Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Firmware 8.1 - Autopilot HW2

This site may earn commission on affiliate links.
I've noticed that whatever database Waze uses for speed limits is way more accurate than the one Tesla uses. The speed limit in my Tesla is wrong a significant amount of the time, whereas Waze is almost always spot on.
 
  • Like
Reactions: MTOman
This is very interesting. I just drove 250 miles in the X with 17.11.3 (clear day with mild traffic) mostly highway and it really is much improved. Not quite AP1 yet, but getting close. Would love to hear if 17.11.10 really is something other then FWD bug fix (which I also have since 8.1).
I confirmed 17.11.10 fixed the FWD buttons missing bug. No release notes specific to release. No other changes observed
 
I've noticed that whatever database Waze uses for speed limits is way more accurate than the one Tesla uses. The speed limit in my Tesla is wrong a significant amount of the time, whereas Waze is almost always spot on.
Not sure where Tesla is getting the data. 60mph in my 30mph neighborhood. I know Waze is accurate in my area, every time I find it incorrect, I login and fix it.
 
I observed that autosteer has improved on streets without lanes. The lines do not jump as much. But, not nearly reliable enough to use consistently. This is one feature that AP 1 did not have. So, at least Tesla is working towards enabling that feature which will be required for autonomous driving.
 
  • Informative
Reactions: EinSV
So you think there is zero concern about Tesla now using two cameras and still not achieving parity with HW1, 3 months after they said they'd achieve parity with HW1?
No, I think the other companies working on autonomous driving systems should be concerned that Tesla Vision is working this well, this soon.
 
So you think there is zero concern about Tesla now using two cameras and still not achieving parity with HW1, 3 months after they said they'd achieve parity with HW1?

Not particularly. I'm a SW engineer (I admit very little knowledge about autonomous driving systems and limited machine learning) and if you have the hardware available to get a quick win I'm fully supportive of doing it.

MobileEye (AP1) seems to have spent a lot of time building a really polished piece of tech around the AP1 camera system. Tesla is way behind (time) and is rushed to reach parity with their new system. Rather than spend time reaching parity with a single camera, I don't find it concerning to use anything you have available. Is it as good as AP1 with a single camera? Nope. Can it be? Probably. Does it have to be before using other cameras? I don't see the point.

So, I'm sure if Tesla had the time they have the engineering talent to reach parity with the one equivalent camera. But they don't have time. They'll have a lot more time to improve behavior under less ideal (less cameras) conditions as time goes on, but right now they need to get AP1 parity out the door safely (key last word!).
 
Right. Tesla is using a second camera and still not at parity with HW1, 3 months after they said they'd be at parity with HW1, but don't actually need the second camera at present. Got it.

You can make whatever inferences you want from the available information. I'll make my
Not particularly. I'm a SW engineer (I admit very little knowledge about autonomous driving systems and limited machine learning) and if you have the hardware available to get a quick win I'm fully supportive of doing it.

MobileEye (AP1) seems to have spent a lot of time building a really polished piece of tech around the AP1 camera system. Tesla is way behind (time) and is rushed to reach parity with their new system. Rather than spend time reaching parity with a single camera, I don't find it concerning to use anything you have available. Is it as good as AP1 with a single camera? Nope. Can it be? Probably. Does it have to be before using other cameras? I don't see the point.

So, I'm sure if Tesla had the time they have the engineering talent to reach parity with the one equivalent camera. But they don't have time. They'll have a lot more time to improve behavior under less ideal (less cameras) conditions as time goes on, but right now they need to get AP1 parity out the door safely (key last word!).

I'm a SW/FW engineer as well, and the only thing I would add is that often times, businesses like to see the end game and structure all the milestones on the way to the end game. While MobileEye had 1 fish-eyed camera lens for basically every function, I'd imagine Tesla's end-game with their heterogenous camera array was different -- the narrow camera will likely only serve a role to track a lead car, and the main camera tracks slight lane curvature, and the wide camera tracks cars going perpendicular and or cutting you off.

It's very likely that they are trying to make the Enhanced Autopilot feature set a stepping stone to FSD, instead of just focusing on copying MobileEye's design. While it's more frustrating in the short run for customers demanding parity with AP1, in the long run it will result in less throwaway work on the path for FSD features.
 
I'm a SW/FW engineer as well, and the only thing I would add is that often times, businesses like to see the end game and structure all the milestones on the way to the end game. While MobileEye had 1 fish-eyed camera lens for basically every function, I'd imagine Tesla's end-game with their heterogenous camera array was different -- the narrow camera will likely only serve a role to track a lead car, and the main camera tracks slight lane curvature, and the wide camera tracks cars going perpendicular and or cutting you off.

It's very likely that they are trying to make the Enhanced Autopilot feature set a stepping stone to FSD, instead of just focusing on copying MobileEye's design. While it's more frustrating in the short run for customers demanding parity with AP1, in the long run it will result in less throwaway work on the path for FSD features.

Yeah that makes a lot of sense. Its something I considered but I actually don't know the specifications of the original AP1 camera. I imagine one of them is identical (or very close) due to the leaked plans that Tesla originally intended to use the AP1 software with AP2 until their new software reached parity...

I think we're starting to see evidence of what you're claiming though. For example, my AP2 car is doing a pretty good job of making lanes on completely unmarked roads. Not nearly good enough for Autosteer [yet] but I don't recall AP1 being able to do that at all... (not sure though since I only experienced AP1 as a loaner). Why would they be doing that if they only intended to reach AP1 parity? So as you said: I think they are designing for EAP/FSD in mind, which is great.

I really wish I could see under the covers! ;)
 
It's very likely that they are trying to make the Enhanced Autopilot feature set a stepping stone to FSD, instead of just focusing on copying MobileEye's design. While it's more frustrating in the short run for customers demanding parity with AP1, in the long run it will result in less throwaway work on the path for FSD features.

I mean, obviously that's what Tesla wants to do. But that doesn't affect anything.
 
I've noted those issues as well. Ping pong once debited from lane lines and headed to concrete divider. When I took over car fishtailed from my overcorrection. Tough to really instantly steer correctly when your headed for disaster. I liked to 55mph version's confidence and the 80mph speed. I think next update will refine these rough corners but for now it's not acceptable on turns.
Yep same here. Kind of bounces around a little in the lane on the freeway. Overall very usable on the freeway. On curves: when the road is empty highway the car is actually able to make curves on highways i found quite well.....a bit jerky and it takes it kinda late (wide turn) But in real life with cars close by in narrow lanes still nerve-racking esp la freeway lanes. terrible on surface streets
 
Is there any way to force the update to download? I've not received it yet. I'm heading out on a road trip in a few days and would really like to use the new upgraded autosteer.
Yup. Make sure your're connected to WiFi, then hold down both scroll buttons on the steering wheel until the main screen goes black. Once it reboots you should receive an update notification within 5 minutes ;)
 
I observed that autosteer has improved on streets without lanes. The lines do not jump as much. But, not nearly reliable enough to use consistently. This is one feature that AP 1 did not have.

my AP2 car is doing a pretty good job of making lanes on completely unmarked roads. Not nearly good enough for Autosteer [yet] but I don't recall AP1 being able to do that at all...

uh oh. It's looking like AP2 is exceeding AP1 on some dimensions now. Then what will people complain about?!
 
  • Funny
Reactions: MTOman
It's been 2 years and 3 months since delivery of my P85D. Still waiting for AP1 to drop me off at the door then park itself on private property.

I am still waiting for it to go pick up my order of burrito and coffee and get back to me. I am disappointed that it still can't do that. I feel cheated. I demand a refund. Would lemon laws apply here..?
 
I think the mistake many are making is to assume AP2 software is some kind of v1.0 of FSD. This is not the case, AP2 is a separate code base that had to be rushed into production due to the relationship between Mobileye and Telsa soured after the AP1 death where the Mobileye solution did not see a giant big rig passing directly in front of it.

AP2 is not v1.0 of FSD, they are completely different solutions. I am sure there is some shared code, but they are different. An example of this is the video that Tesla put out showing a short autonomous trip and showed all the cameras working to identify objects including signs, people and lights. AP1 nor AP2 do those things as a matter of fact, the only way AP1/2 sees people today is with the sonar.

Dont assume that just because AP2 has not surpassed AP1 that there is no development and testing on FSD.