Navigate on Autopilot is Useless (2018.42.3)

[mtndrew1]

Well at least it’s not just my car.

I think this is a byproduct of NoA “abilities” in that the new AP is always ready for NoA even though I have switched it off. I have also noticed that my car drifts back from the lead car and slows whenever there’s a potential freeway transition or lane change opportunity even though I’m not using NoA.

For example in the carpool lane when I get to the dashed section that allows lane changes my car will slow from ~70 to ~65 and create a big space as though it wants to be ready to change lanes even though my destination is far ahead and there’s no slowing of the target car ahead.

This version is a mess. I get that they have to start somewhere but I wish that when NoA was inactive it would just revert to the prior programming.

I rescind any praise I’ve given to 2018.42.x.

 

[electronblue]

The bounding boxes are jumping around on the IC, because they are poorly rendering them, not because they are jumping around on the ape. You can see clearly on @verygreen dev ape videos that the bounding boxes don't jump around compared to the crap they show us on the IC. This is just Teslas poor rendering of the data provided.

I disagree. You can see them jumping around in the videos especially towards the sides, back and in the fisheye periphery where they also tend to have most issues on the instrument cluster.

 

[electronblue]

I suspect V8 using only one (or 2?) forward camera(s) for this task was much less likely to pick up on nearby cars in adjacent lanes. The downside with V8 of course is that sometimes those cars really are coming into your lane, and V8 was bad at noticing that (despite the bogus claims about "side collision avoidance"). V9 tries to notice but because the bounding boxes are jumping all over the place it gets false positives.

Yes and no. Side collision avoidance is not bogus it just has had nothing to do with Tesla Vision as it has been based on ultrasonics.

 

[verygreen]

The bounding boxes are jumping around on the IC, because they are poorly rendering them, not because they are jumping around on the ape. You can see clearly on @verygreen dev ape videos that the bounding boxes don't jump around compared to the crap they show us on the IC. This is just Teslas poor rendering of the data provided.

It should be noted that bounding boxes are drastically different from IC display stuff.

Bounding box is around an object on the (2d) picture. On the other hand IC is a 3d rendition of stuff around the car. Even if your highlight the object on the picture, but are wrong about object distance/speed, the 3d rendition would jump around as you correct those parameters but the box would still be steady on the picture.

 

[rnortman]

The bounding boxes are jumping around on the IC, because they are poorly rendering them, not because they are jumping around on the ape. You can see clearly on @verygreen dev ape videos that the bounding boxes don't jump around compared to the crap they show us on the IC. This is just Teslas poor rendering of the data provided.

Just because they don't jump as much in a 2D representation doesn't mean they're not jumping a lot in 3D. They need to estimate the distance to the object in order to project into 3D. And it is presumably the 3D projection they're using to determine whether the vehicle is in your lane or not.

 

[rnortman]

Yes and no. Side collision avoidance is not bogus it just has had nothing to do with Tesla Vision as it has been based on ultrasonics.

Anything based on ultrasonics is rather useless, especially at highway speeds. Ultrasonics are very slow even when stationary; too slow to help you much in a highly dynamic environment. And at highway speeds they become very unreliable.

Still waiting for a single video showing side collision avoidance at work... you'd think maybe if it actually worked Tesla would have demo videos? Actually, come to think of it, even if it didn't work they would not hold back on the demo videos... just like FSD.

 

[electronblue]

Anything based on ultrasonics is rather useless, especially at highway speeds. Ultrasonics are very slow even when stationary; too slow to help you much in a highly dynamic environment. And at highway speeds they become very unreliable.

Still waiting for a single video showing side collision avoidance at work... you'd think maybe if it actually worked Tesla would have demo videos? Actually, come to think of it, even if it didn't work they would not hold back on the demo videos... just like FSD.

I don’t disagree. I’m just noting that Tesla’s side collision avoidance has been separate feature from vision thusfar at least because it has been solely ultrasonics (or sonar in Tesla hype language).

 

[rcarpen22]

I've had similar results with NoA. It doesn't seem to add anything to EAP except lane change suggestions of dubious quality. Only once has my car actually taken an exit without confirmation while running NoA. Most of the time, it just freaks out and cancels itself. The one area where I think AP2 exceeds AP1 is maintaining the lines of the road in rain or fog. My Model 3 has been able to keep driving in AP in situations where my S would not have been able to see the lines. So, that's nice. I am very much bummed that Tesla has stopped trying to improve AP1 completely. There are certainly a lot of things Elon promised when AP was first announced that AP1 has not (and will not) achieve.

Also, having read some of the posts on this thread, I had to laugh a little. Elon is aware of Jason's talent and has publicly tweeted about it, so I have no doubts he can improve NoA if he wants to. In fact, I'd be really surprised if Tesla hadn't already tried to hire him.

 

[rnortman]

I am very much bummed that Tesla has stopped trying to improve AP1 completely. There are certainly a lot of things Elon promised when AP was first announced that AP1 has not (and will not) achieve.

And the only rational thing to conclude based on past evidence is that there are will also be many of Elon's AP2 promises which are never fulfilled... and yet many on this forum continue to quote AP2 promises as if they mean something.

 

[electronblue]

Many forget that Tesla even announced a version of Navigate on Autopilot for Autopilot 1 — navigation based exit taking.

Not just Elon promised, Tesla announced this very publicly with a timeline. Then it was buried.

 

[rcarpen22]

I remember when AP1 was supposed to eventually read stop signs/red lights, was supposed to be able to use summon to pick you up at your door, and was supposed to be "hands free" on-ramp to off-ramp. Le sigh... I'm sure those presentations and screen shots of the website are still out there somewhere, not that it matters much. I was so pumped about having the car drive itself out of the garage and come pick me up. I mean, there aren't any other cars (that I'm aware of) that you can remotely back out of a garage or parking spot, but Summon was supposed to be so much more than it is. Also, in my opinion, and I know I will draw dislikes for this, the very first version of AP1 was the best AP ever. Excellent functionality, no nags, reasonable following distances... I miss that. I still can't imagine driving on the highway without AP, but the nags take a measurable percentage of my enjoyment away.

 

[Mark Z]

One scary lane in SoCal is the 91 eastbound HOV starting at the 110. In LA County, the concrete divider bulges out frequently near the vehicle as you pass by and both left side parking sensor dash indicators turn red. Manual override was accomplished at one location when it seemed too close. I would feel more comfortable if the AP2 Tesla would turn away slightly from these poorly designed barrier bulges as each one is approached.

 

[scottf200]

Nice demonstration in the Charolette/Moorseville NC area (south of Hickory)

This starts about 30 secs in and then thru 5:30 is the best stuff. I often watch these on my PC at x1.25-x1.50 using youtube gear icon to change.

 

[Kant.Ing]

So for people living their entire lives on highways like in some well developed Asian countries, NoA is already FSD.

 

[rnortman]

So for people living their entire lives on highways like in some well developed Asian countries, NoA is already FSD.

I can't tell if you're being serious. If you are being serious, I honestly don't know where to even start.

 

[diplomat33]

Last weekend, I finally got Nav on AP and got a limited opportunity to test it. Drove on Nav on AP for about 20 miles and took 3 exits. My first impression of Nav on AP is mixed. Auto lane change is very smooth. Auto steer in a lane is very solid. Recommending lane changes for passing slow traffic is not really that useful. I can certainly decide on my own if I need to pass slow traffic or not. I get the impression it is an intermediary step towards the real goal which is auto lane changes without driver input. Lane changes to get into the right lane for an highway exit could be useful in some instances if you are driving on unfamiliar roads and need help knowing what lane to be in. Now instead of paying attention to overpass signs and figuring out what lane to be in to catch a highway change, the car gives you a bit more info to help you. Where Nav on AP was not good for me was on taking exits. It was rather abrupt in taking short exits. It was a bit scary as the car did not slow down but just suddenly veered into the exit lane. Long exits were a little better but still a bit abrupt for my taste. Right now, I think I will take exits myself until that part of Nav on AP improves some more. The rest of Nav on AP is fine.

I do look forward to Nav on AP getting better over time just as EAP as improved over time. When taking exits gets really solid and smooth and auto lane change does not require driver input anymore, Nav on AP will become a really nice semi self-driving feature to have. It's not there yet but I am hopeful it will get there with future updates.

 

[DarkMatter]

If I could make one initial tweak, I would not have the car make the last lane change in to the exit lane until after the previous exit is passed. There are a few exits around here that back up often, so it's silly to try to cut in to that line of cars when you are actually getting off at the following exit instead.

 

[diplomat33]

I would like to see Nav on AP turn the blinker on sooner before taking an exit and also slow down a bit before getting to an exit. We know Nav on AP knows when the car will reach an intended exit so it should not be too hard to set Nav on AP to automatically turn the blinker on say 1/3 mi before getting to an exit.

 

[scottf200]

I would like to see Nav on AP turn the blinker on sooner before taking an exit and also slow down a bit before getting to an exit. We know Nav on AP knows when the car will reach an intended exit so it should not be too hard to set Nav on AP to automatically turn the blinker on say 1/3 mi before getting to an exit.

Certainly seems like a design choice they made ... I don't know about slowing down as that impedes traffic. It absolutely seems like when there is a long exit lane that the turn the blinker on and get into that lane ASAP (i.e. to get out of the way of normal traffic and to allow plenty of room to slow down (think rain, snowy roads, etc)).

 

[diplomat33]

Certainly seems like a design choice they made ... I don't know about slowing down as that impedes traffic. It absolutely seems like when there is a long exit lane that the turn the blinker on and get into that lane ASAP (i.e. to get out of the way of normal traffic and to allow plenty of room to slow down (think rain, snowy roads, etc)).

The reason I said it should start slowing down a little before the exit is because in my experience, Nav on AP takes some exits a bit too fast. Long exits are fine. The car has plenty of exit lane to slow down and it does. But when you are taking a shorter exit, there is less room to slow down. My car on Nav on AP took the following exit at the last second when I had almost missed the exit (moving right to left) at 65 mph and then ping ponged a lot in the exit and then slowed down drastically in the bend. It pulled it off but it was definitely not smooth.



In terms of using the blinker, I just think giving other cars a bit more advanced noticed that you are taking the exit is prudent. It's what a safe human driver would do. My first experience with a long exit, my car just swerved over into the exit lane only putting on the blinker when it initiated the lane change. My second experience with a long exit, I slowed down before the exit (using the scroll wheel) since I thought it might help the car handle the exit better (to avoid the experience with the short exit where the car just swerved into the exit at 65 mph), when I was at the exit, the car asked for a confirmation for a lane change, I hit the stalk to accept and the car put the blinker on and successfully made the lane change into the exit lane. That part was fine. But I wonder about the driver who was behind me and also taking the same exit since they had no indication that I was taking the exit until my car suddenly put the blinker on and swerved into the exit lane. It's certainly not how I would have handed the exit.