Initial 1000 HW2 cars getting AP software 12/31/16

[dhanson865]

As one of those who'll be investing my hard-earned savings on a Model 3, I'd disagree.

The Model 3 has probably at least a year's worth of backlog to go through. People who didn't even see the car before depositing money, followed by people who are waiting 1+ years for actual delivery of the car.

By the time your average non-tech-enthusiast Joe Schmoe will even consider buying a Model 3, it'll have well over a year of public "beta testing" already done. Given that they also have stated new tech will come to the S/X/expensive cars first, I doubt you'll ever see beta testing by a non-enthusiast, lower-income individual.

I don't know about that, I'm a TSLA stockholder, techie, enthusiast, heck what Apple woud call a tech evangelist but I don't have a Model 3 reservation yet.

I know people with more money than me that do have a reservation but don't know squat about technology. One I share a plug with at work doesn't know that its less safe to pull the AC plug out vs disconnecting the J1772 connector first while a car is charging (She has a Leaf now).

I think people that know very little about tech will buy a Model 3 just because it's better than 99% of the ICE cars on the market.

 

[MarcusMaximus]

After 10 minutes you see all that? I havnt noticed anything after 2 hours of driving

I've driven for over 9 hours since the update and still get nothing. There doesn't seem to be much in the way of a consistent method to get the camera calibration to finish. I called customer support last night(actually about a separate issue I'm having with the driver side mirror) and they didn't really have an answer either, even just in terms of types of driving to try to get it to finish.

 

[2BEnhanced]

I drove a 30 mile, mixed street/freeway trip. Nothing active the entire trip. Parked. Twenty minutes later drove again and all was active. TACC worked great. No opportunity yet for low-speed AP. I sure hope they up the AP speed soon to make it something we can use.

 

[hill]

Got the update last night. We only have 60 miles on the car because it had to go back to the shop. Sigh, it flagged all sorts of errors with the air suspension proximity sensor & timeout issues etc. That's all fixed now. But because it's just sitting in the garage - the cameras are likely trying to calibrate, but I don't know how well that can happen sitting in a dark garage. Certainly don't feel like racking up miles just to help calabration, only to get nothing.

.

 

[MarcusMaximus]

Got the update last night. We only have 60 miles on the car because it had to go back to the shop. Sigh, it flagged all sorts of errors with the air suspension proximity sensor & timeout issues etc. That's all fixed now. But because it's just sitting in the garage - the cameras are likely trying to calibrate, but I don't know how well that can happen sitting in a dark garage. Certainly don't feel like racking up miles just to help calabration, only to get nothing.

.

At least according to the reps I've talked to, having it parked in a garage won't do anything. The camera calibration only occurs while driving.

 

[3Victoria]

@Ulmo do you know anything about how the system works? Tesla has said EAP is got onramp to off ramp driving on highways. FSD is supposed to go from house to office with no human interaction. That means FSD has to deal with streetlights, stop signs, crosswalks, and many use cases that don't apply to a divided/limited access highway.

Your description of what you think the team is doing certainly does not reflect my experience in working with control system software development teams.

Agree. @Ulmo's description is not only unlikely but bordering on slanderous. Nobody will be planning ways to crash the car, and intimating such is irresponsible.

 

[IAMGQ]

AP2.0 was definitely rushed for Q4 sales! I see a few responses condoning Tesla for simply making AP 2.0 early adopters drive a car with software in Beta in the name of (Fleet Learning). NO! Fleet Learning is Tesla taking their car pre-go live with AP2.0 and driving it around putting it through different scenarios prior to releasing it to the public. If Tesla wants to release beta software to the public then they should do it the same way Apple does iOS beta which is releasing the software to AP 2.0 enabled cars for a fee. While other AP 2.0 capable cars waited until more (Testing/Fleet Learning) was done and released the 8.1 software. But what do i know? Im only a software engineer.

Tesla's overpromise and underdeliver to create excitement to only let people down will be their downfall. I would rather them be modest and let the car do the talking. Dont sell dreams and give nightmares.

 

[MarcusMaximus]

In fairness, when you sign up for EAP/FSD(and again when you activate it), it tells you outright that the former is a beta feature and the latter does not yet exist.

 

[HX_Guy]

In fairness, when you sign up for EAP/FSD(and again when you activate it), it tells you outright that the former is a beta feature and the latter does not yet exist.

Where does it state that EAP is a beta? I went to the Design Studio page and actually did a search for the word "beta" and nothing.

 

[techmaven]

AP2.0 was definitely rushed for Q4 sales! I see a few responses condoning Tesla for simply making AP 2.0 early adopters drive a car with software in Beta in the name of (Fleet Learning). NO! Fleet Learning is Tesla taking their car pre-go live with AP2.0 and driving it around putting it through different scenarios prior to releasing it to the public. If Tesla wants to release beta software to the public then they should do it the same way Apple does iOS beta which is releasing the software to AP 2.0 enabled cars for a fee. While other AP 2.0 capable cars waited until more (Testing/Fleet Learning) was done and released the 8.1 software. But what do i know? Im only a software engineer.

Tesla's overpromise and underdeliver to create excitement to only let people down will be their downfall. I would rather them be modest and let the car do the talking. Dont sell dreams and give nightmares.

I find that software engineers sometimes have the worst time in understanding how different Tesla' Autopilot software release is from "normal" software releases. I do not believe that it is possible to release software like Tesla's Autopilot in a fashion similar to traditional software where you can run a testing suite and do unit testing or standard regression testing, get the defect count down to a manageable level and ship it. Undoubtedly, they have many terabytes of video and run extensive simulations that are akin to regression testing. But that's not nearly enough to provide confidence of proper behavior in the field. Instead, after all of that testing, they then put the software into the cars for additional validation testing amongst a much larger driving pool, but in shadow mode so it doesn't take action. However, there is a limit to coverage of that kind of validation also. Simply, it isn't possible to develop this kind of software and do any kind of lab testing and pre-recorded video testing in order to achieve the necessary confidence in validation for general release. To top it off, some of the additional hinting is actually crowd sourced to further cut down on both false positive and false negatives, so each run is potentially a new scenario with updated map tiles. Certainly, add to that hardware sensor changes, and the amount of regression testing one can do with earlier recorded real world sensor data is limited at best. The best way is then to get the hardware into vehicles as fast as possible and have the new software run in shadow mode.

In any case, comparing a Tesla Autopilot software release with iOS beta release operating procedure belies a lack of serious thought towards the differences between the software and necessary and possible testing regimens as well as real world consequences.

Fleet learning at this juncture is about shared validation and updating the map tiles. The neural nets training certainly isn't being done in the vehicles... it's more likely deviations from expectations are recorded and uploaded, some of which result in data for updating the training, some of which updates the map tiles.

The only way a smoother transition could have happened is to provide both AP1 hardware *and* AP2 hardware in all the cars, have AP1 hardware actually functioning and blend in AP2. That's cost intensive, power intensive, and possibly the complication of switching and the physical installation and cross connection of the cameras for both make for an actually worse scenario.

 

[Shorty]

I'm noticing the Tesla app doesn't connect as quickly as it used to with the new firmware. Sometimes it doesn't even connect. Hmmmmm

Had that from day one (ie 4 years ago) with my S85

 

[MarcusMaximus]

Where does it state that EAP is a beta? I went to the Design Studio page and actually did a search for the word "beta" and nothing.

Sorry, you're right. On designer, it states that EAP is "expected" in December 2016, "subject to regulatory approval". Beta is denoted next to the button to turn it on in the car once you have the .185 update, and in numerous places on the screen that comes up after you receive said update.

Still, if you don't want to participate in the beta, that just involves not pressing the button next to the giant "BETA" label after receiving the update.

 

[IAMGQ]

I find that software engineers sometimes have the worst time in understanding how different Tesla' Autopilot software release is from "normal" software releases. I do not believe that it is possible to release software like Tesla's Autopilot in a fashion similar to traditional software where you can run a testing suite and do unit testing or standard regression testing, get the defect count down to a manageable level and ship it. Undoubtedly, they have many terabytes of video and run extensive simulations that are akin to regression testing. But that's not nearly enough to provide confidence of proper behavior in the field. Instead, after all of that testing, they then put the software into the cars for additional validation testing amongst a much larger driving pool, but in shadow mode so it doesn't take action. However, there is a limit to coverage of that kind of validation also. Simply, it isn't possible to develop this kind of software and do any kind of lab testing and pre-recorded video testing in order to achieve the necessary confidence in validation for general release. To top it off, some of the additional hinting is actually crowd sourced to further cut down on both false positive and false negatives, so each run is potentially a new scenario with updated map tiles. Certainly, add to that hardware sensor changes, and the amount of regression testing one can do with earlier recorded real world sensor data is limited at best. The best way is then to get the hardware into vehicles as fast as possible and have the new software run in shadow mode.

In any case, comparing a Tesla Autopilot software release with iOS beta release operating procedure belies a lack of serious thought towards the differences between the software and necessary and possible testing regimens as well as real world consequences.

Fleet learning at this juncture is about shared validation and updating the map tiles. The neural nets training certainly isn't being done in the vehicles... it's more likely deviations from expectations are recorded and uploaded, some of which result in data for updating the training, some of which updates the map tiles.

The only way a smoother transition could have happened is to provide both AP1 hardware *and* AP2 hardware in all the cars, have AP1 hardware actually functioning and blend in AP2. That's cost intensive, power intensive, and possibly the complication of switching and the physical installation and cross connection of the cameras for both make for an actually worse scenario.

I understand Autopilot Software releases is not the same as a traditional application software release. Fleet learning in lamen's term is just the autopilot software is learning as it goes (machine learning). Im saying before updates like 8.1/AP2 Tesla should be doing their own real world fleet learning prior to releasing the Beta (could have been done...Im not sure). In my opinion Tesla should have charged to use EAP Beta users now and it would be a lesser charge as its rolled out for all cars. That way they can save themselves from class action lawsuits as the software isn't actually "out" yet.

 

[Sonny Daze]

This is at the bottom of the Autopilot page:

 

[IAMGQ]

Sorry, you're right. On designer, it states that EAP is "expected" in December 2016, "subject to regulatory approval". Beta is denoted next to the button to turn it on in the car once you have the .185 update, and in numerous places on the screen that comes up after you receive said update.

Still, if you don't want to participate in the beta, that just involves not pressing the button next to the giant "BETA" label after receiving the update.

If it doesn't say the software is in BETA and all the features aren't there as described when you configure and buy the car in the design studio but when you receive the car it says BETA and the features you actually get when you purchase the car. Thats very bait and switchess

 

[IAMGQ]

This is at the bottom of the Autopilot page:

View attachment 208996

Tesla covered their arses with the statement "it will become available in December and "Rolled Out" through over the air updates". That essentially means we can release something small in december and not release anything else until Next year if we want too. That sucks but the game is never fair.

 

[HX_Guy]

Tesla covered their arses with the statement "it will become available in December and "Rolled Out" through over the air updates". That essentially means we can release something small in december and not release anything else until Next year if we want too. That sucks but the game is never fair.

Where does it say the part about December in the image posted?

Hell, it actually says Automatic Emergency Braking is a standard feature and it current is not.

 

[MXWing]

Tesla covered their arses with the statement "it will become available in December and "Rolled Out" through over the air updates". That essentially means we can release something small in december and not release anything else until Next year if we want too. That sucks but the game is never fair.

When you are in a relationship with someone like us and Tesla you start to figure out their idiosyncrasies.

Some positives here, some negatives here, exaggerations here, pleasant surprises there.

The whole thing is if you are still happy to drive their cars at the end of the day and trust they are doing their best to make things right.

No one else is close to Tesla when it comes to EV and automation so being the only game in town makes us have to put up with a lot of flaws as well.

 

[MXWing]

Where does it say the part about December in the image posted?

Hell, it actually says Automatic Emergency Braking is a standard feature and it current is not.

I had AEB kick off for me.. only problem was no car was in front of me when it happened.

 

[techmaven]

I understand Autopilot Software releases is not the same as a traditional application software release. Fleet learning in lamen's term is just the autopilot software is learning as it goes (machine learning). Im saying before updates like 8.1/AP2 Tesla should be doing their own real world fleet learning prior to releasing the Beta (could have been done...Im not sure). In my opinion Tesla should have charged to use EAP Beta users now and it would be a lesser charge as its rolled out for all cars. That way they can save themselves from class action lawsuits as the software isn't actually "out" yet.

Ah, no. There is fleet learning as in updates to the map tiles to help with false positives and false negatives in object recognition as well as other driving attributes. However, it isn't like the neural net is being trained by your car's Titan GPU. Instead, the exceptions and the parameters of those exceptions are sent back. They do their own real world fleet learning prior to the release of the software. The neural net is trained. What is going now is validation and I'm telling you that traditional software release procedures do not provide sufficient confidence to allow any rollout. Tesla could drive a test fleet for years and not get to that confidence level.

There is a big debate on whether or not it is possible to make a direct jump to Level 4/5 ADAS. It is quite possible that it will prove to be logistically impossible to arrive at Level 4/5 on a new hardware and software suite without first going through Level 2/3. Thus far, we haven't seen it for general purpose, drive anywhere vehicles. The only ones we've seen thus far have utilized non-scalable extreme mapping and thus restricted the areas in which those vehicles can travel. A general go-anywhere Level 4/5 ADAS may very well always have to go through an extensive validation system, with driving to the excess of hundreds of millions or billions of miles to achieve validation.

Even to replace the Level 2/3 functionality of AP1, they likely needed to have too many miles than what is logistically possible for an internal test fleet.