I suspect your lane change problem is you...if you release upward or downward pressure on the indicator stalk too quickly, ie, before the car has passed over the lane line - not sure if it has to be 50% or more - the car tries to go back to the lane from which you were trying leave. Its actually a good safety feature as it 'thinks" that you have seen something the car hasn't and are canceling the lane change. Try it and see.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Auto steer Not ready for prime time
- Thread starter IROAR
- Start date
-
- Tags
- Auto steer auto steering
What's interesting and kind of almost scientifically intriguing for me is that there seems to be almost no more divisive subject on the form then reports about autopilot behavior. What does it mean that a significant fraction owners report exactly this kind stuff and another large and perhaps smaller but still significant fraction vigorously disagree? I have to wonder if we're getting different functioning across different builds of the autopilot system? I have no knowledge of differences in camera systems or central computing systems, (although there is clearly updating of this Autopilot code across versions of the operating system – and that might explain some of these variabilities), but one would think with cars running the same version of the operating system and having what appear to be virtually identical camera systems that you could not get disparate behavior out of the system.
Perhaps that's just not true! I would challenge the folks who are disagreeing with this post to provide chapter and verse from their own experience with the autopilot that contradicts specific points here. I've observed much of this kind of difficulty and it's part of why I don't have high levels of confidence particularly when the autopilot system has to manage exits and other transitions. But again it may be that some owners are simply not getting this problem manifested – and those owners should step forward and clarify that their system is functioning differently. If that is confirmed, then Tesla has a major mystery on their hands.
canukus
Member
My wife and I have just finished a 6,500 mile trip (Ontario, Canada - San Antonio - Albuquerque - Ontario with many trips in between) and I tried to use the auto steer as much as possible. The slight veer at off-ramps was unsettling, but not very worrisome after we figured out the reason for it. It made driving in the left lane very tempting at times, but I don't like bad manners, so I spared others as much as possible. The automatic passing was generally fine.
This is beta stuff. We need to be constantly aware of that fact and be prepared. I enjoy helping Tesla work out the bugs and I think they are waaay ahead of any competition because of our involvement. Overall, it was fun and I expect Tesla will iron out the veering issue soon.
Onward!
This is beta stuff. We need to be constantly aware of that fact and be prepared. I enjoy helping Tesla work out the bugs and I think they are waaay ahead of any competition because of our involvement. Overall, it was fun and I expect Tesla will iron out the veering issue soon.
Onward!
jrwaters1
Member
I found that the more you use it the better it gets. It seems to function better over time and as you build up a bit of trust you become less sensitive to what might happen and deal with what is really going on. I have seen major improvements over the last year in the software. I have an S but I suspect all models are much the same in this respect.
Joer293
MODEL X P100DL
I’ve driven about 12 vehicles with AP1, 1.5, and 2. Haven’t tried 2.5 yet. Only difference in 1.5 is the updated radar accuracy. 2 is obviously cameras and a custom computer.
The main problem with AP, is the lack of owner configurable options. Those few owners who have rooted the OS, can tweak it for their location and road conditions. One size doesn’t fit all road conditions. I would suggest At a minimum for “performance” trim cars or maybe an “Easter egg” to open a driver profile AP customization panel. Centering behavior, radar, distance, aggressiveness, more than a dozen tweaks are available.
The main problem with AP, is the lack of owner configurable options. Those few owners who have rooted the OS, can tweak it for their location and road conditions. One size doesn’t fit all road conditions. I would suggest At a minimum for “performance” trim cars or maybe an “Easter egg” to open a driver profile AP customization panel. Centering behavior, radar, distance, aggressiveness, more than a dozen tweaks are available.
I do not believe that to be an accurate statement. The cause of the crash was driver inattention. I don't care what you are driving, if your car crashes into a Jersey barrier or other lane divider it's on the driver not the car. I believe in that specific accident forensics said that had been no driver input in over 5 minutes. These are not self driving cars ... yet.
I use it every day going to work and home. It's a short drive but it does a great job and I feel more relaxed and can pay attention to the cars around me, not just focused on the lane and car in front of me. There are tricks to using Autopilot as other have stated, and sometimes I have to turn the Nav on Autopilot (NoA) off when I don't want to take the route provided. I've noticed that NoA will brake if it thinks you are going to miss an exit or interchange, even though I don't want to go that way. I'd rather take an extra 5 minutes but stay on the freeway in Autopilot, than get off and take surface streets where I have to do more driving.
There seem to be two basic camps. Those who want close to autonomous self driving and those who accept that this is a work in progress and take the admonishment to keep a light resistance on the wheel and eyes on the road. I view the feature as a way to relieve stress while staying alert. I use it on local roads as well as limited access highways and accept the hiccups and am always ready for them.
I think you need to run the data against those who paid for and own AP to those who are on the brief trial. The trial users will naturally tend to find fault, as that confirms their decision against purchasing it. Vice versa for those who bought and own it. Like seemingly everything at the moment, the real story depends on your point of view.
Moose408
Member
That is true. He was not paying attention.
However that particular intersection does have unusual (I would say incorrect), lane markings which is confusing to AP. The driver had actually complained about it several times to Tesla that his car would veer towards the barrier often. Sounds like he was relying to much on AP and should have known what it was going to do there.
Moose408
Member
My wife's Mercedes has adaptive cruise control and auto-steer and I would say that Tesla's auto-steer is 10x better.
The Mercedes works OK in stop and go traffic, but is only good for mostly straight highways when going at speed. It doesn't like curves to the right.
The adaptive cruise control on the Mercedes is overall better than the Tesla, much smoother starts and stops. The only issue is that Mercedes minimum follow distance is not close enough for rush-hour traffic (leaves too large a gap).
The Mercedes works OK in stop and go traffic, but is only good for mostly straight highways when going at speed. It doesn't like curves to the right.
The adaptive cruise control on the Mercedes is overall better than the Tesla, much smoother starts and stops. The only issue is that Mercedes minimum follow distance is not close enough for rush-hour traffic (leaves too large a gap).
JeffK
Well-Known Member
Is Tesla using the right mathematics for lane follow, lane change, and exit?
- using certainty polynomial regression when parts of a lane indicator disappears when, for example,
... going past an exit
... going past a highway intersection
... going past other natural breaks on the right side
- using risk (threat divided by avoidance) to determine where within the lane to position the car? For example in N America:
... to the right side when on a two lane highway when a oncoming car is approaching and no risks on the right side
... to the left side when no car is approaching and in deer, elk, or moose country
... somewhere between when different threats of different risks are detected on both sides of the car
... all stuff we learned in driving ed class
- using adaptive speed for curves and corners known by vision and/or GPS
- using adaptive speed for a potential rear-end collision (someone behind you about to collide)
With all of the bright drivers out there, many of who are scientists and engineers, you would think that Tesla would tap into them with special agreements and training to help the firmware and software mature faster.
- let the users suggest better mathematics (many of us can guess, via behavior, what math the car is using and offer better approaches)
- letting qualified/trained users test closed-beta software
- the more driving experiences that can be sent back to Tesla from trained users, the better for safe driving for all.
What Tesla is doing will end up saving thousands of lives. If that process can be safely accelerated, isn't it worth a try?
- using certainty polynomial regression when parts of a lane indicator disappears when, for example,
... going past an exit
... going past a highway intersection
... going past other natural breaks on the right side
- using risk (threat divided by avoidance) to determine where within the lane to position the car? For example in N America:
... to the right side when on a two lane highway when a oncoming car is approaching and no risks on the right side
... to the left side when no car is approaching and in deer, elk, or moose country
... somewhere between when different threats of different risks are detected on both sides of the car
... all stuff we learned in driving ed class
- using adaptive speed for curves and corners known by vision and/or GPS
- using adaptive speed for a potential rear-end collision (someone behind you about to collide)
With all of the bright drivers out there, many of who are scientists and engineers, you would think that Tesla would tap into them with special agreements and training to help the firmware and software mature faster.
- let the users suggest better mathematics (many of us can guess, via behavior, what math the car is using and offer better approaches)
- letting qualified/trained users test closed-beta software
- the more driving experiences that can be sent back to Tesla from trained users, the better for safe driving for all.
What Tesla is doing will end up saving thousands of lives. If that process can be safely accelerated, isn't it worth a try?
JeffK
Well-Known Member
This is the exact opposite of their approach. While there may be some rule based stuff, you'll never create a decision tree capable of FSD. What you're describing is more akin to what Andrej Karpathy refers to as software 1.0 not software 2.0. It's similar to the approach researchers used (unsuccessfully) in the 70s, 80s, and 90s for attempts at FSD.
I completely agree with this post. I have an AP1.0 Model S built in late 2015. Even though I do not have the latest hardware, autosteer has continued to improve and is now quite stable, in my opinion. I have not seen much of the behavior the OP posted, and, at this point, am thinking some folks just aren't willing to use it at this state of development, for their own very good reasons. However, if you believe anything in the press about how other systems, such as Cadillac's, are "more advanced," you are being hoodwinked. Sure, OK, Cadillac's system is less intrusive than the "hold the wheel" nags of AP as it watches your eye movements, but, other than that, it is in no way more advanced. It only works on roads already "mapped" by GM, so is very limited. Tesla is far and away ahead in the move to autonomous driving, but we are not there yet. AP is beta and they mean it. It is still quite useful, in my opinion, for almost any situation today. Just be aware when driving. It is not meant for use while doing anything else but paying attention. It leaves me much more relaxed, in general, especially on long trips. JMO.
NOBODY is going to see full autonomous with only EAP. I suspect Nav on A/P will be GA only for those purchasing FSD.
Keep in mind there are now two videos of a Tesla, the first one definitely an S, the second on perhaps a Model 3, doing FSD, stopping at lights and stop signs, navigating pedestrians and bikers, etc. They do have their own chip going GA within a few months as well. Clearly they are moving away from NVIDIA.
RachH
"Christine" Model 3 Blk Obsidian, LR, AWD
I did a cross country trip from Texas to CA and also a trip from LA to San Fran and back during my 30 day free trial.
My hands were not heavy enough on the wheel so I had to grab the wheel- sometimes a little obnoxiously so. I tried auto pilot here and there and it was the same everytime. So to me, as a woman who may not be as heavy as some of you men, it was not worth it to me.
That being said, the Cruise Assist is AWESOME!!!!! You use it like regular cruise control (w/one pull down of the bar) and when you get close to traffic, it slows down. When the cars get out of the way it speeds back up to where you were. And of course if someone cuts in front of you, it immediately slows down. Btw, I'm sure you all know this and have tried it- I hope
A couple of different times, this option really helped in a big way. On the way to San Fran, traffic went from 60 or so to almost a dead stand still because someone cut in front of someone else. The cruise assist sensed it immediately and slowed down to a stop. It may even have reacted faster than I would have. AND without having to slam on brakes.
I love this feature so much and it is a life saver on roadtrips and even in traffic. I wish they would sell this feature just by itself instead of as part of the whole Auto Pilot or even better, make it standard. I have done some research and it is standard on a lot of luxury vehicles.
Still trying to decide if I want to put out the money for the whole thing for just this one feature.
My hands were not heavy enough on the wheel so I had to grab the wheel- sometimes a little obnoxiously so. I tried auto pilot here and there and it was the same everytime. So to me, as a woman who may not be as heavy as some of you men, it was not worth it to me.
That being said, the Cruise Assist is AWESOME!!!!! You use it like regular cruise control (w/one pull down of the bar) and when you get close to traffic, it slows down. When the cars get out of the way it speeds back up to where you were. And of course if someone cuts in front of you, it immediately slows down. Btw, I'm sure you all know this and have tried it- I hope
A couple of different times, this option really helped in a big way. On the way to San Fran, traffic went from 60 or so to almost a dead stand still because someone cut in front of someone else. The cruise assist sensed it immediately and slowed down to a stop. It may even have reacted faster than I would have. AND without having to slam on brakes.
I love this feature so much and it is a life saver on roadtrips and even in traffic. I wish they would sell this feature just by itself instead of as part of the whole Auto Pilot or even better, make it standard. I have done some research and it is standard on a lot of luxury vehicles.
Still trying to decide if I want to put out the money for the whole thing for just this one feature.
