Forums

What's the endgame for AutoPilot 1.0?

What's the endgame for AutoPilot 1.0?

Before yesterday, I thought I had a pretty good idea what the endgame was for AutoPilot 1.0 and it was:

TACC would improve due to more effective use of RADAR to detect stationary objects
Some unnecessary braking would be removed as the result of fleet learning about overhead road signs
AutoSteer would be more reliable through improvement in its algorithms and through fleet learning
AutoSteer would be able to take a freeway exit analogous to the current lane change feature
AutoSteer would be able to make a freeway exit under navigation (8.1)

That list is an abbreviated version of the Tesla blog post, "Seeing the World in Radar"

Elon Musk has said that Autopilot was reaching its limits under the current hardware, and my expectation has been that the features announced for Firmware 8.1 describe what those limits are. I also had the expectation that the current Autopilot would come out of "beta" and that drivers could have the expectation that, barring some anomaly in the road or traffic, lane keeping would always work (no more "ping pong," no more "truck love," no more surprise lunges, and no more indecision cresting a hill), and TACC wouldn't run into anything. I also had the expectation (maybe the hope) that Summon would become at least as smart as AutoPark and do more than just back in straight.

Now that Autopilot 2.0 hardware is entering the fleet, are those assumptions still valid?

The Enhanced Autopilot feature, according to Tesla, "adds new capabilities to the Tesla Autopilot driving experience" and describes its features to include:

Your Tesla will match speed to traffic conditions, keep within a lane, automatically change lanes without driver input, transition from one freeway to another, exit the freeway when your destination is near, self-park when near a parking spot and be summoned to and from your garage.

What, if any part, of the Enhanced Autopilot feature will existing Autopilot customers get? Obviously we'll keep the ability to match speed to traffic, and keep within a lane. I am assuming that the ability to change lanes "without driver input" will not be available because the current sensor set can't detect a fast approaching car in the passing lane. How about transition from one freeway to another under navigation? No, because that's requires merging into a traffic stream which current sensors can't handle. Exit the freeway when your destination is near? That exactly what we were promised for 8.1 (exit a freeway under navigation).

I just wanted to get the discussion going on what you think the limits of AutoPilot 1.0 are and what you think we'll be seeing over the coming months.

rxlawdude | 20 October, 2016

Remember when Summon was supposed to take AP 1.0 equipped cars from the "garage to your front door?"

rxlawdude | 20 October, 2016

I share your concern that we will not see all promised functionality with the 1.0 hardware.

Haggy | 20 October, 2016

The car does most of what was promised from the beginning, except summon is more limited. But must did talk about the car's ability to read stop signs and traffic lights when AP was first introduced and it would be disingenuous to tell people that the car recognizes them but won't do anything with the data. That doesn't mean that the car will ever resume on its own from a stop sign. It doesn't necessarily mean that it would have to stop at either. Arguably, Tesla could merely put a dashboard indicator that traffic control exists, along with a disclaimer that it can't be relied upon. That way it might tell the driver of something he might miss, but he'd be expected to pay attention. Of course if it can notify drivers, it can stop. Tesla has to work out that part anyway for enhanced autopilot, unless a true limitation in processing power would make something impossible.

rjt | 20 October, 2016

the possible outcomes:

(i) What we have right now in AP features is the end of the road for AP 1.0 HW
(ii) Some selective enhancements with new SW on old AP 1.0 HW

I think (i) is the most realistic scenario. As I own a MS with AP 1.0 hardware, I am optimistic & hopeful we will get some new features.

Also AP 2.0 HW will get replaced with AP 3.0 HW at some point..

Change is inevitable.

Enjoy your Tesla's. AP 1.0 , or 2.0 or no AP, they are still awesome cars.

kevin | 20 October, 2016

BTW, someone with AP and the 8.0 Firmware--the they ever deliver the feature in the 7.1 manual that says the car will slow down when you turn on your blinker to exit from a freeway?

Pleasanton_Ca | 20 October, 2016

Keep in mind that anyone that is getting this now is paying an extra $5,000 (+), for software & hardware that they won't be able to use for some time. Software is still in development and will have to be approved...

Blue85DCalifornia | 20 October, 2016

AP 1.0 has reached as far as it will ever go. The development team has moved on

This is the way with all large software projects.

2015P90DI | 20 October, 2016

It will likely be the same as the P85D. When it was released, P85D owners were promised that the car would continue to improve in performance with software updates. With the exception of regular updates, those cars never got into the 2's for 0-60 time as was originally indicated. Instead, Tesla introduced Ludicrous mode and said if you want more, pay $10,000 more for it. Then came the 90 battery. That'll be another $3,000 please. Then came the 100 battery, that'll be another $10,000 please. Now I think it's $43,000 more to go from a 90D to a P100D.

AP 1.0 will fall into the same category. Other than changes likely for safety reasons, with 2.0 now on its way, don't expect any improvements to what it can do. Basically it's now, if you want more, that'll be another $8,000 please, after you fork out the dough for a whole new car that is.

stephen @newpor... | 20 October, 2016

Endgame is that I will love my car (S85 Dec 15) for as long as I keep it (average 7 years per car) and 5 years from now it will still be better than 99% of the cars on the road. I didn't buy it expecting the hover jet pack upgrade and certainly not for free.

martin | 21 October, 2016

@kevin:
AP1.0 cars will get none of the new features of EAP for all the reasons you mentioned.
So the endgame to me are minor updates, not upgrades any more. I will enjoy my pre facelift MS90D for the next 30 months and then see what my choices are...

tstolz | 21 October, 2016

Tesla continues to improve first generation cars even now .. continuous improvement after the sale is a way Tesla differentiates from the rest. Don't worry .. your car will continue to improve .. you bought a Tesla!

Silver2K | 21 October, 2016

AP 1.0 obviously needs some more refinements that are coming in 8.1.
As tstolz mentioned, they will continue to improve AP 1.0 as time goes on. I doubt very much they are going to just dump semi-autonomous software and leave it buggy.

jordanrichard | 21 October, 2016

What's the end game for us non-AP cars......? As Elon said in the Q&A, AP1 will continue to get better, just as the on-AP cars get more refined.

jhazelwd | 21 October, 2016

I'd be happy if they removed the annoying hands on wheel nag, but that's just me. I was wondering the same thing, as typically beta leads to GA, except Google who keeps things beta for a long long time.

thranx | 21 October, 2016

I would be permanently content with 1.0 if it would just learn to recognize and adapt to street lights and stop signs.
Certainly if it can read speed limit signs and react accordingly, stop signs should hardly be an insurmountable problem.

jhazelwd | 21 October, 2016

Stop signs scare me because it means you are using Autopilot on city streets. Without people recognition, that's a no go for me.

AmineTx | 21 October, 2016

@thranx For reading the stop sign I think the big challenge is once it stops, how does it determine when to go when there are 3 other cars in the intersection.

Haggy | 21 October, 2016

There are a number of things that Tesla can do with respect to the hand warning. They can analyze data to see if there's a difference in accident rate when comparing those who hold the wheel all the time to those who don't. That alone could be misleading since those who do might be more cautious by nature, but if there isn't a difference shown, that could be a relevant factor. Tesla could also track the shadow data, so for anybody who gets into an accident with autopilot off, they will have data showing what the car would have done, had autopilot been in use. They could compare situations where the driver overrode autosteer and see which of the drivers had their hands on the wheel for a significant amount of time before they did so, and compare it to those who had just grabbed the wheel, and see if that factors into anything.

It would be in Tesla's best interest if they could demonstrate that holding the wheel doesn't make anything safer, and that paying attention is the key.

Haggy | 21 October, 2016

Stop signs scare me because it means you are using Autopilot on city streets. Without people recognition, that's a no go for me.

There is people recognition, but Tesla tells you not to rely on it and to pay attention at all times. If they change autopilot to stop automatically at stop signs, I fully expect that they will make you tap a pedal to resume. Without cross traffic detection, it's the only way.

Run4Waffles | 21 October, 2016

@Haggy - "They could compare situations where the driver overrode autosteer and see which of the drivers had their hands on the wheel for a significant amount of time before they did so, and compare it to those who had just grabbed the wheel, and see if that factors into anything."

The problem with this is that the data wouldn't be all that accurate. Unless I'm the exception. When I have auto pilot on, my hands are always on the wheel. The problem is that I don't always apply enough pressure for acknowledgement and I end up getting the please hold the steeringwheel message.

kevin | 21 October, 2016

It would seem to me that AP 1.0 refinements are independent of the entire AP 2.0 development effort, so if it gets refined, it will an effort that has no application to future cars. We might as well start calling it "legacy autopilot".

Realure | 21 October, 2016

I think they owe the AP 1.0 purchasers a non-beta implementation of the features that were promised. Summon, blind-spot detection, auto park all work worse that I would expect in an alpha situation. The blind-spot detection is so bad it is dangerous.

kevin | 21 October, 2016

We have blind-spot detection? I thought it just detected curbs.

Nardypest | 21 October, 2016

I just got the Model S P100D back on September 22nd. I'd never driven a Tesla before I decided to order the new model the day Elon Musk made his announcement. I was given a test drive in the P90D about two weeks before I took delivery of my P100D which only solidified the confidence in my decision! Anyone who owns a Tesla understands how amazing an experience it is to drive. Give the updates...In reality we're all quibbling over the speed in which a car will update itself and become more fully autonomous. No car on the market can update itself! Furthermore, you're not going to find an car on the market that will give you the opportunity to go faster, quicker, or have more range by simply switching out a battery pack! That being said, no car can improve the experience like Tesla can based on an over the air update. I might be new to the Tesla game, but if you want a fully autonomous vehicle right away, just hail a cab! if you want Tesla to get it right, and keep us all safe while doing so, be patient! Tesla is currently leading the race when it comes to fully autonomous technology. Don't be discouraged in the mean time that you'll have to keep two hands on the wheel.

Nardypest | 21 October, 2016

Just give the updates some time, until then, enjoy the experience of owning a special vehicle.

Stiction | 22 October, 2016

I would like stop sign reading (or red stoplight detection) that just warns me I'm not slowing down fast enough, to deal with a sudden distraction on my part.

I don't expect autosteer to ever get good enough to keep hands off the wheel. And I'm not sure more ML will make things much better than we already have. There are limits to the HW and I think we're there.

Why are people so bent on taking hands off the wheel? You are asking for a delayed response to 'corner cases' which can, at the very least, prang your car. e.g. crap falls out of truck ahead, wind blows stuff into road, stalled car in road just around blind corner, deer jump in front of car, etc

Lots of these cases would be trivial for a driver paying attention with hands on the wheel, reflexes can kick in pretty fast!...with hands off the wheel you lose precious milliseconds any way you slice it.

I love TACC and think it's great....but frankly autosteer is not nearly as important to me. Mostly it's a 'weaving' preventer when fussing with the screen for a few seconds.

kevin | 22 October, 2016

I drive with hands off the wheel when I am on a well-marked, controlled access highway when there is almost no traffic. Where I live, the low-traffic scenario is fairly common. If there are going to be cars beside me, then I grip the wheel with two hands.

Not a day passes when I don't have to take control back from AutoSteer to prevent an accident.

SamO | 22 October, 2016

Tesla will soon be able to drive coast to coast in 40 hours without touching the wheel. Start a trip when you are going to sleep. By the time you wake up, you'll be 600-800 mile further along in your journey.

Compared to flying coast to coast, which takes 6-10 hours for flights with early arrivals and airport transfer.

70 hours for train service.

martin | 23 October, 2016

Go on dreaming, SamO...

TaoJones | 23 October, 2016

While I appreciate a good pipe dream and a nice, long road trip as much as the next wayward denizen, there is no way on this earth (or on any other planet, for that matter) that I am taking a snooze in a moving vehicle in which I am the sole human driver *and around which the majority of drivers are similarly human*.

Maybe in 10 years in a dedicated transcontinental robo-lane. Except by then we'll have regional hyperloops - or, tp put it another way... Vroomity.

kevin | 24 October, 2016

Since I wrote my original post, I realized that I had misunderstood the 8.1 freeway exit feature. I updated it with the correction.

It seems to me now that Enhanced Autopilot has only one advantage over the Autopilot convenience feature: the ability to decide when it's safe to merge into a stream of traffic. That is reflected in the ability to change lanes "without driver input" and the ability to transition onto another freeway. I guess another advantage is that it runs on the new hardware.

I think what's really happening with AP 2.0 is that Tesla has raised the price of Autopilot without giving much in additional functionality. That way they can fund the new hardware in all cars and full autonomy research. Once full autonomy arrives, the cost benefit turns heads over heals in favor of the new hardware and software.

mgtesla1 | 24 October, 2016

At this point I'd be happy if the 8.x auto-steer would become as smooth as it was in 7.x. The TACC is much enhanced (except the occasional false alarm), but the steering is uncomfortable.

kevin | 25 October, 2016

@ mgtesla1
I don't have 8.x yet, but i have not been happy with 7.1 Autosteer. I hope that with some fleet learning under its belt, 8.x will be better for both of us.

kevin | 25 October, 2016

It's interesting to read the initial announcement for Summon in 2015.

"During this Beta stage of Summon, we would like customers to become familiar with it on private property. Eventually, your Tesla will be able to drive anywhere across the country to meet you, charging itself along the way. It will sync with your calendar to know exactly when to arrive.

"The release of Tesla Version 7.1 software is the next step toward developing fully autonomous driving capabilities and delivering them through over-the-air software updates, keeping our customers at the forefront of driving technology in the years ahead."

It sounds like Tesla thought the original Autopilot hardware was all they needed.

Realure | 25 October, 2016

At what point will the class action lawyers start going after Tesla for overpromising and underdelivering. AP 1.0 will never be able to do blind spot detection and I predict AP 2.0 will never be able to full L5. Maybe owners should bring in their cars toward the end of their warranty demanding that these features work reliably and not in beta.

Haggy | 25 October, 2016

A lot of the "overpromise and underdeliver" had been hype itself. Much of it really has to do with missing deadlines, not failing to deliver what was promised. AP 1.0 still can't meet me at my front door, but the way it works now is actually more practical for me.

A lot of the "overpromise" statements refer to Tesla meeting delivery goals. What happens time and time again is that Tesla falls short, but ends a quarter producing cars at the volume needed to make cars at the promised rate. It's like running a marathon and not crossing the finish line when expected based on time estimates, but crossing the finish line running even faster than at the start of the race. Only it's a milepost and not a finish line. The snapshots don't line up with the promises, but if you look at how much Tesla missed by when they released numbers a full year ago, and look at the volume they are producing now, it shows that if you had used the 2015 numbers as a predictor, you would have been wrong.

With actual products, the Model 3 is on schedule, and the Gigafactory is ahead of schedule. The new promised ramp up for the Model 3 is more ambitious than the original plan, and for those who doubted Tesla's original production goal, they have been proven wrong time and time again when you look at how many cars Tesla has made compared to the wildest estimates when the Model S came out.

Blind spot detection is functional. It might not work the way I would like it to, but it would be hard to sue Tesla over it.

Elon Musk said in interviews the week that AP 1.0 hardware came out that it wasn't suitable for full autonomy and that would have to wait for a future generation.

kevin | 25 October, 2016

Actually Summon does meet me at my front door, but that is just an artifact of where my front door is.

Haggy | 25 October, 2016

From the way it was described, summon should be able to back out of my driveway and onto the (private) street in front of my house. It's more convenient to have it simply leave the garage and close the door, but I think it would impress others more if it really met me in front of the house on the street. However, I can think of many ways I would rather that Tesla have its engineers spend their time.

dpena23 | 27 October, 2016

There are two things I wish for my AP 1.

#1 Be able to detect traffic lights or at least stop signs.
#2 Smart enough to slow down during steeper turns on the freeway.

I was going about 60mph on autopilot on the I-110 here in L.A and the car went over the line to the other lane, slightly, when the lane curved. I of course was responsible and only tested this when no cars were around me. I lowered the autopilot speed to 50mph and the car was able to clear the turns much better. It would be nice for autopilot to detect the steepness of the turns and adjust speed accordingly.

rxlawdude | 27 October, 2016

@dpena,

I've seen TACC slow down for curves in the OC.

Pungoteague_Dave | 27 October, 2016

@stiction 'Why are people so bent on taking hands off the wheel?'

Because Elon promised exit-to-exit hands free driving for the current cars. The new hands-on nanny regime does away with hands-free-entirely, which is not what we were sold. There's plenty more missing, but that' s a biggie - I fully expected it and said so to much derision here, that some dangerous AP functionality would be removed, as it has, but some remaining elements can never be fixed, like the promised lane change after it is verified safe, or stop sign/traffic signal recognition.