Gore point stripe faded: https://imgur.com/a/xHq57
Street View: https://www.google.com/maps/@37.6346515,-122.104109,3a,75y,8...
Street View: https://www.google.com/maps/@37.6346515,-122.104109,3a,75y,8...
Dirt, snow, rain, fog, dark, and other common conditions could obscure even correctly painted lines, and it just isn't realistic to expect perfect roads throughout such a large network anyway (in particular as human drivers can infer the line is meant to be there from the neighboring lines).
At its core, a system like Autopilot is meant to stand in for humans. Plus in this video where they reproduced the accident there was a full chevron pattern so the argument is flawed/wrong anyway: https://www.youtube.com/watch?v=6QCF8tVqM3I
But what amazes me is that Tesla autopilot apparently won't swerve or stop when it sees an unmoving object straight ahead. Whether it's a semi, a fire truck, or a gore point. But I gather that trying to detect that stuff generates too many false positives. And that overall, it would arguably cause more accidents net.
Edit: I have faced that ambiguity myself. When driving, perhaps foolishly, in fog so dense that I could only see the lines on the road.
It won't even brake unless the thing in front looks like the back end of a car or truck. Even then, if it's stationary, it may still get hit. It won't swerve, ever; it has no logic for that. Four times that we know of, a Tesla has plowed into a vehicle partly blocking a lane.
Putting something on the road which pretends to self-drive and can't detect solid obstacles is criminal.
But seriously, name "autopilot" aside, it's intended for divided highways. With high-quality road markings. And no stop lights.
And to be fair, airplane autopilots handle flying and landing. They don't handle driving between runways and gates.
Tesla's self-driving vaporware being sold since Oct'16 is camera only and should be activated any day now ?
https://twitter.com/elonmusk/status/823632597284691969 " Yes, safety should improve significantly due to autonomy features, even if regs disallow no driver present 23 Jan 2017"
3 months maybe, 6 months definitely 23 Jan 2017
Tesla also uses radar and ultrasonic. Lidar isn’t necessary, just makes depth measurement easier. It’s lower res than cameras, low range, power-inefficient, conflicts with other lidar. Humans drive fine in sunlight, without a coherent IR scanning beam.
Do you have any public source detailing this or do you just make this claim based on your own assumptions?
(Disclaimer: I work at Google but have no idea how our self-driving cars work.)
Doesn't sound like the cars blindly trust potentially stale maps.
My advice to the company BOD & shareholders: have him take the Udacity's self driving car online course and after the first project, he'll clearly understand how limited detecting lanes by CV only can be. To Uber's ex-CEO's point: LIDAR is the SAUCE.
[1] http://www.nsc.org/NewsDocuments/2017/Fatality-estimates-Jun...
And that‘s not even taking into account that we actually understand what we see and can reason about unexpected input and react accordingly.
If it's already fully automated, then why is it not for sale, yet?
https://www.theverge.com/2018/4/2/17189764/waymo-honda-self-...
https://www.theverge.com/2018/3/27/17165992/waymo-jaguar-i-p...
Considering Tesla's must have driven through there hundreds of thousands of times, they should have data to make sure the car at least follows the highway. I don't crash in to a barrier when a paint stripe gets scuffed in between my Monday morning commute and my Tuesday morning commute.
So should this cars follow some line blindly?
I think that this tech should stil be called a drive assist, and assist the driver, Tesla rushed it in enabling the autopilot thing, I am sure if you compare the stats with similar cars(expensive,safe and with drive assist) the stats will be against autopilot, I do not understand why they had to rush the autopilot and not keep it as a driver assist, to warn you and prevent some crashes)though it seems it can't protect from frontal collisions with static objects)
Hopefully when we start removing the driver from cars with autopilot, they will be following some sort of under-road NFC-like guidance system rather than painted lines.
Actually derailing a train takes serious effort; but people still try constantly. Derailing a Tesla seems somewhat easier, so I guess we’ll see.
This self driving cars can be tricked by bad conditions or some optical illusions, like you have a t-shirt with some pattern on it and the Tesla car will coma and hit you
It's even worse: wait until someone does something malicious to an autopilot car to cause it to crash/kill.
I hate to be so cynical, but if you're developing this stuff, you have to imagine worst case scenarios. And there are much much worse scenarios when malice is involved.
I've been had by local traffic cops in $COUNTRY, just a few km after crossing the border: two lanes in one direction, separated by a solid line ("no changing lanes"), running parallel for about a km. Left lane suddenly ends - asphalt gradually narrows from the left until there's only the right lane left. Your options are a) crossing the solid line into the right lane, and...wait, that's all, there's no way to go off the road. And guess what, traffic cops conveniently positioned right there: "you have crossed the solid line, that's a ticket and XYZ in local currency" "but there's nothing else I could have done!" "we don't care, you crossed the solid line!" Repeat until ticket is paid - note that this is obviously targetted at people who are not likely to fight the ticket (and presumably a mistaken/illegal road marking) in local traffic court.
I could imagine a very similar scenario set up to trip up SDVs (and fleece their presumably affluent occupants). The problem comes when con men underestimate the risks, and accidentally kill people.
The same suggestion as before, autopilot needs to be turned back into learning only. It never should have left. There were way too many videos just on youtube showing how bad it could be. It should simply be recording what it would decide to do versus what the driver did and flag every exception. It certainly can see barriers and its routines could save every instance where it might have hit something.
The NHTSA really should demand that Autopilot features be disabled. There are even cases of it failing to park without hitting objects.
Musk should be lauded for his EV push, it opens doors to all sorts of new technologies being incorporated in cars and will one remove the need for gasoline/diesel as the primary drive in all cars. However he needs to learn to step back on technologies that are clear cutting edge and not ready for prime time.
At this time features like "AP" should be used not to take over for the driver but instead assist the driver by keeping them from being unsafe. This means doing similar to what blind spot monitors to and lane drift systems work on.
(i) That there is an obstacle fast approaching.
(ii) That all cars in front of it is doing something else -- time to slow down.
But why can't it?
If you think about a sharp left turn with a wall on the right side of the road, there is a "stationary object" (the wall) in front of the vehicle throughout the turn that will be detected by radar and cameras. In order to navigate a situation like this, an autonomous vehicle has to have additional decision-making in place to effectively override "stationary object in front of me = stop" logic. But once you do that, it gets tricky to determine when you should ignore the "stationary object" and when you shouldn't.
On top of this, Tesla opted not to include LIDAR as a sensor in its vehicles, so it does not get a 3D point cloud to help identify what an object is and what to do with it. Instead, there's a high reliance on the cameras and radar, but these have some limitations. For example, it's difficult for a camera to distinguish between two objects of similar brightness and color, so for example if there's a white semi-trailer in front of a white wall 100ft away, the camera might not "see" the trailer. This can create situations where sensors are reporting conflicting data, which is bad - for example in the trailer situation, the camera would be saying "there's nothing in front of me until a wall 100ft away", but the radar would be saying "no there's something right in front of me". Or the radar might not "see" an obstacle that is a few feet off the ground while the camera is saying "there's something in front of me". Resolving these conflicts is hard.
That's not tricky. You never ignore it. A stationary object means that you will have to either stop or turn. You don't know which one yet, but you definitely know you'll have to take an action. If you haven't decided by some close point, you just pick one.
There's no reason to ever ignore it.
Tesla seems to have figured that, It is better to kill some of your customers then pissing every single one of them off...
The dichotomy between ignore and stop is a false one.
Combine that with the normal event of a garage wall or building directly in front of you while parking and we have three different "stationary object in front" situations that all have to be handled differently. In all three, the radar and camera are saying "stationary object ahead". But in the left-turn-with-wall situation you need to turn instead of stop. In the following-another-car situation you do nothing, and in the parking situation you need to stop instead of turn. And that's all without even starting to account for things like rain (opaque to some LIDAR and cameras = "we're literally crashing right now!" while radar reports nothing amiss). In order to decide what to do you have to pull in additional environment information, but then that's going to have similar caveats and conflicting reports to account for, so you need additional rules and information, etc. and now you've snowballed into a highly complex model with corner cases and potential bugs everywhere.
If it was easy we would have had self-driving cars a decade ago.
Vehicles in front moving at the same speed aren't stationary in the external reference frame. They're stationary in the vehicle's reference frame. If one stops, though, it becomes stationary in the external reference frame, and has closing velocity in the vehicle's reference frame.
Sorry to be pedantic.
> Combine that with the normal event of a garage wall or building directly in front of you while parking and we have three different "stationary object in front" situations that all have to be handled differently. In all three, the radar and camera are saying "stationary object ahead". But in the left-turn-with-wall situation you need to turn instead of stop. In the following-another-car situation you do nothing, and in the parking situation you need to stop instead of turn.
If the object is stationary, you turn-or-stop. If it's another car, it's really easy to measure that it's not stationary. If your sensors can't figure out the distance or velocity for multiple seconds in a row, your equipment should not be allowed to drive.
> highly complex model with corner cases and potential bugs everywhere
That's all in object detection, though. And none of that noise looks at all like a solid object at a specific location.
> If it was easy we would have had self-driving cars a decade ago.
I disagree. Even if you have a system that competently avoids stationary-object collisions, you're nowhere near a full self-driving car.
And while the overall problem of avoiding collisions has hard parts, none of the hard parts are in the "What do we do about a stationary object?" logic.
I feel like we're just going in circles here. I could keep giving you examples of cases where sensors will report conflicting information that makes decisions like turn or stop difficult or false-positive prone, and you'll keep insisting that it's just object recognition and you should always know to turn or stop. I'll stop here and stand by my original point - sensor limitations and immature algorithms are the reason for issues like the ones Autopilot has; completely preventing these issues is not straight forward.
You said the camera/radar would be reading another car as "stationary", which is a velocity number of zero. Add/subtract your speed and you get the real speed.
A non-defective sensor suite is going to give you either distance or velocity, and you can use distance measurements to calculate velocity really easily.
> A radar return from a rear bumper at relative zero miles an hour is exactly the same whether the car is actually in motion or not.
Doppler effect.
> "stationary" is always relative to something
Once you have a velocity number, no matter what it's relative to, it's trivial to convert it to any reference frame you want.
> I feel like we're just going in circles here. I could keep giving you examples of cases where sensors will report conflicting information
That's because such things are irrelevant to my argument. I'm only talking about how algorithms should handle objects that have already been found.
> I'll stop here and stand by my original point - sensor limitations and immature algorithms are the reason for issues like the ones Autopilot has; completely preventing these issues is not straight forward.
I fundamentally disagree that the navigation algorithms are difficult here. Sensor issues are huge, but your original scenario was based on the sensors working and navigation failing. That specific kind of navigation failure is utterly inexcusable. It is actually easy to force a car to either slow down or turn to avoid a known obstacle. That specific part should never ever ever fail. It doesn't matter how hard other parts are.
I'm quite familiar with how vector math works. Your comments above seem to still be missing that I am talking about situations from the perspective of individual sensors, which have no ability to make judgments about different frames of reference. That only applies at the level of the full driving system. For example, to a radar sensor alone, there is no difference between a car parked 10ft in front of you while you are motionless relative to the ground and a car 10ft in front of you while you are both moving at 60mph relative to the ground. Both are "stationary" as far as the return signature is concerned. There will be no doppler shift in either case. The driving system has to combine this with other sources of information for it to be useful, and that's where problems creep in.
> That's because such things are irrelevant to my argument. I'm only talking about how algorithms should handle objects that have already been found.
We're arguing two perspectives of the same higher-level opinion (algorithms are insufficiently developed to be safe enough for autonomous driving). What I'm trying to say is that there are a number of fuzzy steps to get from sensor readings to actual object detected, and then from actual object detected to "known obstacle" classification, as you put it. I don't think I'm going to be able to argue this case clearly enough in the comments here, so I'm going to add this to my longer article writing list. Thank you for the constructive debate on this.
Saying one very very specific thing is easy is not the same as calling engineers idiots. It's bad to conflate "the system is hard" with "every single piece of the system is hard"
> The driving system has to combine this with other sources of information for it to be useful, and that's where problems creep in.
There are many places where integrating information can let errors creep in.
The car adding its own ground velocity in is just... not one of them.
> What I'm trying to say is that there are a number of fuzzy steps to get from sensor readings to actual object detected, and then from actual object detected to "known obstacle" classification, as you put it.
Agreed. But while some steps are fuzzy, some are easy. The chain from start to finish is fragile and difficult. But some of the individual links are rock-solid.
> I'm going to add this to my longer article writing list. Thank you for the constructive debate on this.
I'm not sure if we really got anywhere productive here, but good luck!
The hardware that may be actually required to make this work well may still cost tens of thousands of dollars, which means car makers will avoid using it and will use cheaper options like cameras and radar.
While on the software side, we may be 80% of the road there, but the other 20% and solving every single edge case out of potentially millions will take another decade or more.
And that's without even mentioning software bugs that could make the whole system crash. I think someone mentioned recently that there are on average 15 bugs every 1,000 lines or code or something. Either way, we should expect a lot of bugs to be in this software. Remember Toyota's "spaghetti code"? And now we're just going to assume that Toyota's self-driving cars are going to drive us safely within 3 years?
And of course nobody wants to talk about how these cars will be hacked remotely.
On Teslas theoretically the camera is there to prevent stuff like this but vision AI isn't all there yet.
Other companies use lidar which gives pretty accurate distance and direction but not object speed. Pretty useful for avoiding obstacles but current automotive grade lidars are very expensive. Much cheaper solid state lidars seem to be imminent.
(iv) That it has (literally) been down this road before and had to revise its estimate of what it thought it was a lane (i.e. jerk back into the lane once it realized).
(v) That other Teslas have been down this road and learned that.
(vi) That the lane is suddenly getting wider than a freeway lane should, and that doesn't normally happen.
(vii) That the paint is worn, which indicates poor road maintenance, which necessitates caution.
(ix) That it saw a freeway exit sign a short while back, which logically indicates there must be a V-shaped spot ahead somewhere that it shouldn't drive into, and this could be it.
I love you qualified this ;)
Assuming no exit-only lanes, there is a V-shaped area you can drive into: the beginning of the new exit lane. And then it's followed by another V-shaped area that you can't: the resumption of the shoulder.
One way you can tell the difference is lines, if they are painted properly. But you can also tell from context. If you know (from experience and/or road signs) that only one lane peels off for this exit, and if you have already seen the V-shaped area for the exit, then the next V-shaped area is not OK to drive into.
There have been at least two other video near collisions of people driving on that section - it seems dangerous.
This reminds me when I was living in the Midwest and there was this one spot where I had to do a u turn in my commute which was very dangerous because of some tree branches obstructing view of oncoming traffic. One day I almost got into an accident and since it shook me quite a lot, I wrote an email to the city that the tree branch needed to be trimmed so I could see better. Sure enough I got an email back a few days later from a woman saying they “understand my frustrations” or something like that. I thought I did my part and went on with my day. Fast forward about a month, and I almost get in an accident at the same spot, man I was fucking furious that my communication with city was just “ohh we understand” —- instead of employing a sea of men and women telling me that they “understand” why could they have just not just trimmed the tree and improved safety.
If I was still living there I think I would have just gone with my chainsaw and done it myself.
Seems to work for Portland Anarchist Road Care: https://www.citylab.com/equity/2017/03/portland-anarchists-w...
The sticker, was made to Caltrans and DOT specifications, and affixed in broad daylight. Since the initial installation, Caltrans replaced the sign, but now adds the I-5 signage themselves.
Guerrilla public service can work people. ;)
Degraded paint does not a bad road make. Roads are optimized for human drivers that can leverage additional context to make their decisions. Many countries don't even bother to stripe their roads at all.
Now imagine trying to teach a car these things!
Now consider how you may feel about a car which has decided based on all available data that this strip of road is a lane. Its crystal clear to that car until its suddenly not.
You reacted with what appeared to me to be very restrained aggression (typical and acceptable, no offense intended) as you struggled to figure out if they were just being dense on purpose or were actually ignorant. I dont feel either were accurate but your message conveyed that was what you felt to me.
A car simply says “hey i dont know what im doing anymore. Help!”
Chicago even has this codified in the traffic code; while the rest of the state has "improper lane usage" as a violation, Chicago has "failure to maintain a lane".
Even on roads with painted markings they get scraped off by snowplows almost every winter. I'm reasonably sure that's what happened to that gore.
Los Angeles has a very interesting set of problems. Chief among them is the repercussions from closing roads/highways for repairs. It’s often times better, I think, to let them suffer in disrepair than it is to close them for any length of time. With semi-autonomous cars, however, I wonder if there will be greater pressure to keep them in repair.
Wealthy individuals/corporations but poor roads (and other government infrastructure/services) isn’t unexpected, it’s more or less the goal.
California had a lot of fires it had to fight last year. Where did that budget come from?
Streetview:
https://www.google.com/maps/@37.4112791,-122.076645,3a,75y,9...
March '18 was at 101/85
More faded than the last StreetView pics.
I also wonder if the line -- where the asphalt meets concrete -- contributed. Maybe it looked like the shoulder. Granted, there is also a yellow line.
I havent heard of GM killing anyone with their car.
At the very least, the system should slow down significantly and warn the driver if there's not high confidence in where it should be going.