-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autopilot Fail
- Thread starter AlexHolmes
- Start date
Agreed on both hating to be Debbie, but also recognizing that there are about a billion different scenarios that AP needs to anticipate/decifer on a real time basis. Posters that open with "I would never use AP in a situation where..." need to remember that there are an infinite number of subtle potential problems.
lunitiks
Cool James & Black Teacher
What if Tesla streamed the entire trip live on their website?
How cool wouldn't that be!! 8-camera visual coverage, a couple of interior go pros (one at the driver, one at the IC), full radar, sonar and GPS status --- all live on tesla.com!
Man that would be a killer marketing stunt. I'd be glued to the screen
How cool wouldn't that be!! 8-camera visual coverage, a couple of interior go pros (one at the driver, one at the IC), full radar, sonar and GPS status --- all live on tesla.com!
Man that would be a killer marketing stunt. I'd be glued to the screen
lunitiks
Cool James & Black Teacher
Let's hope not. But seriously I think it would be a brilliant move.Maybe literally
IMO, Tesla doesn't *have* to prove they can make it by december, but at least they've promised to try. And what a great opportunity that is to have the whole world watching, analyzing, potentially being amazed out of their pants
We’re kind of getting off topic here, but my guess is that Tesla is already making a number of LA to NY runs to see how their technology works and to plot out the best route. I suspect the ‘official’ run will be well publicized and maybe live streamed, but I’m sure that will be far from the first run and far from the first successful run. Tesla wants to make sure it will work and it will be repeatable before they show the world that they can do it.
I also think that even if or when the LA to NY run succeeds it doesn’t mean that FSD will be released to our cars anytime soon. I suspect it will be more of a proof of concept thing and it will still be months (or more likely years) before we can start the car and close our eyes. They will still need time to work out solutions to all of those little one off situations that may come up.
And while I’m speculating here, I also suspect that we will see FSD come in stages. The car will first be able to do things like switch highways or stop and start at traffic lights and stop signs, then maybe it will be able to make turns and navigate along a route, but we will still need to pay attention and keep our hands on the wheel through all of these improvements. Then, once the technology is proven maybe will we be able to let go of the wheel and stop paying attention.
I also think that even if or when the LA to NY run succeeds it doesn’t mean that FSD will be released to our cars anytime soon. I suspect it will be more of a proof of concept thing and it will still be months (or more likely years) before we can start the car and close our eyes. They will still need time to work out solutions to all of those little one off situations that may come up.
And while I’m speculating here, I also suspect that we will see FSD come in stages. The car will first be able to do things like switch highways or stop and start at traffic lights and stop signs, then maybe it will be able to make turns and navigate along a route, but we will still need to pay attention and keep our hands on the wheel through all of these improvements. Then, once the technology is proven maybe will we be able to let go of the wheel and stop paying attention.
Actually it is fun only when you come here and create a new topic with screaming headlines, "TESLA TRIED TO KILL ME".
I hate to be a Pollyanna
but an AP1 car could achieve that run today using AP 95% of the time. I've several 1000 mile runs under my belt and rarely take it out of AP on the highway(of course I understand AP1s limitations and don't do stuff like travel in the right lane etc.) Really isn't too much of a stretch to get close to 100%...just leave New York on a highway and enter LA on a highway. The only tricky part with AP is exiting the highway for SCing.
HankLloydRight
No Roads
Yes that 5% piece would certainly require a hella lot of luck.
His statement: “In November or December of this year, we should be able to go all the way from a parking lot in California to a parking lot in New York with no controls touched in the entire journey.”
It could turn into one of those "first time..." contests e.g. first transcontinental flight etc. "First Driverless Car Coast to Coast"
His statement: “In November or December of this year, we should be able to go all the way from a parking lot in California to a parking lot in New York with no controls touched in the entire journey.”
It could turn into one of those "first time..." contests e.g. first transcontinental flight etc. "First Driverless Car Coast to Coast"
Xminus6
Member
My X on 17.17.17 did this when I was entering the Caldecott Tunnel going West a couple weeks ago. Tried to follow the old lane lines that designated the alternating direction travel periods of Bore #2 in the tunnel. It nearly steered me directly into the barrel barriers. Bore #2 is only one direction now permanently and has newer markers for the new direction but the X ignored them. Pretty freaky.
Todd Burch
14-Year Member
Twice a day, every day, I drive through a 10-15 mile construction zone with lanes jogging back and forth and Jersey walls right next door. While I'm guarded when the walls or other cars are close, AP1 is remarkably good in these situations. Perfect? No. But even in places where you can see the old lane lines that they've partially scraped off, almost always my AP1 car follows the correct lines. When I say "almost always", I think maybe once or twice out of a hundred trips will it choose the wrong lines.
Of course, performance depends on a lot of conditions. In my mind part of operating AP is paying close attention to the conditions it can handle, the conditions that confuse it, and being extra vigilant when you encounter conditions where its performance is questionable.
Of course, performance depends on a lot of conditions. In my mind part of operating AP is paying close attention to the conditions it can handle, the conditions that confuse it, and being extra vigilant when you encounter conditions where its performance is questionable.
I think it's important to point out that there were three things that impacted driving on AP in that situation.
One was the jersey barriers being so close. This isn't where someone wants to use AP as you don't have any time to correct.
Secondly was the old previous lines. The problem with those is they can still be there years after the construction has finished. How is a person supposed to know where a construction zone used to be?
Thirdly is the side spot monitoring of the car. As the software currently stands it doesn't use the side camera to detect if there is a car there. Instead it's only a slightly better side monitoring system than HW1, and it's not very good. It's not very good because it relies on ultrasonic sensors.
You can judge the driver all you want, but the system failed.
I hope the OP emails a link to the video to the autopilot team.
One was the jersey barriers being so close. This isn't where someone wants to use AP as you don't have any time to correct.
Secondly was the old previous lines. The problem with those is they can still be there years after the construction has finished. How is a person supposed to know where a construction zone used to be?
Thirdly is the side spot monitoring of the car. As the software currently stands it doesn't use the side camera to detect if there is a car there. Instead it's only a slightly better side monitoring system than HW1, and it's not very good. It's not very good because it relies on ultrasonic sensors.
You can judge the driver all you want, but the system failed.
I hope the OP emails a link to the video to the autopilot team.
Yeah, right now I have a mental list of things that are dicey (HW2). This is mainly for highway driving, since I don't use AP on local roads. I'm always on alert for these conditions:
1) Anything on the road that could be interpreted as unclear or conflicting lane boundaries. Doesn't actually have to be an actual lane mark. Could be a shadow (think, power line shadow along a road). Could be something spilled on the road.
2) Double yellow lines. Not sure why, but they are sometimes weakly detected as a lane marking even when the marks themselves are good.
3) Trucks or vehicles with a large profile - being behind, passing, or being passed by.
4) Narrow lanes, lanes that split or merge, lanes with cones, poles, or any barriers in or near them. Tunnels and bridges.
5) Approaching and passing under overpasses and large highway signs.
6) Curves that are more than just a gentle curve.
7) Passing exits. I usually want to stay on the highway but sometimes the car doesn't
8) Approaching stopped or slowed traffic.
9) Using automatic lane changing. Sometimes AP thinks a car two lanes over is in the lane I'm trying to change into and decides to slow way down.
10) Vehicles cutting you off or merging in heavy traffic - AP may not notice them in time.
1) Anything on the road that could be interpreted as unclear or conflicting lane boundaries. Doesn't actually have to be an actual lane mark. Could be a shadow (think, power line shadow along a road). Could be something spilled on the road.
2) Double yellow lines. Not sure why, but they are sometimes weakly detected as a lane marking even when the marks themselves are good.
3) Trucks or vehicles with a large profile - being behind, passing, or being passed by.
4) Narrow lanes, lanes that split or merge, lanes with cones, poles, or any barriers in or near them. Tunnels and bridges.
5) Approaching and passing under overpasses and large highway signs.
6) Curves that are more than just a gentle curve.
7) Passing exits. I usually want to stay on the highway but sometimes the car doesn't
8) Approaching stopped or slowed traffic.
9) Using automatic lane changing. Sometimes AP thinks a car two lanes over is in the lane I'm trying to change into and decides to slow way down.
10) Vehicles cutting you off or merging in heavy traffic - AP may not notice them in time.
Totally agree with this list, speaking for 17.24.28. To add one, cars further ahead in a different lane in a turn can cause speed reductions.Yeah, right now I have a mental list of things that are dicey (HW2). This is mainly for highway driving, since I don't use AP on local roads. I'm always on alert for these conditions:
1) Anything on the road that could be interpreted as unclear or conflicting lane boundaries. Doesn't actually have to be an actual lane mark. Could be a shadow (think, power line shadow along a road). Could be something spilled on the road.
2) Double yellow lines. Not sure why, but they are sometimes weakly detected as a lane marking even when the marks themselves are good.
3) Trucks or vehicles with a large profile - being behind, passing, or being passed by.
4) Narrow lanes, lanes that split or merge, lanes with cones, poles, or any barriers in or near them. Tunnels and bridges.
5) Approaching and passing under overpasses and large highway signs.
6) Curves that are more than just a gentle curve.
7) Passing exits. I usually want to stay on the highway but sometimes the car doesn't
8) Approaching stopped or slowed traffic.
9) Using automatic lane changing. Sometimes AP thinks a car two lanes over is in the lane I'm trying to change into and decides to slow way down.
10) Vehicles cutting you off or merging in heavy traffic - AP may not notice them in time.
To elaborate on 2), I also gets trouble when you have a wide yellow strip ,with or without rumble, next to the asphalt edge and fence. Car struggles with the 2 or maybe 3 different "lanes".
Totally agree with your list. Will add two, 11) disturbance from trafic ahead in other lane, in bends. 12) if in a trafic jam at high speed highway, lane markings has so much distance that it will have troubles stayin in lane in 5-15 km/h.Yeah, right now I have a mental list of things that are dicey (HW2). This is mainly for highway driving, since I don't use AP on local roads. I'm always on alert for these conditions:
1) Anything on the road that could be interpreted as unclear or conflicting lane boundaries. Doesn't actually have to be an actual lane mark. Could be a shadow (think, power line shadow along a road). Could be something spilled on the road.
2) Double yellow lines. Not sure why, but they are sometimes weakly detected as a lane marking even when the marks themselves are good.
3) Trucks or vehicles with a large profile - being behind, passing, or being passed by.
4) Narrow lanes, lanes that split or merge, lanes with cones, poles, or any barriers in or near them. Tunnels and bridges.
5) Approaching and passing under overpasses and large highway signs.
6) Curves that are more than just a gentle curve.
7) Passing exits. I usually want to stay on the highway but sometimes the car doesn't
8) Approaching stopped or slowed traffic.
9) Using automatic lane changing. Sometimes AP thinks a car two lanes over is in the lane I'm trying to change into and decides to slow way down.
10) Vehicles cutting you off or merging in heavy traffic - AP may not notice them in time.
To elaborate on 2) my car also has troubles with huge yellow lines where it does not seem to find the proper edge to follow. Getting too close to the fence. See example (don't mind the burning car).
JSkrehot
Member
The problem, is they are going to drive the car on the exact same route multiple times so it can learn, before ever doing what they say they will. Much like the video that shows "what a Tesla sees," where it drives a guy to work and parks itself, navigating city streets, pedestrians, and the highway. They had to have driven that exact route hundreds of times before making that video, as no car mentioned on this forum has been able to duplicate that kind of precision to date.
I'm not surprised . AP works well on highway driving and well delineated main streets. However it fails every time I drive around traffic circle in my community. And last two weeks it seems to move along extreme left of lane on occasion which seems inappropriate and had not happened before. I trust the AP less and less in my tesla 70
Yeah, right now I have a mental list of things that are dicey (HW2). This is mainly for highway driving, since I don't use AP on local roads. I'm always on alert for these conditions:
1) Anything on the road that could be interpreted as unclear or conflicting lane boundaries. Doesn't actually have to be an actual lane mark. Could be a shadow (think, power line shadow along a road). Could be something spilled on the road.
2) Double yellow lines. Not sure why, but they are sometimes weakly detected as a lane marking even when the marks themselves are good.
3) Trucks or vehicles with a large profile - being behind, passing, or being passed by.
4) Narrow lanes, lanes that split or merge, lanes with cones, poles, or any barriers in or near them. Tunnels and bridges.
5) Approaching and passing under overpasses and large highway signs.
6) Curves that are more than just a gentle curve.
7) Passing exits. I usually want to stay on the highway but sometimes the car doesn't
8) Approaching stopped or slowed traffic.
9) Using automatic lane changing. Sometimes AP thinks a car two lanes over is in the lane I'm trying to change into and decides to slow way down.
10) Vehicles cutting you off or merging in heavy traffic - AP may not notice them in time.
So basically everything?
Similar threads
