I've had the opposite experience. I couldn't use EAP before the NoA update in the left lane because when the lane widened for the HOV entrance it would straddle it and if I let it probably drive into the pylons (no real gore point, just pylons). Now it correctly sticks to the inside lane line and ignores the HOV entrance.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Navigate on Autopilot is Going to Cause Accidents
- Thread starter Superspectral
- Start date
- Status
sperkin
Active Member
I agree, there are areas that improved. For example on the I-10E before merging into the I-60E in LA, lane #1 will split into 2 lanes by going double wide before lines are drawn. Before with EAP, it will just stay centered and ride the lines down the middle taking up both lanes. Now with Nav on AP, it will go into lane #2 even before the lines are drawn.
Good fix here, but is it 100% like this all the time? What if one day it freaks out and swerve as cars are trying to pass not knowing if I'm going to #1 or #2?
I think with current state of NoA it should be closed alpha test. I really feel this is rushed out and not very safe at all. Unlike EAP, it was down to 95%ish perfect as long as you know the limitations. As for NoA. It doesn't work most of the time as what the release notes describes. I don't even know it's limitations since it's suppose to change lanes and stay in the correct lane when the freeway splits but it doesn't do it most of the time. I have never gotten lane change to work on a normal work day commute.
Also, there are many fail points I didn't even mention yet because it's not safely related. Just poor traffic knowledge. For example, NoA wants to get off the I-10E to take the long exit road which seems fast but at the end it's 3 lanes merging into 1 lane. Taking that route on a Friday night adds 30min to the commute. It's better to stay on the freeway going 5mph. NoA doesn't know everyone slams their brakes in lane #1 on the I-60E near Atlantic Blvd exit but lane #4 is free flowing. NoA keeps trying to get me away from lane #4.
For 4 years on TMC I have been seeing posts similar to yours; will Tesla owners get into trouble using AP? Or in this case Navigate on Autopilot?
My answer is “yes”, some people will get into trouble. That is clear based on four years of Tesla Auto Pilot usage. Some owners don’t read the manual before using AP and don’t understand its limitations. Some owners make risky assumptions about what AP can and cannot do. Some owners seem incapable of understanding the clear warnings in the Tesla owners manual. Some get complacent and stop paying attention when using AP. And yes, sometimes AP makes mistakes and the driver does not take control quickly enough.
From the very beginning of Tesla AP some owners made serious errors, in a few highly-publicized cases costing them their life. The new Navigate on Autopilot is not fundamentally different; the driver still has to understand its limitations and pay close attention to the road.
Tesla stated recently that their data showed that Teslas with AP far less likely to be involved in an accident than the average accident rate of all other cars. See Tesla
QUOTE:
—————————————————————————
Here’s a look at the data we’re able to report for Q3:
- Over the past quarter, we’ve registered one accident or crash-like event for every 3.34 million miles driven in which drivers had Autopilot engaged.
- For those driving without Autopilot, we registered one accident or crash-like event for every 1.92 million miles driven. By comparison, the National Highway Traffic Safety Administration’s (NHTSA) most recent data shows that in the United States, there is an automobile crash every 492,000 miles. While NHTSA’s data includes accidents that have occurred, our records include accidents as well as near misses (what we are calling crash-like events).
That data of course does not include owners using the new Navigate on Autopilot. I’ve been using it extensively for weeks now. To me it is just a small but significant extension of what Enhanced Auto Pilot did before. It’s not fundamentally different. I do not anticipate a rash of new accidents among Tesla owners using Navigate on Auto Pilot.
Of course with so many Teslas on the road now, it is inevitable that there will be a sensational news story at some point saying “Tesla’s Navigate on Autopilot feature tried to kill me!”. It’s inevitable...
My experience has been different. I find that Navigate on Autopilot gets the lane changes and freeway interchanges right the vast majority of the time. Sometimes it doesn’t. And sometimes it has recommended a lane change in advance of an interchange and I thought “Why is it doing that” but then when I got to the interchange I realized that the lane change recommendation was appropriate! And this was for an interchange that I had used many many times.
insaneoctane
Well-Known Member
The other new result with NOA is that other drivers will continue to notice a higher rate of odd driving maneuvers from Teslas as NOA is used by more drivers. With frequent braking on lane changes and aborted lane changes, it's quite plausible that Tesla owners will begin to get a "bad rap" as unpredictable drivers.
In California, we might already have that reputation. A coworker asked me why the Teslas he sees on his commute often make very jerky weaving moves. He commutes on a freeway which has several lane splits that AP does not handle well, so I imagine he is seeing the results of the car waffling between the two lanes in the split, then the owner taking control.
I tell my friends I am just trying to keep the tires warm like they do with Indy cars while the caution flag is out.
⚡️ELECTROMAN⚡️
Village Idiot
NoA sounds a lot like AP2 when it first got activated. Scary and unpredictable. It took about a year before AP2 actually got pretty good. I would like to start using NoA so I can can continue to watch Tesla develop EAP, but it doesn’t sound like it’s worth downgrading my V8 UI to V9 yet.
Very well written response. I appreciate backing up your position with statistics.
But, you're in the SF Bay Area, near Tesla HQ. One would think that the dev team has done a lot more dogfooding and testing in the Bay Area than other parts of the country given where the most (all?) of the dev team for EAP, NoA, etc. is located.
I can't speak to all of So Cal w/authority as I didn't live down there w/a car for very long (nor did I have much time to explore and this was before any civilian GPS units w/map existed) but there are some pretty unique/unusual situations down there such as exits on the left and some really weird exits around downtown LA (not sure if it was the 110 or 101). Some (most?) of these problem reports are coming from outside the SF Bay Area.
Glamisduner
Active Member
I have used it several times, I'm using it as much as I can to help tesla with 0 faith in the system. In medium traffic where there is oncoming traffic and outgoing traffic on the highway when changing interstates its pretty scary and I almost always have to take over. I have already posted a video from the first day it was out, where it starts to follow the truck in front of me taking the wrong exit instead of the freeway offramp. it probably would have taken me right into a median if I didn't grab the wheel. I tried it here a few times now. The second time I got hard braking for some reason instead, and the 3rd time it moved past this exit but still failed to take the proper exit which is right after.
However over the weekend in lighter traffic it drove me like 80% of the way on one of my longer drives, I only had to take over once or twice just to too much merging cars, and once for a construction zone.
Another time it followed the line to the right which expanded about 3 lanes of distance and it felt like it was headed for the wall (exit was supposed to me into the left lane).
When I get the car back I think I will just use it in light traffic to help telsa collect data, but I don't know that I will ever actually trust it to work.
Also I have noted the car often does not slow down enough on off-ramps with tight turns, seems like it will only slow down to the posted speed limit? so if the recommended speed is 35 and the posted speed is 65 it would probably fly right off the road (at least some of the time, sometimes it does in fact seem to slow down?). You end up having to hit the brake and take over anyways (hopefully tesla gets data from this). I already noticed it often drives over the lane markers when exiting onto the next intersection, regardless of speed.
Sometimes it does not try to take every exit... Sometimes it will still do that.
You can see it tried to exit behind the buss, even though that is NOT the exit it is supposed to take.
However over the weekend in lighter traffic it drove me like 80% of the way on one of my longer drives, I only had to take over once or twice just to too much merging cars, and once for a construction zone.
Another time it followed the line to the right which expanded about 3 lanes of distance and it felt like it was headed for the wall (exit was supposed to me into the left lane).
When I get the car back I think I will just use it in light traffic to help telsa collect data, but I don't know that I will ever actually trust it to work.
Also I have noted the car often does not slow down enough on off-ramps with tight turns, seems like it will only slow down to the posted speed limit? so if the recommended speed is 35 and the posted speed is 65 it would probably fly right off the road (at least some of the time, sometimes it does in fact seem to slow down?). You end up having to hit the brake and take over anyways (hopefully tesla gets data from this). I already noticed it often drives over the lane markers when exiting onto the next intersection, regardless of speed.
Sometimes it does not try to take every exit... Sometimes it will still do that.
You can see it tried to exit behind the buss, even though that is NOT the exit it is supposed to take.
Last edited:
barjohn
Member
Yes, it isn't consistent and it is too unpredictable to trust. As I said in another post, where the 55 fwy ends in Newport Beach, it quit NoA but stayed on AP and sped up toward cars stopped at the traffic light requiring me to hit the brakes hard. The one consistent behavior I have witnessed is where the concrete barrier on the left comes out to the lane line where the columns support the over pass and it veers away from them into the adjacent lane until I take control. Here is a video clip where you can see that I cross the adjacent lane line. It shows at about 24 secs in.
I do not find Navigate on Autopilot “scary and unpredictable”. It is simply an extension of Auto Pilot, and anyone who has been using AP should know enough to realize that the driver has to stay alert and ready to take over at any time. If someone is not comfortable doing that, they should just drive their car the traditional way.
As for “downgrading” your UI to V9 I think that is an illogical attitude. If you want to continue to get Tesla firmware updates you have no choice but to accept the UI changes that are part of that process. That is the reality of 21st century software driven consumer products; they are not static during the period of ownership.
That reminds me of what happened in late 2014 when Tesla introduced AP and the S/X driver display changed to show the car image and got rid of the data display that had a connection to an analog gauge. Some people got upset and refused to accept any future firmware updates because they were so attached to the original driver display UI. I suspect by now even the most extreme diehards have either given in or sold their Teslas...perhaps to buy a newer one.
I watched your video 3 times. I honestly could not see the “veer”. But I’ll take your word for it, of course.
Agreed, a clear error, the car appeared to track the bus and then corrected itself and got back on the nav route.
Glamisduner
Active Member
That's me correcting it, I took over as I didn't want to drive into the median and wasn't sure what it would do with drive on nav enabled.
Yes so far the 3 has failed all 3 attempts to merge onto the 8 freeway right after there. The cross traffic is a real hazard even for human drivers. I have not had a chance to try since 42.4 update though, car will be at the body shop for probably a couple weeks now too.
Autopilot 2.0 + is much better than the AP1 car has been treating me though, with the new gray reflective paint they are putting down the freeways here the AP1 loaner car constantly disengages, won't engage or drives over the lines. It's a very noticeable difference! AP2.5 car's handle it better but still go over the lines sometimes.
barjohn
Member
I will try and capture a better image of what is happening since it does it every time. @ecarfan, tell that to my wife because it scared her so bad a couple of times she threatened not to ride with me in the 3 unless I agreed not to use NoA. We sometimes forget that as the driver we have learned to anticipate when AP will do something weird and compensate for it but our passengers are not as attuned and it can be very startling and unpredictable. Your programming of yourself to anticipate when something may happen and being prepared to take control when it does is the very definition of unpredictable. It may happen and it may not and you can never be certain.
Yeah it looked like it lost the left lane line as it was obstructed by the bus and followed it. Definitely not ideal, I'm sure it will improve. I've had mostly positive experiences with NoA. The only problems I have had is when it wants me to go right to pass and I know that lane is ending soon.
After reading the beginning of this thread, I tried NoA briefly on Sunday by taking the long way home on the freeway instead of taking the more direct way home on surface streets. For those familiar with the western Silicon Valley, I was traveling SB on CA-85 and took the ramp to NB I-280. 85 is 3 lanes wide and widens to 4 southbound lanes just before the Homestead Rd. exit immediately before the 280 interchange. The lines are extremely faded and AP predictably had trouble as the #3 lane grew into a "wide lane" before the new dashed line for the new #4 lane started. AP started out following the shoulder line as it should to take the exit, but in the middle of the wide lane it got confused and abruptly moved to the center. A car behind me thought I was moving to the #4 lane and started to overtake. They were probably alarmed at the sudden move and after the car had moved completely into the #4 lane they quickly passed me. Slowing for and following the offramp was smooth and uneventful and required no input even though there was some sort of message on the screen midway through the ramp curve. I did not manage to read it because I was too busy looking for other cars in preparation for the merge. This merge is relatively complicated because you have one lane from SB 85 merging into two lanes of traffic from NB 85 immediately followed by the #1 ramp lane merging into the #4 lane of I-280. The #2 ramp lane becomes an exit-only lane for Foothill Expressway. Anyone in the exit-only lane has to merge into the #4 lane of I-280 within a short distance. Both ramp lanes have to merge into the freeway within 280 yards while I-280 travelers that want to exit Foothill have to cross over in this same short merge area. Since there was little traffic on this Sunday morning, a quick application of the turn signal allowed the car to make the lane change out of the exit-only lane. As I recall, the car did not speed back up to the posted speed limit, I had to do that manually with the right thumb wheel.
As much as Elon and his team want to make Nav on AP a "general solution", there are definitely places with problematic features that would clearly benefit from high precision mapping to make the car handle the situation more like a human driver. I would very much like CalTrans to maintain better lane markings, but I think we all know how likely that is given the number of highway miles they have to maintain. However, I do appreciate the way they have painted alternating white and black paint for the lane lines on the concrete surface of I-280, especially where the lanes don't follow the concrete seam lines. I appreciate it from both a human vision and machine vision perspective. In certain lighting, it would be very difficult to discern the lane lines without the black paint due to the way the white paint blends into the concrete.
As much as Elon and his team want to make Nav on AP a "general solution", there are definitely places with problematic features that would clearly benefit from high precision mapping to make the car handle the situation more like a human driver. I would very much like CalTrans to maintain better lane markings, but I think we all know how likely that is given the number of highway miles they have to maintain. However, I do appreciate the way they have painted alternating white and black paint for the lane lines on the concrete surface of I-280, especially where the lanes don't follow the concrete seam lines. I appreciate it from both a human vision and machine vision perspective. In certain lighting, it would be very difficult to discern the lane lines without the black paint due to the way the white paint blends into the concrete.
- Status
Similar threads
