krazineurons
krazineurons
i recall @MXWing who was part of first 1000 AP2 updates had mentioned that he saw cars in adjacent lanes after getting the subsequent bugfixes update.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autopilot for HW2 rolling out to all HW2 cars today!
- Thread starter HumanGenome
- Start date
3Victoria
Active Member
Do you also have experience in AI? Many have underestimated Tesla (repeatedly). The new hardware is only two months old, and as has been previously explained, its AI needs to learn over miles. The best way to get those miles is to let the new cars record them. The system is optional, and you can just not activate it for another couple of months.
Hello fellow Ap2 folks... so got in some serious driving today with both Auto Steer and TAAC .. honestly both work amazing ... for those in Charlotte - tried on I-77 to/from the work .. autosteer will go to Zero and then start back up again - same with TAAC - autosteer at 45 mph already pretty awesome - great there is more to come!
This is my second post on TMC - after watching the forum for many months, and then I pulled the trigger on my first Tesla MS 60 as a commuter car replacement in October. I was one of the weird cases that had to make a choice between AP1 and AP2 after I ordered my car. I chose AP2 on faith, and have been driving the car myself since delivery in early December. I have enjoyed driving it so much that I have not been too concerned about the wait for EAP. Yesterday I got the update, and took it for a quick spin right after the update installed. I received no camera calibration messages that I can recall, and everything worked immediately. This morning I found enough slow traffic that I could engage about 20 miles of Auto Steer with EAP this morning on the North Dallas Tollway. TACC was already working great, and when the traffic slowed down then I engaged Auto Steer. Turned on fine, and even followed the car in front of me through a new construction zone (a common occurrence on the NTTA). The only concern I had is when the car would hug the right side of the lane, seeming to be pretty close to the cars in adjacent lanes, but still within the lines. The steering wheel is constantly moving, making micro adjustments - even at very slow speeds. It almost seems (totally guessing here) as if the front camera is the only one driving the show right now, and that the system is sensing the input from the side sonar (radar?), but is not fully utilizing the cameras on the side pillars in order to see the cars in adjacent lanes. I am making an un-educated guess here, but it seems like the side pillar cameras are not yet fully activated at this point, and that is causing the uncertainty and waffling in the lane. The "lane dancing" feels like the car is trying to hard to play the game, yet has one hand tied behind its back. I never felt like there was any danger, it felt more like a new driver that was uncertain and needed guidance. It seems like there is a key ingredient missing that is pending and yet to be added (hardware and/or software), and when it is turned on, then Auto Steer will be ready for primetime. TACC worked perfectly for me all day. I am having a hard time thinking about driving in any other way now - especially with my 100 mile round trip daily commute. I came to a full stop multiple times behind cars at red lights, and the car smoothly continued on with traffic several minutes later when the light changed, just like I would have done.
Recommendation based on today's experience with TACC - I would turn off the TACC as you approach crowded city driving, especially if you are transitioning from a fast highway (70 MPH) to a more dense city environment with traffic signals. The speed will stay locked at 79 MPH unless you turn off the TACC or adjust it downward for the new speed limits in town. Cars leaving the lane in front of you could cause you to GO REAL FAST depending on where you last left the cruise control. Got a little coffee on my shirt this morning as a result (LOL) but it served as a great reminder of how much fun it was to get through some of the worst traffic in North America. For the first time in a very, very long time, I am excited to go tackle the tollway in the morning. This is why I bought the car, and I know that it will get better over time.
Recommendation based on today's experience with TACC - I would turn off the TACC as you approach crowded city driving, especially if you are transitioning from a fast highway (70 MPH) to a more dense city environment with traffic signals. The speed will stay locked at 79 MPH unless you turn off the TACC or adjust it downward for the new speed limits in town. Cars leaving the lane in front of you could cause you to GO REAL FAST depending on where you last left the cruise control. Got a little coffee on my shirt this morning as a result (LOL) but it served as a great reminder of how much fun it was to get through some of the worst traffic in North America. For the first time in a very, very long time, I am excited to go tackle the tollway in the morning. This is why I bought the car, and I know that it will get better over time.
Yep I had the lane dancing today as well. I hope they smooth out its twitchyness in an update and maybe you are right, maybe its using limited camera input for the initial release - although I don't know why they would do that !
TACC worked well for me, I adjusted the gap to 2 or 3 car lengths and it helped my heart rate, gap of 1 was braking too late for me
My brain tells me you are right and they still haven't figured out how to make it all work. But then how do we explain the FSD video that was released? My heart wants me to believe my brain is wrong and I'll be using AP2 in a month. I'd hate to go there on this thread, but if I feel I was purposefully lied to by Tesla regarding AP2 it will not be pretty. I'm trying very hard to control myself and think positive.
acarlos1000
Member
I tried Autosteer today on my way to the office and it worked perfectly on Stop and Go Traffic on the highway.
See video below.
That same poster qualifying their statement with non relevant credentials is even fuzzier.
I was able to use Autosteer today on the highway for maybe 10 miles . Honestly it worked better than I thought it was going to. As for TACC, it worked fine too. I've had a lot of experience with Subaru's system (which is excellent - I believe it is stereo-optical) so I'm pretty comfortable with that already. I did have a slowdown going under a sign on the highway, and an erroneous collision warning about nothing, but it is beta.
The fact that this is an AI system as opposed to a linear, or even traditional non-linear, control system makes it almost impossible to divine what the design philosophy is. So what sensors are doing what or what is going to improve when whatever is enabled is all just idle speculation. Honestly, in neural net systems even the designers don't really know what the network is using whatever sensors for! That being said, I hope it gets better soon because its understanding of the world around the car does not engender a great deal of confidence. The dancing lane lines are a perfect example; why do they move SO MUCH when the car is stopped? Or, I was stopped at a light and watched a phantom car drive off in front of me and into space ... but there were no cars moving anywhere! It was just me and a dozen cars waiting at a light. It leaves you with the idea that Tesla has sat a gazillion monkeys down at a gazillion typewriters and gotten "It was the best of times, it was the blurst of times," which is pretty good but not quite there.
I'm honestly not worried because Mr. Musk has devised at least one control system that does a pretty fantastic job of driving cars. Oh, and another system that flies rockets backwards through space into the atmosphere and lands them on floating barges on the ocean ... AT NIGHT! I just wish that in a fairy tale world I'd be privy to the design philosophy because how they're going to take this goofy world of dancing lane lines and phantom cars to a completely operable self driving car is going to be really cool. In the meantime though, I hope they don't lose confidence of less geeky customers.
The fact that this is an AI system as opposed to a linear, or even traditional non-linear, control system makes it almost impossible to divine what the design philosophy is. So what sensors are doing what or what is going to improve when whatever is enabled is all just idle speculation. Honestly, in neural net systems even the designers don't really know what the network is using whatever sensors for! That being said, I hope it gets better soon because its understanding of the world around the car does not engender a great deal of confidence. The dancing lane lines are a perfect example; why do they move SO MUCH when the car is stopped? Or, I was stopped at a light and watched a phantom car drive off in front of me and into space ... but there were no cars moving anywhere! It was just me and a dozen cars waiting at a light. It leaves you with the idea that Tesla has sat a gazillion monkeys down at a gazillion typewriters and gotten "It was the best of times, it was the blurst of times," which is pretty good but not quite there.
I'm honestly not worried because Mr. Musk has devised at least one control system that does a pretty fantastic job of driving cars. Oh, and another system that flies rockets backwards through space into the atmosphere and lands them on floating barges on the ocean ... AT NIGHT! I just wish that in a fairy tale world I'd be privy to the design philosophy because how they're going to take this goofy world of dancing lane lines and phantom cars to a completely operable self driving car is going to be really cool. In the meantime though, I hope they don't lose confidence of less geeky customers.
DoubleACL
Member
Having some difficulty finding roads where Autosteer is enable-able. Freeway or tollway onramps seem to be the most often enabled locations when under the 45 mph limit. The only problem is that many of these are sharp 270 degree curves and I had to disengage Autosteer several times for fear of the car not turning fast of sharp enough to avoid the guard rail.
Mobileeye has been around for a long time. The fact that Tesla is anywhere near AP1 on HW2 makes me hopeful, not worried.
On my way home on my exit there is an arrow marked in the middle of the pavement, and it is not painted but the arrow is made of raised bots, so that when you drive over them it vibrates the car like driving over rumble strips.
The width of the arrow is only a few inches smaller than the distance between the tires. I always aim to drive exactly dead center on that lane so that I don't run over the bots. I get it only half the time and the other times I end up running either my left or right tires over atleast one of the bots.
AP does it perfectly all the time. Never once it ran over the bots and you can do that only if you manage to get the car centered within an inch or two.
My dream job would be working for AP team in Tesla. I would even work for free !
The width of the arrow is only a few inches smaller than the distance between the tires. I always aim to drive exactly dead center on that lane so that I don't run over the bots. I get it only half the time and the other times I end up running either my left or right tires over atleast one of the bots.
AP does it perfectly all the time. Never once it ran over the bots and you can do that only if you manage to get the car centered within an inch or two.
My dream job would be working for AP team in Tesla. I would even work for free !
My car was also one of the first 1000 and i never saw adjacent lanes car being shown..
see videos from my playlist
DJ 240V
Member
Sudden slowdown with TACC under a bridge. TACC was engaged at 75mph . Fairly light traffic . The car suddenly slowed down as I went under a bridge. I have noticed this before too. Happened to me about 3 times in my 600 odd miles driving with TACC over the past 3 weeks. This time there was no one behind me and the slowdown was a quick sudden hesitation .
Stay safe and alert , especially if someone is tailing you close.
Stay safe and alert , especially if someone is tailing you close.
DJ 240V
Member
So on HW2 I only see a single lane display with EAP. Are you referring to AP1?
Last edited:
Miss the AP 1.0 on my P85D.... Had the first one in NJ. Custom build vehicle. Found a P100D in December and got it delivered end of Dec. AP 1.0 was amazing. Wish I had kept my car and placed an order instead but the Paramus Tesla found exact color, specs of what I wanted. They got the vehicle somewhere in Alabama and brought it to NJ. But so far no AP 2.0 upgrade yet. Ive read through these posts, and you guys should be lucky you have got the software push for AP 2.0. Can't wait. Wish I had stayed with my original plan and ordered HW 2.0 in Jan for delivery in spring. Miss my old AP 1.0 already. I like the concept of being a guinea pig but it sucks right now not having the features I enjoyed for the past 2 yrs. When I got my new HW2 car in Dec, I was told only a few weeks for the AP. Any part of it would be helpful right now, instead of being penalized for getting the car in Dec when I did. Its a cool car but honestly wish they had never found it, and I kept my car and ordered my Tesla like I did 2 years ago. They pushed super hard to buy it before the year was over, something I had never experienced at Tesla. My second Tesla at Paramus, the experience has been nothing like the adventure 2 years ago.
Scott-P
Member
I'll chime in, too.
Almost 2 hours today in Seattle traffic. I agree that TACC and AutoSteer perform as advertised, with some interesting caveats. Actually, TACC seems to work precisely as advertised. I would love to see it go all the way to 0, and re-engage (which AutoSteer can do now), so it is great for slow and go, but not for stop and go traffic. It has difficulty with traffic merging into your lane ahead, since it isn't seeing cars in other lanes reliably, so I found the need to keep playing with the follow distance to make merges work safely.
AutoSteer is more interesting. It is somewhat jittery, which seems worse at slower speeds, almost as though it has trouble seeing the lane edges at low speed (say 5-15 mph). It does a very good job of playing follow-the-leader (especially in the speed range from about 20 to 44 mph), but like TACC doesn't handle merging traffic well at all (it seems to get confused as to where the traffic is coming from, as though it can't see the merging vehicle until it is fully in front of the car)--I had to take over several times for merging. It also has trouble with off ramps, wanting to veer off onto the off ramp especially if the car ahead exits. It also seems to want to follow the car ahead into another lane when the car ahead changes lanes right up to the point where it sees the lane edge (you can almost hear it "follow-follow-follow--Oops, it's leaving, don't follow. Where's my new leader?"). Similarly, I don't think it is capable of handling a lane change yet, probably as it isn't really picking up the cars behind and along side. However, in a long stretch of stop and go and slow and go in a middle lane, it works perfectly.
My only real negative comment is that I don't like the way AutoSteer disengages as the high end. When traffic lightens up, and we all speed up to 45, and the time comes to add some accel to go up over 45, I don't need a bunch of warning bells and a graphic on my screen; I just added power, so I know that the AutoSteer limit has been exceeded. Better would be a nice simple "ping", and a release of the tension on the wheel as AutoSteer disengages when manually overriden. Perhaps a piece of code to see if it is getting force feed back on the wheel as the accelerator is pressed, and then the car knows that I know that I am driving. This would be consistent with what it does when I overpower the wheel to make a lane change (or prevent it from following the leader onto an off ramp) as well as what it does when I hit the brake (it is also more like how a typical aircraft autopilot disengages when overriden, which seems to be the model here).
This is definitely a step in the right direction. I can't wait to see how it evolves!
Almost 2 hours today in Seattle traffic. I agree that TACC and AutoSteer perform as advertised, with some interesting caveats. Actually, TACC seems to work precisely as advertised. I would love to see it go all the way to 0, and re-engage (which AutoSteer can do now), so it is great for slow and go, but not for stop and go traffic. It has difficulty with traffic merging into your lane ahead, since it isn't seeing cars in other lanes reliably, so I found the need to keep playing with the follow distance to make merges work safely.
AutoSteer is more interesting. It is somewhat jittery, which seems worse at slower speeds, almost as though it has trouble seeing the lane edges at low speed (say 5-15 mph). It does a very good job of playing follow-the-leader (especially in the speed range from about 20 to 44 mph), but like TACC doesn't handle merging traffic well at all (it seems to get confused as to where the traffic is coming from, as though it can't see the merging vehicle until it is fully in front of the car)--I had to take over several times for merging. It also has trouble with off ramps, wanting to veer off onto the off ramp especially if the car ahead exits. It also seems to want to follow the car ahead into another lane when the car ahead changes lanes right up to the point where it sees the lane edge (you can almost hear it "follow-follow-follow--Oops, it's leaving, don't follow. Where's my new leader?"). Similarly, I don't think it is capable of handling a lane change yet, probably as it isn't really picking up the cars behind and along side. However, in a long stretch of stop and go and slow and go in a middle lane, it works perfectly.
My only real negative comment is that I don't like the way AutoSteer disengages as the high end. When traffic lightens up, and we all speed up to 45, and the time comes to add some accel to go up over 45, I don't need a bunch of warning bells and a graphic on my screen; I just added power, so I know that the AutoSteer limit has been exceeded. Better would be a nice simple "ping", and a release of the tension on the wheel as AutoSteer disengages when manually overriden. Perhaps a piece of code to see if it is getting force feed back on the wheel as the accelerator is pressed, and then the car knows that I know that I am driving. This would be consistent with what it does when I overpower the wheel to make a lane change (or prevent it from following the leader onto an off ramp) as well as what it does when I hit the brake (it is also more like how a typical aircraft autopilot disengages when overriden, which seems to be the model here).
This is definitely a step in the right direction. I can't wait to see how it evolves!
Thanks for this video, but I have to say "589 Wh/mi over the last 15 minutes!"
