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

Autopilot swerving on freeway with distinct lane markings!

This site may earn commission on affiliate links.
I was hoping that there would be a learning system - like swarm mentality...where every tesla driving down a road would "broadcast" what that road looks like to the cloud. Lane marker, manual driving, all have GPS information to improve the mapping. Next time you drive that path - rain or shine, the GPS will influence the AP.
Contributing to this learning mode - I understand that the air shocks will learn where the speed bumps are and inflate when the GPS say to do that. Speed bumps...lane marking...lots of cameras watching lots of lane markings every day ...updating maps so stormy nights are on known roads. Seems a bit futuristic..but not out of range in the technology I know of.
 
Why *both* hands?

Because the message on the dashboard every time you engage autopilot says "hands". Not "hand".
keep hands on wheel.JPG
 
More generally: The extensive agreement that you must acknowledge when you enable AutoSteer in Settings; the extensive cautions in the Owner's Manual; the notice at the bottom of the instrument panel when you engage AutoSteer; its nags when it lacks sensing of the wheel being held -- all these count for nothing?

No, they count for something. They're just not related to the topic we're discussing, which is Tesla misleads prospective buyers about autopilot's capabilities. It doesn't say Autopilot is in Beta. It doesn't say you have to keep your hands on the wheel. It doesn't say the car will swerve unexpectedly into the lane next to you whether there is a car there or not. Those are important things they should disclose - not after you buy the car but when you are looking at the car as a possible purchase. To represent autopilot as a mature product that always keeps the car in your lane is false and misleading. In my opinion.
 
  • Disagree
Reactions: gsxdsm and kort677
...Tesla misleads prospective buyers about autopilot's capabilities. ... In my opinion.
In my opinion they do not mislead prospective buyers about its capabilities. I'm pretty sure Tesla's attorneys agree with me as well, or the warnings would be written in a different way.

If you feel that Tesla has misled you in some substantial way, perhaps you should sell your Tesla and buy something with more proven technology - like a 1965 Dodge Polara.
 
As has been clearly established, each is entitles to their opinion, but I have to think that anyone buying a Tesla
who doesn't realize they're jumping on to a giant learning curve (with all that that entails) really hasn't been paying
attention, regardless of what Tesla may or may not say prior to purchase. Either that, or they're willfully ignoring
information that's right in front of them.
 
  • Like
Reactions: kort677
it's only a beta, it's only a beta.

I agree, I have a neighbor who is up there age wise and was thinking of buying a tesla, he was quizzing me on the ownership experience and I kept getting back to the fact that owning a tesla is not a casual relationship with the car. unlike every other car he's ever owned with the tesla he is going to have to go deeper than just fueling it, putting the key in and going. He'd need to learn about charging, the effects of climatic conditions and other things. the best analogy is that when driving a tesla you have to think like a pilot. to bring this back to AP on the tesla, when a pilot puts the plane onto auto, he doesn't take a nap, he is still monitoring everything that is going on, just like tesla tells you to do when using AP.
 
  • Like
Reactions: JenniferQ
Did our first big road trip this weekend -- round trip from Lafayette to Chicago. Mostly interstate and tollway in between, but plenty of city while there. Is the AP perfect? Certainly not. Does it make the mind-numbing interstate travel through the barren wastelands of NW Indiana just fly by? Absolutely! The TACC made bumper-to-bumper stop-and-go traffic along Lakeshore Drive just "not a problem" (assuming you're not in a hurry and fuming about being late). I was hoping we could do the whole trip on one charge, and the range estimate supported that hope, but both the hope and the estimate turned out to be a bit optimistic. Fortunately there was a Tesla dealership with a couple of superchargers less than ten minutes from our hotel (which offered no charging service with their valet parking -- c'mon, let's get with the program, Kimpton!).

After this experience it is my belief that Tesla should disable auto-lane-change above, say, 55 mph. At interstate speeds it is just plain dangerous to rely on it -- it just doesn't "see" cars coming up behind you on the left. IMO they should, for now, put there energies into refining the already-pretty-amazing lane following and defer the much-less-useful lane changing until it works much better than it does now.
 
I wonder if the new radar placement on the updated Model S nose will help the issue with hill crests and line of sight? It seems like moving the radar sensor from it's current location (close to the road surface) to a higher position (where it can see further down the road over more road imperfections) is an attempt to improve this behavior.
 
It seems like moving the radar sensor ... is an attempt to improve this behavior.
This sort of thing makes me nervous that the whole "continuous improvement through software and OTA updates" thing may
be something of a myth. The more Tesla relies on hardware change to fix/improve things like AP the less incentive they're
going to have to make changes that don't rely on that new hardware, leaving existing owners increasingly in the dust. If they
come out with AP 2.0, based on a whole new sensor suite, what incentive will they have to lift a finger improving AP 1.x?
 
I wonder if the new radar placement on the updated Model S nose will help the issue with hill crests and line of sight? It seems like moving the radar sensor from it's current location (close to the road surface) to a higher position (where it can see further down the road over more road imperfections) is an attempt to improve this behavior.
The hill crest problem people most often describe is auto steering losing track of the lane. It is the camera on the windshield that detects the lane lines. Moving the radar will not improve this.
 
  • Like
Reactions: mtavel
That's funny you mentioned what the sales guy told you - I asked an Audi sales guy when they were going to get autopilot recently and he said "what's that?". So I told him a little about Tesla's version and he said "our car does that now - lane keeping assist will drive it in your lane". I didn't press him on it but I think it probably doesn't do the same thing. Oh well, they're sales people - they can't sell if they tell the truth! I'm waiting for AP 2.0 - hopefully in the Model 3 so when I trade again it won't be as expensive. Swapping Model Ss every couple of years is quite expensive.

Other car companies are calling Adaptive Cruise Control, Auto Emergency Brkg, and Lane Keep Assist, as line items (ACC/AEB/LKA), not a single technology feature.

Other automakers are being a lot more cautious than Tesla when advertising the capabilities. Companies have been sued for floormats, can you imagine a class action against a driver's aid?

It's a really bizarre realm for automakers. While fatalities are down, crashes are up. People no longer desire to look out the windshield while driving. These driver's aids WILL reduce crashes, but cannot ever be 100% safe which appears to be the litmus test used in court against automakers when a safety system is not perfect. 1% defect level isn't much different than 100% defect level in the court's eyes.

Prediction, the next multi-billion dollar class action will be concerning driver's aids but will not be Tesla. Not deep enough pockets. They will target another brand with more assets and less political support. Most automakers were working on what Tesla calls "autopilot" in 2000. It has recently accelerated since it appears the government will mandate it's existence, yet not protect the industry like they did with airbags, or commercial airplane travel.
 
Most automakers were working on what Tesla calls "autopilot" in 2000. It has recently accelerated since it appears the government will mandate it's existence, yet not protect the industry like they did with airbags, or commercial airplane travel.

I was wondering about that. Surely the tech didn't exist in 2000, not to the level it is today.

I wouldn't think Tesla is immune to a class action suit because of their shallow pockets. If someone dies in a Model S or X while autopilot is engaged, I hate to think what the results will be. It could threaten the existence of the company. Even if Tesla wins, the cost in sales and bad publicity will be enormous. Hopefully we never find this out. If they're smart they are working hard on improving they system with the existing hardware, since thousands of us will be driving this rendition for years, and are at risk every time we use it.
 
Hillhold is still a tap for me. Lane keeping seems a bit more loosy goosy since the rainbow road update.

AP has not yet attained what I was sold which was that it will be completely autonomous from entrance ramp to exit ramp. It's about where I expected it to be by now though, despite what the sales folks told me even when I insisted that I didn't think it'd do what they promised by when they promised (and it hasn't). I still hope and expect it will get there... soon.
 
I've noticed the same anomalies on rising over a hill with a significant drop on the other side.
It's almost as if the sensors briefly lose track of the road.

I'd guess this could be prevented by averaging the road markings over a slightly longer time... but who knows.
The fact is, I always have my hands on the or near the wheel. When the car starts to veer toward the center line (which invariably for me has been when a transport has been immediately adjacent to me....my confidence is clearly shaken. I grab the wheel but this results in a slight "veer" toward the shoulder.

I'm guessing this isn't good for those drivers behind me who are looking at the Tesla inditia on the car and seeing a less than elegant driver assist response :)

Doug
 
  • Like
Reactions: Zybd1201
I am having a lot of issues with autopilot too. I understand everyone's frustrations. For me, the key statement that keeps sticking in my head that is really bothering me is when Elon said "it is probably better than a human at this point."

I could not disagree with that statement more. A number of times during every 45 minute commute I prevent it from doing something ridiculous. At times, it functions way outside of what I would expect it to do by swerving into other lanes.

It makes me very happy when it works well but the unpredictability of its mistakes is concerning.
 
Last edited:
My car has swerved into the lane next to me twice when a truck was in the path. I think my car hates pickups! I've sent the dashcam video to Tesla asking why its doing this. The first time was a couple months back and they never answered. I probably won't get an answer this time either but we need to keep telling Tesla that autopilot needs to be fixed. Tesla tells you not to trust it and be ready to take over at any moment, but what's the use of having autopilot if you can't relax and let the car drive? To me its not doing its job as advertised (on ramp to off ramp).

Thank you for sharing. Please keep doing it and don't let others convince you not to. I agree, it is not functioning as expected and this shouldn't be happening. The same thing is happening to me as well and I'm sad that posting some valid concern like this on the forum invokes some pretty negative responses. We are responsible drivers who bought a product and would prefer it not to literally drive into other cars. That is a very reasonable request.
 
It is puzzling to me how and why some of you are reporting these issues, while I and others haven't seen anything remotely of that nature.

Yes when you are cresting a hill it sure loses track of the lines and so wobbles a bit, but that is behavior I would expect since the forward sighting of the lines is diminished greatly.

On the right lane when there is an exit, it sure is confused of which way to go. But how would it know which lane is the continuation of the highway?

Bots? Yes I wouldn't trust AP. Cones? No way.

Every one of the situations where AP lane keeping is confused, are the ones you would expect it to be confused.

But on a typical decently lane marked highway lane, it just friggin drives.
 
I noticed this driving down the 405 at 1 am in the morning last night. Little to no cars on the freeway, clear lane markings, straight roadway, and fully illuminated by headlights. I was in the carpool lane and the car would randomly swerve out of the lane into the left shoulder and would correct itself after about 3-5 seconds. It did this twice on a 25 minute drive, both times the lanes were clearly visible. And it also swerved within its own lane as well for no apparent reason.

I have had the same experiences here in Florida. On the highway with little to no traffic around, sunny day, and randomly swerving while on autopilot. I would say it happens at least once or twice on my routine trips between Miami and Florida.
 
  • Informative
Reactions: Electricfan