would behave, e.g., how it behaved when the clip was recorded or how it would behave with a newer network.

I believe Tesla will process the backlog of data with the newest neural networks to detect if there are inconsistencies or jitter, e.g., pedestrian appearing and disappearing from predictions, and create targeted training data with the understanding from the full clip. Basically, this avoids training the network on things it already knows (avoid overfitting) and biases the predictions to be more consistently correct.

The old data is still useful as newer versions of the neural networks can regress some behaviors as well, so having an existing library of theoretically "FSD Beta driver thought this situation was interesting" can be used to quickly detect regressions and create new targeted training data.
 
It would first be helpful if we don't have to guess what's important to the Tesla FSD developers. Better comunication on their part would be an important first step.

We need to know how are they dealing with the data they are currently getting from the beta testers? Are they equipped to handle the data and feedback they are currently getting from the beta testers? Can they handle even more as they add more and more testers to the current pool?

Or are they already backed up with data? If so how, how is the data being currently handled? Is too much of it being sorted and dealt with by hand? Are computers ready now to do most of this work? Otherwise, anything more we send them may just sit, like all the ships carrying goods, currently off shore in California, that can't be unloaded.
Perhaps, not telling us what they’re looking from beta testers is precisely what they want to do. They don’t want people modifying the way they drive and then give them tainted data. The whole purpose is to get real world driving data so that FSD works as it should. I think that if Tesla were giving specific instructions as to what kind of things to do in order to provide data, those instructions themselves would skew the data
 
  • Like
Reactions: Taser8612

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