Am I actually teaching?

[Jtrader]

Can anyone share insight into how the Neural Net actual learns to drive from ME? I would like to contribute if I thought it would make a difference.

Does it learn from:

• My every day driving?
• My driving on Autopilot/Navigate on Autopilot?
• Me manual overriding Autopilot?

Does it matter if I have HW2.0 or does it need HW2.5 or 3.0?

Does a human observe every instance of a manual override or does the Neural Net record and learn from every manual override? If it does learn something - does it come out in the next update?

Prior to several updates ago, it used to dive into the same turn lane every day. I would have to steer it back into my lane thereby cancelling Autopilot. It doesn't dive anymore and stays in the lane it's supposed to. Did it learn to not dive into the turn lane at this specific spot or did it learn in general or is it just a coincidence?

I'm not sure I even know enough about it to ask the right questions.

Thanks in advance for any insight.

 

[ucmndd]

Your car sends lots of data back to Tesla, which is then fed into machine learning algorithms to modify/improve the neural nets in the next release.

So yes, you are helping to train the fleet just by driving.

 

[ItsNotAboutTheMoney]

Your car sends lots of data back to Tesla, which is then fed into machine learning algorithms to modify/improve the neural nets in the next release.

So yes, you are helping to train the fleet just by driving.

As they said at Autonomy Day, it's more that "errors" are identified from the data, and then Tesla requests matching data from the fleet. So there's a probability that a Tesla helps, but no guarantee.

 

[brianman]

Recommend for OP: Watch the Autonomy Day presentation.

 

[Jtrader]

Recommend for OP: Watch the Autonomy Day presentation.

I watched but it doesn't say whether I have to manually override or not.