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

Navigate on Autopilot is Useless (2018.42.3)

This site may earn commission on affiliate links.
"Navigate on Autopilot" has now replaced the advertised "On-ramp to Off-ramp" EAP feature on tesla.com/autopilot.

Happened this month.

tesla_autopilot update.JPG


Off-topic I guess, but "FSD" text also changed.

tesla_com_autopilot_FSD update.JPG
 
"Navigate on Autopilot" has now replaced the advertised "On-ramp to Off-ramp" EAP feature on tesla.com/autopilot.

Looks like they cleared up lots of things that never made any sense to begin with.

The 4 versus 8 cameras never made sense to me. All the cameras were really needed to see what was going on around the car.

The changing lanes unconfirmed lane changes never made sense for EAP. Sure it's what I was looking forwards to the most, but I had a hard time understanding how it could be done safely while also not being FSD. It seemed to be more of a FSD feature than an EAP feature, but Tesla had it as an EAP feature. This also explains why we haven't seen unconfirmed lane changes switched on despite having way more miles than needed to validate it.

The only thing I'm surprised about is the snake chargers (or their equivalent) is still on there. I'm surprised because I haven't heard anything about them since the demonstration of an early proof of concept. I'm sure handicap people would love to have at least one or two snake chargers at superchargers.
 
I think that might be a bit of rose colored glasses. The first AP1 loaner I had in 2016 was impressive, but also freaked me out with a phantom braking event on a metal bridge. The most recent AP1 loaner I had could not make it through a corner I have been using as an AP2 test corner without crossing the center line and hitting the brakes. I would trust AP2 in more situations than AP1, though in the situations that AP1 has down, it does tend toward more confidence. Caveat: I only have a few hundred miles on AP1, compared to thousands on AP2.

I'm not basing it just on my experience.

I was even more heavily involved on TMC back in 2015, and 2016 than I am now. I only saw a handful of mentions of false braking during the early days of TACC. It only increased after Tesla tried to do more with it. There was a particularly bad update where they tried to brake if an emergency vehicle was stopped in the lane next to you, but it kept mis-detecting things that weren't emergency vehicles at all. At some point they pulled the update because it was so awful. I can't remember if it impacted TACC or only AP.

In any case I had a really positive experience with TACC during my entire time with my Model S between 2015 and 2018.

On any given drive from Seattle to Portland I knew the odds of a false braking event with AP1 was low, and if it did happen it was going to be pretty soft. Sure it still happened occasionally, but a really low amount. Probably on-par with what a Subaru with a later version of eyesight has. I didn't have good experiences with AP so I didn't use it. Most of my drives was 90% TACC only.

With my Model 3 I know that any drive from Seattle to Portland is going to have false braking. I tend to know where it will occur and I could easily mark them if Tesla had any way of marking a problem spot using the nav map. I know any time I go under a sign where the road dips under it that it might trigger a false braking event. I believe there is one on 405 Northbound near where you get on it from I5N.

The difference with the Model 3 is that most of my driving is AP driving as AP2 is so much better at AP driving than AP1.

The insurance institute also tested AP1 (with an older version of firmware) against AP2 (with a new firmware version) and they concluded the same thing. That there were way more false braking events with the AP2 car than with the AP1 car. But, the AP2 car (with the later firmware) was a lot better at staying in the lane.

Keep in mind I'm not opposed to Tesla trying to make sure their cars stop for stopped object. Most of the events of a Tesla running into an emergency vehicle have happened with AP1.

The problem I have is it was imposed on me when I don't have a problem with crashing into stopped cars.

I should have a way of dictating to it how sensitive I want the TACC to be.

There are sensitivity setting for Emergency braking, and there is a binary enable switch for prevent acceleration into stopped objects (which I have off because that's had false detections). But, there is no way to adjust the TACC sensitivity to stopped cars.
 
@S4WRXTTCS The wording does not suggest to me that automatic lane changes have been retired entirely as an idea, ”makes adjustments” can mean many things. I can see your concerns over it for sure though — how would it work in practice...

The old language was a commitment, and the new language doesn't commit to anything that it doesn't already do.

Makes adjustments simply means to me that it's going to try to position itself between the cars to really encourage you to acknowledge the requested lane change.

I'm not opposed to how it is today if it had really good maps, and if they significantly improved the automatic lane changes to where it didn't abort the lane change 10% of the time for no reason at all.

When you're faced with a lot of interchanges NoA has the opportunity to be extremely useful even without lane changes to get around slower traffic.

Like the other day I was in Oregon and to get somewhere I had to go on 26W, 405, I5, and then 84 (or something like that). It was just a crazy route where traffic tended to be backed up in the lane I needed so I needed to get over early. NoA wasn't getting me over as soon as I needed. I wanted to test it, but didn't want to risk being stuck in the wrong lane because the traffic in the lane I needed was backed up.

The maps seemed better in Oregon than WA so it's a fun place to test NoA, and there are lot of interchanges along with bridges to test it out on.

I can't test it in WA because it doesn't have good maps for I5 going through Seattle (at least not going North) or 405S getting onto I5S.

On large stretches of 405 there isn't a whole lot for it to do because it's in the HOV lanes protected by a double lane barrier.

Tacoma is a mess where even Google/Apple have a hard time keeping up with all the changes.
 
Last edited:
UK and US texts are not identical.

Btw, US text says (emphasis mine) “All Tesla vehicles have the hardware needed in the future for full self-driving in almost all circumstances, at a safety level we believe will be at least twice as good as the average human driver. The system is designed to be able to conduct short and long distancetrips with no action required by the person in the driver’s seat.”
 
Great analysis. AP2 has been training me on when to take over since it arrived. It's more often than any of us would like. The annoying double chime, both on & off, bothers my passengers once they know what they mean (I wish I could at least turn down the volume). The nag is getting easier to handle. At least I haven't had my hand slapped recently "for the remainder of the trip" ;0-
 
  • Like
Reactions: scottf200
UK and US texts are not identical.

Btw, US text says (emphasis mine) “All Tesla vehicles have the hardware needed in the future for full self-driving in almost all circumstances, at a safety level we believe will be at least twice as good as the average human driver. The system is designed to be able to conduct short and long distancetrips with no action required by the person in the driver’s seat.”


Not that I’m interested in it, but what does that mean for the whole Tesla ride hailing service?!
 
  • Funny
Reactions: rnortman


Not that I’m interested in it, but what does that mean for the whole Tesla ride hailing service?!

Given that the details for Tesla Network have been coming ”next year” since 2016 and they can be expected to come ”next year” for the foreseeable future I would encourage you to wait until ”next year” for the answer.

In all seriousness it seems like Tesla is giving up on the car responsible driving part (summon without a driver) for the time being.

So much for the Tesla announced ”Level 5 capable” hardware in 2016 AP2 cars?
 
Given that the details for Tesla Network have been coming ”next year” since 2016 and they can be expected to come ”next year” for the foreseeable future I would encourage you to wait until ”next year” for the answer.

In all seriousness it seems like Tesla is giving up on the car responsible driving part (summon without a driver) for the time being.

So much for the Tesla announced ”Level 5 capable” hardware in 2016 AP2 cars?

The true believers will continue to believe that driverless ride-sharing is coming. We just need to wait for V8. I mean V9. I mean HW3. Yeah, that's it, HW3 has 100x more power! All we need is more power and suddenly SkyTeslaNet will save us all.
 
I know this thread is about 2018.42, but I finally used 2018.48 NoA a few days back. At least in light traffic, it's much better about suggesting lane changes to overtake. If you take the suggestions it does not end up slowing down before making the pass. It's still a bit too slow on lane changes to get out of the way of someone who comes up to undertake, or to make a quick pass, though. That seems OK at this stage in development as there's still a human who can take care of those situations, and it picks the safe thing (abort and slow).
 
Aborting and slowing is not always safe. This is why doing a half-assed job is worse than not doing it at all.
The abort seems to happen before it even crosses the line in to the next lane. Given that it is still tracking and reacting to the car in front, I don't see a better choice. You are not cutting off the driver in the new lane and not rear ending the driver in the current lane.

I'm curious what you think is a better choice. It matches the choice I would make as a human driver, and that I see other drivers make.
 
The abort seems to happen before it even crosses the line in to the next lane. Given that it is still tracking and reacting to the car in front, I don't see a better choice. You are not cutting off the driver in the new lane and not rear ending the driver in the current lane.

I agree, it's no big deal as long as you are aware of your surroundings. Some people lack situational awareness so they might worry about their safety in a dynamically changing situation like an aborted lane change. Personally, my first 8 years as a motorist was 80% on motorcycles and I think I developed better situational awareness because of that so I don't get worried about my safety too easily, especially in a car as maneuverable and responsive as the Model 3.
 
The abort seems to happen before it even crosses the line in to the next lane. Given that it is still tracking and reacting to the car in front, I don't see a better choice. You are not cutting off the driver in the new lane and not rear ending the driver in the current lane.

I'm curious what you think is a better choice. It matches the choice I would make as a human driver, and that I see other drivers make.

I've had it (try to) abort when it is partially into the lane already. This is very scary.

The better choice is to not make promises you can't (safely) deliver on. False positives matter and they are hard to deal with. Saying that the driver is always in control is a cop-out; it ignores the fact that things can go bad in a fraction of a second while the driver takes control. Elon simply doesn't get it. He sees a slick demo and assumes that in a few months they'll iron out the tiny little bugs and then it'll be great. 99% isn't good enough in ADAS and AV systems. You need more 9's than that. This is why the other ADAS systems do less than Autopilot tries to do -- because they can't do it reliably, they don't do it at all.

That said, with corner radar Autopilot could probably avoid most of these lane change aborts. This is a self-inflicted wound -- these radars are relatively cheap and bring so much benefit, and it's only Elon's hubris and ignorance that prevented them from including them. Given that they do not have radar coverage to the rear and in neighboring lanes, and the ultrasonics are largely useless at highway speeds, they need to make their camera-based detection much, much better. Like, add a couple of 9's at least. This is not easy.
 
  • Like
Reactions: Inside