Is HW3 required to fix the problems with AP / EAP / NoA / AutoHighBeam / AutoWipers?

  • SUPPORT THE SITE AND ENJOY A PREMIUM EXPERIENCE!
    Welcome to Tesla Owners Online, four years young! For a low subscription fee, you will receive access to an ad-free version of TOO. We now offer yearly memberships! You can subscribe via this direct link:
    https://teslaownersonline.com/account/upgrades

    SUBSCRIBE TO OUR YOUTUBE CHANNEL!
    Did you know we have a YouTube channel that's all about Tesla? Lots of Tesla information, fun, vlogs, product reviews, and a weekly Tesla Owners Online Podcast as well!

  • It's OK to discuss software issues here but please report bugs to Tesla directly at servicehelpna@teslamotors.com if you want things fixed.

John Di Cecco

Active member
Joined
Sep 25, 2017
Messages
27
Location
Crystal Lake
Country
Country
What I see leads me to think the phantom braking is a conservative overreaction because the video/sonar processing pipeline is too slow and too decimated (lower frame rate) in order to accommodate to the hardware. Then in order to work around the slowness there are events that cut thru the pipeline to conservatively slow the car. Often it’s reacting to someone cutting in that is far enough ahead for them to match speed by the time the car would intersect their position, but the car won’t realize that until the processing pipeline is done and it may miss knowledge due to low frame rates. New HW 3.0 can’t come soon enough!
You lost me with 'Sonar'
 

M3OC Rules

Top-Contributor
TOO Supporting Member
Joined
Nov 17, 2016
Messages
626
Location
Minneapolis, MN
Tesla Owner
Model 3
Country
Country
What I see leads me to think the phantom braking is a conservative overreaction because the video/sonar processing pipeline is too slow and too decimated (lower frame rate) in order to accommodate to the hardware. Then in order to work around the slowness there are events that cut thru the pipeline to conservatively slow the car. Often it’s reacting to someone cutting in that is far enough ahead for them to match speed by the time the car would intersect their position, but the car won’t realize that until the processing pipeline is done and it may miss knowledge due to low frame rates. New HW 3.0 can’t come soon enough!
Interesting theory. It is curious how it slows just a little bit sometimes and then resumes. But it maybe it's just not smart enough yet. I don't think it is programmed to anticipate what a car's speed will be or if it will be out of the way before you get there. It does try to figure out if a car is going to merge but once it decides a car is going to merge it doesn't re-evaluate that decision. You have to wait until the car goes away(which can be quite a ways sometimes) or it merges in.

It would be interesting to know if currently, HW3 is running the same code with the same processing time(meaning they are slowing it down so it performs the same as HW2.5.)
 

DocScott

Well-known member
TOO Supporting Member
Joined
Mar 6, 2019
Messages
345
Location
Westchester, NY
Tesla Owner
Model 3
Country
Country
My sense is that, even with HW 2.5, AP has gotten much more sophisticated recently (I'm on 20.4.4). It's anticipating other cars' behaviors in a much more human-like way than was the case a few versions ago. For example, if I signal to change lanes in to a lane with a slower flow of traffic, it gently slows to match the speed in that lane. It slows more gently than it used to when a slow-moving car crosses in front (i.e. crosses multiple lanes) and resumes speed once it's crossed more quickly. It slows a bit in advance when it sees a curve coming. It also slows a bit (but not too much) when uncertainties increase; e.g., when there's someone ahead driving erratically, or when visibility is limited, or when there's stopped traffic in an adjacent lane.

While the slowing for lane changes in to a slower lane is easy to replicated, the "slowing for uncertainty" behavior is hard to pinpoint, because it's a kind of a judgment call, and it usually involves some kind of corner case. For example, yesterday I was taking an interchange on AP (I don't have EAP or FSD). The two lanes on the right split off for the interchange, and I was in the left one of the two. Just after the lanes split, was one of those spots where you're not supposed to have a car, but it's still paved and you can fit one in if you try. And there was a vehicle (SUV, if I recall) sitting in that spot, with a blinker on (turn signal? hazard? it's a bit of a blur and I didn't hit the dashcam to record it). But the vehicle wasn't moving at that moment.

AP clearly did not like that situation, and I don't blame it. It slowed down about 20 mph, but did so fairly gradually. In fact, my wife, who is certainly familiar with me driving on AP, didn't think it was AP--she thought I was just unsure as to whether to take the interchange. That's different from "phantom braking" of a few versions ago, where it was sharp enough that she'd know quickly it was AP misbehaving! I did use the accelerator to maintain a higher speed than AP was choosing past the stopped car, because I judged it wasn't about to jump out in to traffic, but I think the car reacted reasonably and that if I'd let it AP would have calmly driven by the person who wasn't where they were supposed to be at maybe 35 mph, and then picked back up to highway speed.

So I think there's still software gains being made that affect HW 2.5.