What if Autopilot Software is Already Alive/Aware

[MsElectric]

So I was thinking about the whole Autopilot situation and I was wondering if the Auto Pilot software logic has already been pushed to all the cars and if the code is currently "alive" and actively validating what the software would do and comparing that to driver input. The software can then report to Tesla HQ about cases where the driver input deviated from what the software was intending to do to "learn" how to handle special cases as well as optimize the Autopilot Logic.

Would be pretty interesting if when they release Auto Pilot lane keeping functionality they come out and say that the software was being validated for weeks or months by running in "parallel" with driver input before the system ever fully came alive. By continuously refining the Autopilot Logic they can reach a point where driver input and autopilot logic are very closely matched.

 

[trils0n]

This is how they tuned the Forward Collision Warning / Auto emergency braking. Tesla would compare driver actions to what the dormant software would have done. I assume they are doing similar things for the rest of auto pilot.

 

[donkeykong]

I have wondered about this too. With lane keeping assist warnings, tesla should have already been validating the "software would have done" versus "the human actually did".

However, I don't think they would be sending that much data over 3G or LTE.

And on on my home wifi, I setup a transparent reverse proxy to see exactly how much data is sent. I couldn't see the exact data because they are using end to end encryption. I guess I could see it if I really wanted to , but I didn't. But given the size and frequency of my model s communicating back to the mothership, I don't think they are sending this theoretical autopilot data.

But sure would be a good idea, even if it's opt in.

 

[steph280]

Except when you are driving, you are unlikely to be using your home wifi.

There has got to be a reason why they allocate so much uplink bandwidth compared to downlink for 3G/LTE.

 

[donkeykong]

Except when you are driving, you are unlikely to be using your home wifi.

There has got to be a reason why they allocate so much uplink bandwidth compared to downlink for 3G/LTE.

To preclude that possibility, I retried the same experiment with my phone acting as a hotspot and Tesla linked to that wifi network. What I see is, the car checks into the mothership at predefined times through the day (2,3 times a day). It does not communicate all the time. I tried this on a 1.5hr driving day.

 

[ItsNotAboutTheMoney]

To preclude that possibility, I retried the same experiment with my phone acting as a hotspot and Tesla linked to that wifi network. What I see is, the car checks into the mothership at predefined times through the day (2,3 times a day). It does not communicate all the time. I tried this on a 1.5hr driving day.

They'd spread it out. Just needs to send some logs...

 

[mspohr]

The lane keeping assist warning has been active in my car since delivery in February and I've kept it on continuously. I've found it to be very accurate in alerting me when I am out of the lane. I'm sure this will be used for the auto-pilot steering so it's been reassuring to see it working so well already.

 

[donkeykong]

They'd spread it out. Just needs to send some logs...

The size of logs should change based on driving habits then. It didn't.

 

[gizmoboy]

The size of logs should change based on driving habits then. It didn't.

I'm not saying it's alive, but do keep in mind that the size of data may be quite small, and only statistically would it need to change based on amount of driving activity. It only needs to send details when your actions differ from its predicted actions for reasons it can't explain.

Having said all of that, based on what you've observed, if you forced me to bet, I'd bet it's not happening.

 

[MsElectric]

I'm not saying it's alive, but do keep in mind that the size of data may be quite small, and only statistically would it need to change based on amount of driving activity. Keep in mind that it only needs to send details when your actions differ from its predicted actions for reasons it can't explain.

Having said all of that, based on what you've observed, if you forced me to bet, I'd bet it's not happening.

It might not be active for everyone and it might be done only with prior permission/opt-in for a small segment of owners. Perhaps those in the beta program.

Even if .5 % of the owners took part in such a program that's probably all the data they are going to need to refine the software... For the rest of the cars it can simply be a percentage figure for each build of the Auto Pilot software as in, ver. x.yy predicted driver input to 86% accuracy.

 

[donkeykong]

It might not be active for everyone and it might be done only with prior permission/opt-in for a small segment of owners. Perhaps those in the beta program.

Even if .5 % of the owners took part in such a program that's probably all the data they are going to need to refine the software... For the rest of the cars it can simply be a percentage figure for each build of the Auto Pilot software as in, ver. x.yy predicted driver input to 86% accuracy.

Youre one smart MsElectric. Now the opt in beta program - totally possible.

 

[commasign]

I sure hope the car notifies Tesla whenever I have to hard brake during TACC, which is usually when a car slowly pulls in front of me but Tesla doesn't feel the need to brake until collision is imminent.

 

[MsElectric]

I sure hope the car notifies Tesla whenever I have to hard brake during TACC, which is usually when a car slowly pulls in front of me but Tesla doesn't feel the need to brake until collision is imminent.

Each time you do that the Tesla engineer responsible for that subroutine gets a mild electric shock in real time

 

[kanylbullen]

To preclude that possibility, I retried the same experiment with my phone acting as a hotspot and Tesla linked to that wifi network. What I see is, the car checks into the mothership at predefined times through the day (2,3 times a day). It does not communicate all the time. I tried this on a 1.5hr driving day.

The car could still be connected to Teslas servers through the 3G connection while being simultaneously connected to your Wi-Fi hotspot for normal usage such as browsing, radio and navigation.

Otherwise, if someone steals a Model S and connects it to a hotspot with a firewall preventing the connection to Teslas servers, it would make it impossible for Tesla to track the vehicle.

 

[donkeykong]

The car could still be connected to Teslas servers through the 3G connection while being simultaneously connected to your Wi-Fi hotspot for normal usage such as browsing, radio and navigation.
Otherwise, if someone steals a Model S and connects it to a hotspot with a firewall preventing the connection to Teslas servers, it would make it impossible for Tesla to track the vehicle.

You can fool GPS and 3G and Wifi by wrapping the mirrors in aluminium foil. That's where all the antenna's are. If that doesn't work a $100 GPS jammer is really all you need. Another $100 cell jammer, and for $200 you can take the Model S offline with a flick of a button.

Plus, for Tesla to pick a specific 3g interface preferentially in their code, would be a huge undertaking. Every line of code in every open source library they've picked would have to be split apart in picking the proper network preferentially. I really doubt they'd spend a lot of attention and time, while they have autopilot behind schedule, sucky mobile apps, a not open API, no open source, etc. etc. etc. Looks like hitech skills atleast on the computer side of things, aren't Tesla's forte. Sweet car, but the computer inside isn't being used to 10% of it's potential.

What I'm trying ot say is, what you're suggesting is an extremely low probability occurence.

 

[MsElectric]

You can fool GPS and 3G and Wifi by wrapping the mirrors in aluminium foil. That's where all the antenna's are. If that doesn't work a $100 GPS jammer is really all you need. Another $100 cell jammer, and for $200 you can take the Model S offline with a flick of a button.

Plus, for Tesla to pick a specific 3g interface preferentially in their code, would be a huge undertaking. Every line of code in every open source library they've picked would have to be split apart in picking the proper network preferentially. I really doubt they'd spend a lot of attention and time, while they have autopilot behind schedule, sucky mobile apps, a not open API, no open source, etc. etc. etc. Looks like hitech skills atleast on the computer side of things, aren't Tesla's forte. Sweet car, but the computer inside isn't being used to 10% of it's potential.

What I'm trying ot say is, what you're suggesting is an extremely low probability occurence.

You do realize Tesla offers perhaps 10X the computing capability of pretty much any other car available to buy via the touchscreen software interface right?

NVidia themselves said at their last show and tell how already Tesla is asking for more computing capability as they are close to maxing out the capabilities of the chip that was included in the car.

Yes, their software needs a bit of work but I have every confidence they will get it right. Have some faith

 

[donkeykong]

You do realize Tesla offers perhaps 10X the computing capability of pretty much any other car available to buy via the touchscreen software interface right?

NVidia themselves said at their last show and tell how already Tesla is asking for more computing capability as they are close to maxing out the capabilities of the chip that was included in the car.

Yes, their software needs a bit of work but I have every confidence they will get it right. Have some faith

Yup, I hope they introduce more and more cool features
- A better media player
- Better nav
- Better voice recognition and functions
- Autopilot (LOL)
- Better UI
- Dashcam
- App Store (doubt this makes economic sense for app developers, but ..)
- Choice of carplay android auto.
- etc.

 

[kanylbullen]

Plus, for Tesla to pick a specific 3g interface preferentially in their code, would be a huge undertaking. Every line of code in every open source library they've picked would have to be split apart in picking the proper network preferentially. .

Not really, it can easily be done on the network level, a simple route add command where connections to some ip addresses, eg Teslas servers, always choose the 3G connections default gateway, even though the Wi-Fi gateway has a lower routing metric.

 

[MsElectric]

Yup, I hope they introduce more and more cool features
- A better media player
- Better nav
- Better voice recognition and functions
- Autopilot (LOL)
- Better UI
- Dashcam
- App Store (doubt this makes economic sense for app developers, but ..)
- Choice of carplay android auto.
- etc.

A bunch of those features are coming. Be patient and don't beat up on Tesla too hard as I honestly feel they are doing the best they can with limited resources.

I doubt Tesla will ever offer a dash cam as it opens up a huge can of worms. What they should do is offer a spare 12V connection that you can plug into without messing with the wiring of the car and by most accounts it seems newer Model S cars now have an extra 12V power source that can be tapped in so the dash cam part of it is up to you and Tesla has done what they can in providing a useable power source.

 

[gizmoboy]

A bunch of those features are coming. Be patient and don't beat up on Tesla too hard as I honestly feel they are doing the best they can with limited resources.

I doubt Tesla will ever offer a dash cam as it opens up a huge can of worms. What they should do is offer a spare 12V connection that you can plug into without messing with the wiring of the car and by most accounts it seems newer Model S cars now have an extra 12V power source that can be tapped in so the dash cam part of it is up to you and Tesla has done what they can in providing a useable power source.

What can of worms would a dash can open?

They are lady have cameras looking fore and aft, right? Just record the last ~x hours of video to an SSD and offer the ability to offload it to USB and/or view it via app.