Blog See the “Smooth as Silk” Autopilot Update in Action

[vc_row][vc_column][vc_video link=”https://vimeo.com/221199071″ video_title=”1″][vc_column_text]The “smooth as silk” Autopilot algorithm promised by Elon Musk is beginning to roll out to Tesla’s equipped with Autopilot 2.0.

Model S owner James Hansbert published a video of the new firmware (17.22.46) in action and offered some first impressions, including:



Some “dancing” between lines, but better at tracking lanes through intersections where lane are missing lines

No problems in rain

Smooth braking at traffic lights

Difficulty getting Autopark parallel parking to detect open spots



Hansbert also commented on two new features highlighted in the release notes:

Autopark: Perpendicular Parking

To make it easier to park in a variety of situations, Model S can now also back up into perpendicular spaces using Autopark. Drive completely past the space at no more than 10 mph until the ‘P’ appears in the instrument panel. Then use the same Autopark functions as in parallel Autopark – starting canceling, pausing, resuming, and so on.

Display Brightness

With Display Brightness, you can clearly see the touchscreen and instrument panel throughout the day and night. This feature adjusts screen brightness based on your Model S surroundings. It also learns from your preferences: if you adjust the brightness manually, Display Brightness uses your update to make future adjustments.

Display Brightness is enabled by default. To disable, uncheck Auto-adjust in Controls > Displays > BRIGHTNESS & MODES.

A wide release of the firmware is expected this week.[/vc_column_text][/vc_column][vc_column][/vc_column][vc_column][/vc_column][/vc_row]

 

[pilotSteve]

Thanks for the video, its a good set of tests and commentary. Sadly the silk seems to be badly wrinkled still/again. Ah well, for sure "next version" will be for the win. Or at least different!

 

[AmpedInAustin]

Was surprised to see it fail so much. Still a lot of work to be done. I look forward to more feedback from others that have it.

 

[Matias]

I have AP1 and I think it would have made the same mistakes.

 

[Mark Z]

Thanks for posting.

For those who need to see AP2 at its worse, start the playback at 11:33.

I was hoping for a major improvement in this area. Unfortunately, not in this release.

 

[apacheguy]

Hilarious! Thanks for sharing.

 

[BestRadar]

So much as smooth as silk.. After 11:30 in your video is what mine seems to behave like which is why I keep sayings its still dangerous.

 

[oktane]

Total fail. What a let down. Tesla probably will never achieve parity, at least not in the time I own the car.

 

[brianman]

I noticed today that at least one of the tweeks uses the adjustive "longitudinally" for the smoothness. Is that new? And is that basically a comment about TACC rather than Autosteer.

 

[Electroman]

From the video it seems that Autosteer and TACC behaves just as good as AP1.

It struggles when there is a curve *and* the lane markings are missing or confusing, as in two lanes being split it took the left lane, but it happened to be left-turn lane.

I am not sure what the incessant whining is all about.

 

[chillaban]

From the video it seems that Autosteer and TACC behaves just as good as AP1.

It struggles when there is a curve *and* the lane markings are missing or confusing, as in two lanes being split it took the left lane, but it happened to be left-turn lane.

I am not sure what the incessant whining is all about.

Totally agreed. This update was about smoothness of control, not miraculously new holistic path planning.

It seemed to behave much like how I expect AP1 would have, and the situations where it failed are pushing the current definition of Autopilot past its limits.

EDIT: I wish the test course had more stop-and-go traffic, more smooth continuous curves (particularly the sharper ones). Those situations are where AP2 is dodgy compared to AP1. It sometimes makes quite sudden speed adjustments in stop and go traffic, and takes most curves by tugging the steering wheel once a second, resulting in a jerky feel. (And there seems to be some sort of steering angle limit, because on sharper curves it will smoothly depart the lane and the screen even shows the car knows it's falling out of the lane, but it refuses to correct with more steering).

 

[Electroman]

I hate the first video that comes out, takes the most edge cases and tries to rally the perennial complainers.

I am sure there are so many improvements in highway conditions and high stop and go traffic, that is completely not addressed on that video.

 

[arnis]

Tailgater. Vuush.... I would slow down slightly for that

 

[whttiger25]

I am looking forward to trying it out. I just took delivery of my AP2 car (coming from AP1) and have noticed it is far worse in general multi-line divided freeway situations. I wouldn't dare try it on a 2-lane undivided road. So far, i've noticed the car tends to bounce around between the lane lines like a pong game, and on a mild left curve with clear markings it felt like the car rapidly started drifting to the right and without my intervention would have careened into another car...and while I would intervene 100% of the time I wonder what it would have done if I didn't....

Anyway, luckily I like driving the car so I can be patient, but it's still frustrating to see how much worse AP2 is than AP1...

 

[David29]

Was surprised to see it fail so much. Still a lot of work to be done. I look forward to more feedback from others that have it.

Actually, as an AP1 user, I thought it did very well, especially as the testing was done on undivided roads, which we were told from the outset were not the intended environment for the system (even though it sometimes does work pretty well there). I was particularly impressed with how well it carried straight through intersections at which there were breaks in the lane markings.
I'd like to also see tests on a divided, limited access highway, and I imagine those test videos will show up soon.

 

[ecarfan]

@David29 I also noted that the video showed AutoPilot being used on undivided roads with cross traffic. I realize that the driver appeared to be paying close attention to his surroundings, but I wouldn't use AP or EAP on that type of road. I don't understand why some people continue to do that.

 

[chillaban]

@David29 I also noted that the video showed AutoPilot being used on undivided roads with cross traffic. I realize that the driver appeared to be paying close attention to his surroundings, but I wouldn't use AP or EAP on that type of road. I don't understand why some people continue to do that.

I think for the purposes of constructing a difficult AP test course, it's fine. As you said, he seems to have had a hands-free recording setup and intervened appropriately. Certainly it's not where I'd use AP1 or AP2 on a day to day basis.

I really wished the test was done on an interstate / multilane continuous highway, as that's where AP2 lacks a lot of smoothness compared to AP1. Whether it's tracking curves, stop and go traffic, switching lanes, or dealing with cut-ins, AP2 currently gives a lot of rough and unnecessary steering and accelerating/braking commands. That's what I took Elon to mean when he teased silky smoothness. I didn't expect it to magically start understanding turning lanes and navigating city roads without a ton of extra fanfare and weeks of teasing

 

[brianman]

I am looking forward to trying it out. I just took delivery of my AP2 car (coming from AP1) and have noticed it is far worse in general multi-line divided freeway situations. I wouldn't dare try it on a 2-lane undivided road. So far, i've noticed the car tends to bounce around between the lane lines like a pong game, and on a mild left curve with clear markings it felt like the car rapidly started drifting to the right and without my intervention would have careened into another car...and while I would intervene 100% of the time I wonder what it would have done if I didn't....

Anyway, luckily I like driving the car so I can be patient, but it's still frustrating to see how much worse AP2 is than AP1...

Standard request: Please include firmware version number when providing experiential data on your vehicle with AP. This stuff moves fast.

Thanks!

 

[oktane]

@David29 I also noted that the video showed AutoPilot being used on undivided roads with cross traffic. I realize that the driver appeared to be paying close attention to his surroundings, but I wouldn't use AP or EAP on that type of road. I don't understand why some people continue to do that.

Also would prefer to see what it does on the freeway. As to why people use it on local/undivided roads, it's because that's how Tesla has marketed it. Don't you remember seeing the FSD video? Tesla's EAP description included local roads. Never did Tesla suggest use of EAP was restricted to freeways. If that were the case, they could easily make it so in their firmware and issue us all refunds.

 

[chillaban]

Also would prefer to see what it does on the freeway. As to why people use it on local/undivided roads, it's because that's how Tesla has marketed it. Don't you remember seeing the FSD video? Tesla's EAP description included local roads. Never did Tesla suggest use of EAP was restricted to freeways. If that were the case, they could easily make it so in their firmware and issue us all refunds.

I'd rather this not derail into a highway vs local AP debate. We've had this in many threads and I'm guessing someone is already preparing screenshots of various manual snippets and whatnot.

At any rate, we can probably agree that as-is, the way EAP/AP1 handles local surface roads is not great. This video is a good demonstration of that. But the subtext is that this video discredits the claim of "silky smooth" control algorithm improvements, but for me, it doesn't quite do that yet. If we see AP2 erratically hunting on a freeway or losing track of curves, then that's more of evidence that 17.22.46 is more hype than fact.

But for now, I eagerly await more videos and impressions of this update