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

Fatal Flaw in AP2 Design

This site may earn commission on affiliate links.
The diving for a lane is why I seldom use EAP on local roads. I've given up.
I come close to agreeing with you. Instead, I'm selective on when I use EAP on local roads as I want to provide as much information back to Tesla to help them improve their software. I keep engaged with my hands on the steering wheel getting a sense for the times that the Tesla goes stupid on me and take over control. I make a mental note of the problem it likely encountered and pay particular attention to it recurring. Just keep in mind that we're all Beta testers.
 
THIS! Thanks for pointing it out... Drives me crazy, whether it is a wall on my left, or a truck/car on my right, the car responds AFTER I already am alongside or even past that obstacle. I keep thinking in my mind "well, that was useless, but cool that I can see it is trying".

Yes, it must be predictive - video games do this well for video game AI. We need the same, but it will require cameras as you actually want to make that move before the car is in sensor range.

An example: in 17.17.4, the system moves the car away from trucks which are in the adjacent lane, if it can. But it does not plan to do this when overtaking a truck. It is not until the car is alongside the truck that the lane positioning changes. If the path prediction was nailed then the car would move into position within the lane before the nose of the car was level with the back of the truck.

Better path prediction would resolve the sensor glitches that cause the car to do weird things when cresting a hill, or when the road lines disappear temporarily.
 
  • Like
Reactions: J1mbo
"fatal flaw." Man with enough of these, we're just not going to have AP at all, watch. #ruinthingsforeverybody. It's been repeated many times that autopilot is a driver assistance aid, not a substitute for driving. Use it when appropriate, but not 24/7.

Im surprised cruise control still exists given the many dangers having a car stay at one speed indefinitely poses.
 
THIS! Thanks for pointing it out... Drives me crazy, whether it is a wall on my left, or a truck/car on my right, the car responds AFTER I already am alongside or even past that obstacle. I keep thinking in my mind "well, that was useless, but cool that I can see it is trying".

Yes, it must be predictive - video games do this well for video game AI. We need the same, but it will require cameras as you actually want to make that move before the car is in sensor range.

Yes. I noticed this from the early firmware, that side collision "blips" on the instrument cluster (the yellow reflections) happen about 3 seconds after the threat is already passed. If there is a low wall or a truck adjacent to you, it gets detected almost after you've already passed the object, and THEN the car veers off center in the lane as if to demonstrate it is thinking of your safety. Somehow these sensors seem to be biased facing rear rather than front.
 
Yes. I noticed this from the early firmware, that side collision "blips" on the instrument cluster (the yellow reflections) happen about 3 seconds after the threat is already passed. If there is a low wall or a truck adjacent to you, it gets detected almost after you've already passed the object, and THEN the car veers off center in the lane as if to demonstrate it is thinking of your safety. Somehow these sensors seem to be biased facing rear rather than front.

That's strange... I've been paying attention to those blips as I pass cars(after you mentioned them happening too late in a previous post) and haven't noticed any delay. The car definitely moves to the other side of the lane only when already next to the adjacent vehicle, but I haven't seen any noticeable delay in that detection. Along the same lines, I don't see the delay in the blind spot detection you reported before. On my car it tends to be proactive and shows up the whisker/blip before the car moves out of view of my mirrors.
 
  • Informative
Reactions: EinSV
My car does this to a less dangerous extent on a road with railroad tracks. The tracks are on an elevated short section and when the car's nose goes up and comes down it will pull right abruptly and then go back to the lane. It moves at most 2 feet and doesn't actually cross the lane lines but there could, possibly, be a pedestrian on the road near there so I've stopped using AS on that section. I am unsure this can be fixed other than, as you say, keep going straight but that's not always the right call.
 
I've been watching Tesla roll out their driving assist features and while they still claim it is in Beta status, it is obvious they are improving all the time and going in the right direction.

Everyone wants it to be perfect, but nothing is perfect in driving personal vehicles, including human drivers themselves.

You can be the best driver in the world, but when you get a piece of grit in your eye, you own concentration goes away too.

This auto driving stuff is getting better all the time, but no where near being able to the ultimate goal of telling your vehicle where to go, and it just gets you there in all circumstances, with no human intervention.

That is perhaps going to be available in the future, but not available right now.

I imagine that in the future, road designers will take into consideration the capabilities of auto driving vehicles when laying out road improvements. They will provide marking on the lanes that will be easier for the cars to follow. They will have indicators when exits are coming up, and design them for smooth flow by the auto piloted vehicles.

This combination of smart cars and smart roads might be the ultimate answer to automatous personal transportation.
 
  • Like
Reactions: RacerX and GSP
That's strange... I've been paying attention to those blips as I pass cars(after you mentioned them happening too late in a previous post) and haven't noticed any delay. The car definitely moves to the other side of the lane only when already next to the adjacent vehicle, but I haven't seen any noticeable delay in that detection. Along the same lines, I don't see the delay in the blind spot detection you reported before. On my car it tends to be proactive and shows up the whisker/blip before the car moves out of view of my mirrors.

I wonder if it's some sort of calibration issue with the sensors. Perhaps they don't come perfectly calibrated from the factory? I don't know what is considered "normal" behavior but your data point helps.
 
I've been watching Tesla roll out their driving assist features and while they still claim it is in Beta status, it is obvious they are improving all the time and going in the right direction.

Of course, there is no argument about that. I just wish Tesla hadn't claimed otherwise before I purchased my car, insisting the features would be ready by December 2016. I would have enjoyed a nicer ICE car (Audi R8 V10 Plus) in the meantime while they figured stuff out.
 
  • Like
Reactions: NerdUno
I think that Kerfluffle with Mobile Eye caused many of the current issues.

Tesla needed to go back to the drawing board and re-imagine their way forward.

Took a small step backwards in order to allow more giant steps forward in the future.

Have a friend with an R8. Loves the car, but it has it's own issues. He tends to drive his Model X much more often, and uses the R8 mostly for enthusiastic romps in the back roads of San Diego.
 
  • Informative
Reactions: oktane
That depends on how the "event" looks to the computer. If it thinks everything is A-OK and is just responding normally to the information it has, then no.
Right, and I understand that. What I guess I wasn't clear on was that I think that any time AS is manually aborted by a corrective steer, it should be flagged as not OK. When my car dives, I don't have to nudge AS off -- I have to reverse the veer. It is not insignificant. It should not be considered normal.

For example on a road near me with a 35mph speed limit and single lane: at a certain point the single lane becomes two as a new lane is introduced to the right. The line on the left hand side of the car is double-yellow and continuous. I can understand (but am disappointed) that the Tesla is not smart enough to continue to use the yellow lines and drive straight. But here is what happens EVERY TIME: the Tesla veers to the left OVER the yellow lines. It is in fact the only option that makes no sense, yet it is what the car does every time. WTF.
 
I don't think people people realize they are playing with tomorrow's technology -today-.

I picked up a 2017 Chevy Volt two months before the Model X and guess how I stay in between lines........ - I turn the wheel ever so slightly to the left or to the right depending on which the lines move. I *think* that's how I've done things for over 20 years and so far hasn't been *fatal*.

Until they remove the steering wheel completely, any *fatal flaw* is from poor driver judgment and nothing to do with the car.

AP is an amazing convenience and a toy WHERE APPROPRIATE.

*PROTECT YOURSELF AT ALL TIMES*
 
You'd think with so much life threatening stuff going on someone would have the presence of mind to post a few videos of it actually happening.

If it's a repeatable event, getting a video should be straightforward.
 
Maybe this example is why Cadillac made HD Lidar maps of all roads that SuperCruise is made to drive hands-free on.

We'll have to wait and see but Cadillac has enough confidence in the system to market it as 'hands free'.
That has more to do with Cadillac having a driver attention detection system that is independent of keeping hands on steering wheel, than supposed confidence in the system. The NHTSA will not allow L2 systems to operate without ensuring driver attention and almost all cars use hands on the steering wheel to do so.

We'll see how it operates in the real world when it comes out.
 
  • Helpful
Reactions: GSP
No one should be calling Tesla. That's akin to hard coding things into the machine learning for driving. The car/software knows it lost tracking, sees what you do to react, and as that information gets processed in the cloud will eventually be fixed. That's how the process works....not calling some human to physically write something in code. :rolls eyes:
In other words, do nothing? :shakes head:
 
  • Like
Reactions: GSP and NerdUno
I use AP a lot - probably 40 miles a day split evenly between highway and local use. AP1 users have posted a lot about how when their car crests a hill it will go berserk and turn the wheel. (That may have been fixed and I haven't read because it seems the AP1 and AP2 threads have largely diverged on TMC). I also noticed this on AP2. I have also noticed the lane "hunting" swerve when entering an intersection or some other situation where lane lines disappear. I live in a city, and there is a lot of traffic. That means I oftentimes have a lead car. When there is a lead car, no lane lines don't seem to matter too much because it follows the lead car.

That all being said, what the OP writes does happen and I agree that it is weird. Under what circumstances is the default behavior for losing the lane to turn the wheel? If you're in fog and can't see, do you just turn the wheel left? :eek: No, you slow down and proceed along. Eventually you stop if it gets really bad. In virtually every case of "Lane Panic" (that's trademarked by me now, FYI) the situation would resolve itself and the car would find a lane if it had just continued straight. Even more confusing to me is why if the car has truly lost its way, why don't you get the "Take control, we're going to die!" Armageddon Hands? It's always seemed weird to me that it works this way.

That being said, I'm sure it will be addressed. I personally don't put a lot of stock in the HD mapping technique, but I would imagine that at some level they'll use some mapping technique to at least nudge the car along at the correct heading. I think there are a lot of intersections that I get through that require I know which direction I want to end up going in the end. Still, I don't think the Lane Panic reaction is a basis of operation for the system that will permeate it forever. I think of it more as an edge case (albeit a kind of big one) that just hasn't been addressed yet.
 
  • Like
Reactions: hiroshiy