Navigate on Autopilot is Useless (2018.42.3)

[avoigt]

Interesting so see how different the views on NoAP are.

I do not want to judge on what has been said here and call it right or wrong but only collect information as of now.

One point I want to make though is that in this thread a couple of statements has been made what V9 does not do although Tesla confirmed that its not yet doing it so I wonder why this is something that should even be discussed.

Example:

• Car does not suggest too get back to the right lane(s) - Tesla did confirm comes later
• Car does not work well in construction zones - It was never said AP2.5 NoAP can do that yet
• Car does not read speed signs - Well, again it was never said it can do this
• I have seen some other statement where people got pretty emotional and believe it does not add value

Also I believe all the experiences shared here are fair and accurate but what is very interesting to me is that it seems the personal experiences are very much different and also how people recognize improvements as well as step backs on the current version.

Finally there has been many statements what the car did in a specific situation. Some said they experienced the exact opposite, others said they experienced the same. So thats kind of odd I have to say or better said it does not create a consistent result.

Example, car does suggest a takeover but would collide with another car.... all Videos I did review did not show that and usually bad situations are posted immediately.

Its more than fine to criticize but its helpful to have an illustration or video otherwise its not very valuable for most readers. This is because the specific situation when it did not perform well needs to be looked at and most of the conversation here did miss that level of detail unfortunately.

To illustrate I want to share the below video from someone obviously very positive about V9 NoAP.

 

[Boomer19]

@wk057
excellent information across your posts here, and also good +/- feedback from most others.

i’m a noob (or a boob) when it comes to the tech of AP, and don’t even have the latest version yet (kinda miffed about it as well).

my question, to everyone really, is why?
why is tesla not up to speed on this? is it internal strife about a direction to take? disagreement on methodology? budget? head butting with old guard versus avant- guard?
is it that karpathy hasn’t been there long enough, to cultivate a team, set a game plan, get everyone on board, and start executing?
are they on the cusp?

any thoughts on this, i think, would be as interesting as the tech itself. FSD seems such a monumental challenge, yet people speak as if it’s so apparently going to be as plain as day...i’m just trying to grasp what forces have to align in order to find success on this front.

thanks.

anyone have point of view on this?

 

[lunitiks]

anyone have point of view on this?

 

[AWDtsla]

I actually did this mod on my S a while back when I had a broken radar bracket that would eventually slide out of usefulness. The problem with it is that you get used to TACC and it limits your reaction time, even if you've been told by the car you're in dumb mode now. So, honestly, this is something that shouldn't be done. It actually is a problem resolved on the facelift cars, X, and 3... since the radar is protected.

I'll sign a waiver. If I can successfully drive an older car with dumb cruise without crashing, pretty sure I can handle the same in the Tesla. The use-case is driving in inclement weather where I'm already hyper-vigilant about what's ahead of me, the problem is accidentally going faster than desired. For example driving downhill a mountain highway, staying at 50 mph, no faster, no slower, minimum inputs at all times. To prevent mode confusion, I think maybe radar cruise should never work unless autosteer is engaged, they should be the same mode.

I think the RLS on AP1 works great, and they should put this back into production for AP2 cars. More options are better than less, though, in most cases.

If behavior on AP1 is considered good, can't imagine AP2. Very inconsistent here. Goes into fast wipe stopped at a light, negligible mist. Refuses to wipe, lots of water, bad visibility. Both on the same drive, same lighting/weather. Never once complained about RLS with BMW.

 

[DarkMatter]

I'll sign a waiver. If I can successfully drive an older car with dumb cruise without crashing, pretty sure I can handle the same in the Tesla. The use-case is driving in inclement weather where I'm already hyper-vigilant about what's ahead of me, the problem is accidentally going faster than desired. For example driving downhill a mountain highway, staying at 50 mph, no faster, no slower, minimum inputs at all times. To prevent mode confusion, I think maybe radar cruise should never work unless autosteer is engaged, they should be the same mode.

Manufacturers and safety experts do not recommend using any cruise control in inclement weather, such as heavy rain, ice, or snow. You are an adult and can do whatever you want, just please know that there are accidents regularly caused by using cruise in poor weather.

 

[jschwefel]

Similar to Boomer19’s question above, what are the reasons behind AP2 deficiencies and why hasn’t Tesla able to address them? I understand the breakup with Mobileye, but that was two years ago. Why hasn’t Tesla progressed any further? Do they have the wrong approach? Are they incompetent? Are they being conservative?

What is the consensus on what’s holding them back?

A couple of the contributors to this tread seemingly have the solution, or at least some good ideas. Will these ideas not scale to what Tesla needs? Most of us want Tesla to be successful. How can this community help get us where we want to be?

 

[Bladerskb]

The result is that the car is usually able to seemlessly overtake a slower vehicle without ever decelerating.

The software will also recursively overtake another vehicle that appears ahead while already in an overtake maneuver, and eventually work to return to the original lane unless a number of factors make it clear that it shouldn't (like that target lane being exit-only or something). Lots of sanity checks. For example, a lane must exist for a full mile before being considered a target lane. It also checks nav data about the current road to double check if a lane is a ramp or not, when possible... although this data is pretty bad.

There is a lot of logic coded in... it makes the "thoughts" of the custom setup a bit more defined, mainly due to the predictive nature of the model used. It will continue to project vehicle positions for every vehicle ever encountered, until a max number of objects is encountered. It's kind of cool, although limited since it assumes a constant speed for the other vehicles it can no longer see. I also wrote an algo that predicts longitudinal movement of vehicles seen by the ultrasonics, which augments that data a bit and works okay. The result is that, in practice, rarely does it begin the process of an overtake/lane change while there is a vehicle in the way in the target lane. The only exception tends to be when moving back to the right near a merge... which I don't even think AP2 can handle as of yet, either.

I did try to program in courtesy lane changes for interchanges/merges, but just not enough data to work with to make it reliable
.

Are you sending control commands or are you just calling a changeLaneRight() method?

It would be interesting if you can control the steering and throttle and brake.

I'm more interesting in the throttle/brake since that's safer.

I'm currently working on open source autonomous driving layer that you can feed perception data and it will be able to perform advanced lane changes, planning, merging, cut-ins. based on a model based reinforcement learning using mobileye's rss.

So even if you are just calling a changLaneLeft() and a setSpeed() function. I could probably still get it to work sufficiently.

With this system you would be able do a cross country drive (highway only) without ever touching any controls.

I still have some kinks I'm working out. But will be interested in working with u on ap2.

Also can you access and grab frames from model 3 driver facing cam?

 

[Wooloomooloo]

• NavOnAP has no concept of "Keep Right, Pass Left". It never suggests lane changes back to the right in any of the available modes..

Neither do about 97% of American drivers... sorry, I had to comment on this one.

 

[AWDtsla]

Manufacturers and safety experts do not recommend using any cruise control in inclement weather, such as heavy rain, ice, or snow. You are an adult and can do whatever you want, just please know that there are accidents regularly caused by using cruise in poor weather.

Accidents are caused by control inputs that exceed the traction envelope. Using cruise to maintain speed downhill is the minimum possible input. Given that you don't have any audible feedback of speed in a Tesla, it's quite easy to exceed your desired speed while you're concentrating on looking at road, leading to a non-minimal input event to slow down. I've literally driven through hundreds of snowstorms without the aid of even stability control. I have no problems either putting the car into a drift or recovering from it, in any conditions.

 

[mtndrew1]

Similar to Boomer19’s question above, what are the reasons behind AP2 deficiencies and why hasn’t Tesla able to address them? I understand the breakup with Mobileye, but that was two years ago. Why hasn’t Tesla progressed any further? Do they have the wrong approach? Are they incompetent? Are they being conservative?

What is the consensus on what’s holding them back?

A couple of the contributors to this tread seemingly have the solution, or at least some good ideas. Will these ideas not scale to what Tesla needs? Most of us want Tesla to be successful. How can this community help get us where we want to be?

I’ve been using AP2 since January of 2017, first using a friend’s car and now on my own (HW2.5).

The progress since just January of this year has been remarkable. Ten months ago AP2 could not crest a hill on the 101 freeway in Los Angeles without totally losing its composure and darting across lanes. It would bang into curbs auto parking. The steering inputs were hundreds of little twitches. It would accelerate straight into stopped traffic. Etc.

In the first release in January of ‘17 the system couldn’t render lane lines without them dancing wildly in the instrument cluster. The maximum speed the system would allow was 25 MPH for a while. Then 30 for a month or two. Then 35. And so on.

Because NoA is rubbish at the moment doesn’t mean that AP2+ hasn’t made enormous strides.

Since I bought my car it has mastered lane changes above 30 MPH and has added the (currently clunky) ability to do so below 30. It can show cars in other lanes. It has smooth, natural steering inputs now.

If you haven’t been along for the ride since 12/2016 it’s easy to lose sight of how much Tesla’s system has grown (IMO). From my viewpoint the new skills are accelerating and I’m looking forward to what it brings in 2019. NoA is indeed crap (for me) at the moment though.

 

[jschwefel]

I’ve been using AP2 since January of 2017, first using a friend’s car and now on my own (HW2.5).

The progress since just January of this year has been remarkable. Ten months ago AP2 could not crest a hill on the 101 freeway in Los Angeles without totally losing its composure and darting across lanes. It would bang into curbs auto parking. The steering inputs were hundreds of little twitches. It would accelerate straight into stopped traffic. Etc.

In the first release in January of ‘17 the system couldn’t render lane lines without them dancing wildly in the instrument cluster. The maximum speed the system would allow was 25 MPH for a while. Then 30 for a month or two. Then 35. And so on.

Because NoA is rubbish at the moment doesn’t mean that AP2+ hasn’t made enormous strides.

Since I bought my car it has mastered lane changes above 30 MPH and has added the (currently clunky) ability to do so below 30. It can show cars in other lanes. It has smooth, natural steering inputs now.

If you haven’t been along for the ride since 12/2016 it’s easy to lose sight of how much Tesla’s system has grown (IMO). From my viewpoint the new skills are accelerating and I’m looking forward to what it brings in 2019. NoA is indeed crap (for me) at the moment though.

I’ve had my Model 3 with EAP since February, and I’ve noticed a number of improvements in the software since then. Frankly, I honk it’s pretty good and I’m not aware of another system out there or being offered today that is better. I want it to stay that way; however the comments from some very smart people in this thread give me concern of some fundamental problems within Tesla’s EAP/FSD program. Are they on the right path or will they soon fall behind? I guess only time will tell.

 

[jschwefel]

I’ve had my Model 3 with EAP since February, and I’ve noticed a number of improvements in the software since then. Frankly, I honk it’s pretty good and I’m not aware of another system out there or being offered today that is better. I want it to stay that way; however the comments from some very smart people in this thread give me concern of some fundamental problems within Tesla’s EAP/FSD program. Are they on the right path or will they soon fall behind? I guess only time will tell.

honk = think, darn spell check

 

[jeffro01]

If only they'd get rid of the stupid autosteer speed restrictions... That by itself would be a massive improvement... Their database for this is inaccurate and incomplete so it seems like a questionable decision to rely upon it...

Jeff

 

[AWDtsla]

If only they'd get rid of the stupid autosteer speed restrictions... That by itself would be a massive improvement... Their database for this is inaccurate and incomplete so it seems like a questionable decision to rely upon it...

Jeff

LOL I almost forgot about this. It's fun driving through a tunnel at 65-70mph when the car reads the 20mph speed limit of the surface street above you.

Super fun.

 

[Santa Claws]

@wk057 - Solid original post!

I'd say I logged 150-200 miles this weekend with NavOnAP enbaled. I saw a lot of what you mentioned. It has me way more involved than regular Autopilot ever did. I get it... It's brand new. There are a lot of features being introduced with NavOnAP, and it needs to be properly sorted. I just hope Tesla can make something actually useful for us in the near future.

At a minimum, it would be great if Tesla got us the following implemented ASAP:

1. Read speed limit signs
2. Overtake a vehicle with zero deceleration (or even just very limited deceleration)

 

[S4WRXTTCS]

I’ve had my Model 3 with EAP since February, and I’ve noticed a number of improvements in the software since then. Frankly, I honk it’s pretty good and I’m not aware of another system out there or being offered today that is better. I want it to stay that way; however the comments from some very smart people in this thread give me concern of some fundamental problems within Tesla’s EAP/FSD program. Are they on the right path or will they soon fall behind? I guess only time will tell.

For EAP I think it's best to think of it as lot of components need to be in place for it to work right.

I think what we're seeing is the result of some elements working well, but other elements not being where they need to be.

In my use of it my biggest annoyance was the lack of a maps driven approach. Like it sometimes an exit on the right would trick it into moving slightly into it. Why? The car knows where I'm going, and I'm on drive-on-nav so why would it even flinch for that?

There were also numerous cases where it simply didn't know the correct speed. This meant I kept having to check my rear to make sure I wasn't annoying anyone by letting Drive-On-Nav do the driving.

So most of the issues have nothing to do with NN's, but it seems like NN's were the major focus of V9.

For me personally I'm going to reserve judgement for 6 months on it.

Where I recognize it as something that brought a lot of things under the hood, and it takes time to steadily improve it.

 

[S4WRXTTCS]

LOL I almost forgot about this. It's fun driving through a tunnel at 65-70mph when the car reads the 20mph speed limit of the surface street above you.

Super fun.

One would think Tesla would have a solid solution in place for speed limits, and recommended speeds for each part of a white-listed drive on nav road BEFORE it was allowed to initiate drive on nav for it.

 

[S4WRXTTCS]

@wk057 - Solid original post!

I'd say I logged 150-200 miles this weekend with NavOnAP enbaled. I saw a lot of what you mentioned. It has me way more involved than regular Autopilot ever did. I get it... It's brand new. There are a lot of features being introduced with NavOnAP, and it needs to be properly sorted. I just hope Tesla can make something actually useful for us in the near future.

At a minimum, it would be great if Tesla got us the following implemented ASAP:

1. Read speed limit signs
2. Overtake a vehicle with zero deceleration (or even just very limited deceleration)

3. Better speed control for changing lanes into a lane behind someone slower. Sure it's okay to slow down a LITTLE bit before changing the lane, but don't slow all the way before getting over. Just move over, and then slow down.

 

[eli_]

I wonder how many of the issues with NoA is from errors in the "ADAS map tiles". Like mine will inexplicably put a turn signal on as if it's taking a ramp or fork when there is nothing actually there. And it doesn't correlate with anything in the turn-by-turn nav directions.

 

[MagicXCarpet]

Navigate on Autopilot is Useless (2018.42.3 Sounds like this was written by one of the desperate bears trying to depress TSLA.