I understand that emergency maneuver system was disabled so the car did not brake between t minus 1.3 and t. But why didn't it brake from t minus 6 to t minus 1.3? Looks like it detected that the car's and object's paths were converging, so why didn't it brake during that interval?
Based on a TED talk by someone from Google[1]. I think having the car apply the brakes when there’s a possible disturbance causes the car to apply the breaks too much and makes for a really uncomfortable ride.
I think a major effort in self driving is solving the Goldilocks issue of reacting properly to impending accidents, but also not apply uncomfortable breaking if it’s not needed.
Seems like it was too insensitive at that distance.
There's the third option of slowing down. That's what most human drivers do subconsciously when we see something that we're having trouble identify, and feel it could turn into an obstacle.
This too. An electric car would simple decelerate by letting off the gas pedal and having the regen kick in. On a gas car this would be equivalent to a partial braking.
Shouldn't the brake application be not boolean but 0-100% strength based on confidence levels?
>I think a major effort in self driving is solving the Goldilocks issue of reacting properly to impending accidents, but also not apply uncomfortable breaking if it’s not needed.
This is also an issue with current production automatic braking system. One that was largely solved with on track testing, and on road testing logging false triggers with a driver driving. There's no need to risk lives unless you're just cutting corners to avoid the cost of a test track.
Because then the car would break every time anyone approached the car from an angle (which is constantly). Think every intersection ever, every time driving near a sidewalk ever. The car would be herky/jerky as crap.
They should have had it set to spike the breaks once collision was imminent though, that's (maybe) the biggest programming omission here.
They should have set it to slow down gradually when approached from an angle at T-6 and then speed up once past the intersection risk, so that when the scenario emerged at T-1.6 it could emergency stop safely.
> Think every intersection ever, every time driving near a sidewalk ever. The car would be herky/jerky as crap.
I'm not sure that'd be a huge issue. The vectors have to be intersecting first of all, which most vectors emanating from sidewalks wouldn't be, and then a little hysteresis would smooth out most of the rest.
I don't know if you've been to New York or any other places where people walk, but vectors would absolutely be intersecting on a regular basis up until a fairly short time when the pedestrian would stop. Constantly I walk toward an intersection where, if I kept going for three more seconds, I would be pasted to the street by a passing car. But I stop at the end of the sidewalk, before the road begins, so the vector changes to zero in those last three seconds. It would be super weird if cars would brake at the intersection every time this happened. Cars would be braking at every major street on every major avenue, constantly.
What's actually needed here is some notion of whether the pedestrian is paying attention and will correctly stop and not intersect the path of the car. Humans are constantly making that assessment based on sometimes very subtle cues (is the person looking at/talking on a phone, or are they paying attention, for example).
Yeah, eye contact is a very important signal. Maybe there needs to be some specialized hardware to detect eyes and determine the direction they're looking in.
We use eye contact because we can't infer what another person is thinking and we can't react quickly enough to their actual movements at car speeds. This latter isn't the case with automated vehicles, so eye contact shouldn't be necessary, as long as you get the vector algorithms right.
> Constantly I walk toward an intersection where, if I kept going for three more seconds, I would be pasted to the street by a passing car.
These autonomous systems are evaluating surrounding vectors every few milliseconds. A timescale of 3 seconds simply isn't important, as they would instantly detect you slowing down and conclude that you wouldn't intersect with their vector.
> But why didn't it brake from t minus 6 to t minus 1.3? Looks like it detected that the car's and object's paths were converging, so why didn't it brake during that interval?
You're missing the context of this thread. The software in the Uber car has a clear failure condition. That has nothing to do with whether it's possible to infer such vector collisions without jerky driving, which is the point I'm addressing.
The question was why the car doesn’t brake early. “Because scanning every few milliseconds” is not an answer. Scanning frequency is irrelevant to the fact that emergency braking is not a reasonable strategy in general.
Safe driving often does require slowing down in the face of insufficient information. If a human driver sees an inattentive pedestrian about to intersect traffic, they will slow down. “Drive until collision is unavoidable” is a failing strategy.
And anyway, jerky driving is a symptom of late braking, not early braking.
> And anyway, jerky driving is a symptom of late braking, not early braking.
I see it as more than just jerkyness, I see a massive safety issue in traffic. If your autonomous car is slamming on the brakes spontaneously there's a lot more opportunities for other drivers to plow into you from behind.
They can't detect me slowing down before I start slowing down. So if it's t-4 until impact and I'm still moving at full speed, they would need to start braking now if they can't stop in 4s (assuming the worst case that I continue on my current trajectory).
That being said, I'm happy to find my assumptions about stopping time are incorrect and a car traveling at 25mph can stop in less than a second. So on busy NYC streets this wouldn't be an issue. Even at 50mph it appears that stopping time is sub 3s, so the vehicle could probably have avoided this collision if it were running a more intelligent program.
> They can't detect me slowing down before I start slowing down. So if it's t-4 until impact and I'm still moving at full speed, they would need to start braking now if they can't stop in 4s (assuming the worst case that I continue on my current trajectory).
Right, collision is basic physics accounting for the stopping time and distance of pedestrians and cars. So the question is whether pedestrians on sidewalks really have so many collision vectors with traffic such that autonomous vehicles would be jerky all of the time as the initial poster suggested.
I claim reasonable defaults that would far outperform humans on average wouldn't have that property. Autonomous vehicles should be programmed to follow the rules of the road with reasonable provisions to avoid collisions when possible.
I think the situation might be all the transitions. All the time people on bikes switch from a driveway to a bike lane, during which they could continue straight into the road. Or people step out of a building and walk diagonally across a large sidewalk, they could keep going straight into the road.
Which simply means it is because the car's AI isn't good enough to classify that object as something it should slow down for versus something it can ignore (like an empty plastic bag drifting across the road.)
Well, it is good enough, it just that it develops that confidence over the period of so many seconds. In this case it took until T minus 1.6 seconds to realize "ok this is something we should stop for".
I don’t know their internals but from that report it looks like their recognition system is probabilistic and routinely hops back and forth between “car collision RED ALERT!” and “lol there’s no problem”. If it were to randomly slam on its breaks every other second then it would cause all kinds of other accidents.
Sensors were fine, victim was detected, software was crappy.
That's vanilla testing edge-case stuff, really, and it's known that uber are unter when it comes to this, but the removal of all the useful safety layers after that (braking, alert, second human, hardware system) is reckless and stupid.
Well, exactly. Nothing wrong with the sensors, and the classifier was getting consistent pings. This is the critical failure that led to the crash, as much as the shitty final-second emergency non-process.
I observe that cars and other objects often come very close to each other, so it would seem impossible to simply brake based on "converging paths". It's necessary to know what an object is and how it's going to behave. If you don't, I don't see how you can go anywhere.
People slow down from 35 to 30 or whatever. Humans don't slow down to the point at which an accident is physically impossible for all unexpected movements, because that would be zero considering there are frequently objects within feet or inches.
Self driving cars can emulate humans, but that won't bring them to human level performance without the ability to model other actors. If they try to mathematically rule out the possibility of accidents without such models, they won't be able to go anywhere.
That's not how traffic works at the moment. I often have pedestrians walk towards the street and I assume they will stop so I don't slow down unless they are children or similar. Almost every day I could hit pedestrians if they kept walking.
You should slow down. Those people you are describing sound like they want to cross the street, and they probably have right of way, so yield for them.
Also, my two year old will sometimes walk towards the curb, but she is very good with streets, so I am not worried. She always stops and waits to hold someone's hand before crossing. This behavior freaks some drivers out, causing them to slow or come to a complete stop, which is the nicest outcome because then I can take her hand and cross the street. When I am walking by myself drivers rarely yield even as I am stepping into the street, even at marked crossings.
I guess my point is if my two year old exhibits the behavior you ascribe to a hypothetical non-child pedestrian then how can you be sure your hypothetical pedestrian won't just "keep walking"? What if they are blind, or drunk, or reckless? Perhaps you have been lucky before and never struck a pedestrian but I strongly urge you to assess your behavior. Stop for pedestrians, it's the nice thing to do and it's probably the law where you live.
You make me sound like a crazy driver :). Just watch the traffic along a busy road with pedestrians on the side. Nobody slows down if they pedestrians behave the usual way. You also often have pedestrians walk into the road to just stop right before the zone where cars are. Nobody slows down because they see that the pedestrian is observing traffic.
I have made this observation, I even mentioned it in my comment. I am urging you and anyone else who reads my words to assess your behavior so as to affect a change in the status quo. Sometimes pedestrians do walk into the road and they do get struck. The only foolproof way to prevent this is to change driver behavior which is why the requirement to yield to pedestrians at intersections is codified into most laws.
When I am driving I often see people standing at the curb just staring at their smartphone. Usually these people are wasting time because they don't expect traffic to stop. When I stop for them they are usually pleased, they cross the road and get on with their life. Sometimes these people are just waiting for an Uber or something, when I stop for them they get confused and look at me funny. I don't mind, I just smile at them and resume driving. I am in a car, so I can accelerate and travel very quickly with almost no effort. It is no trouble for me to spend a few seconds stopping for a false positive.
In the context of self-driving cars though, they can't read expressions and exhibit them. They can't necessarily even say whether something is a person or not. So your driving methods are not applicable. A computer can say "given the physics of how an average person can move, it is possible for them to leap in front of me in X amount of time" and then what? I think that a self-driving vehicle that follows your principles without your analytic ability is going to have so many false positives it will be useless. And I think the fact that they aren't attempting to follow your principles is evidence they don't have the ability.
I agree, and perhaps self-driving cars are not yet ready for "prime-time". The solution for the current state of the art might also just be maintaining a lower speed with automated drivers, which may also necessitate limiting the types of roadway on which they can operate. A slower average speed shouldn't be a big problem for automated cars since they don't experience the frustration of human drivers. Given wide-enough adoption, accomodations can be made to traffic signalling apparatus, car-to-car communication, and car-to-cloud integration to develop near-seamless traffic flows, allowing shorter travel times even at slower speeds. From here the tech could be iteratively improved to provide faster speeds without compromising safety.
This only works if you don't have other cars run into you when you slow down unexpectedly. I am not saying that current traffic is sane but just saying "always also down when you see pedestrians " just doesn't reflect reality. In CA you often have speed limits of 45 right next to houses with driveways. Either you play it safe and go 20 or less and get cursed at by other cars or you go way too fast to respond to unexpected obstacles.
When I am driving I routinely check my rearview mirror and assess the following distance of the cars behind me, so I usually know I will not be rear-ended when I am stopping for pedestrians or for any other reason. If I am driving and I notice someone is following too close for our speed I will tap the brake lights so as to encourage them to increase their following distance. If this fails I will slow down to a speed where their following distance becomes appropriate. If they are an uncommonly aggressive driver I might even pull over or change lanes and allow them to pass, I certainly don't want to be rear-ended! That said, even if I were to fail at this, I would prefer to be rear-ended stopping for a pedestrian that would have stopped than to strike a pedestrian who failed to stop walking into the path of my vehicle.
The speed limit is a reference to the maximum allowable speed of the roadway, not the minimum, only, or even recommended speed.
You clearly never have driven a busy four lane street in LA with bicycles and pedestrians mixed in. What you are saying makes sense in theory but nobody drives that way.
I do drive this way, most often in Seattle which has no shortage of the behaviors you're referencing. You can drive safely too because you are in control of your vehicle.
I didn't always drive like this, but I was in an accident that was my fault that totally upended my life, so I made an effort to change my ways. You can do it too, before you get in an accident that sets your life back, or irreparably shatters it...
I recommend taking an advanced drivers education course if you seriously decide you want to improve your driving. A lot of this stuff is covered.
Stopping unexpectedly does not cause accidents, locking down on your brakes unexpectedly causes accidents. In fact, in the US the person who hits you from behind will be held at fault no matter how hard you hit your brakes or why, because they are expected to maintain sufficient distance and attention to stop when you do.
If a light application of the brakes causes the car behind you to slam into you, the fault is with the idiot tailgating you, while playing with their phone.
Nobody's suggesting that anyone should slam the brakes every time a moving object intersects your vector of motion.
> I often have pedestrians walk towards the street and I assume they will stop so I don't slow down unless they are children or similar. Almost every day I could hit pedestrians if they kept walking.
With respect, I don't know what you meant to say but that sounds like a description of a bad (or at least inconsiderate) driver to me.
In any case, when I think about how I would design a self driving car, an "auto-auto", the first principle I came up with was that it should never travel so fast that it couldn't safely slow down or break to avoid a possible collision. This is the bedrock, foundational principal.
> I often have pedestrians walk towards the street and I assume they will stop so I don't slow down
> the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path
Little different, huh? If you see something that looks like it might be in your way, and you aren't sure what it is, you just keep going?
And if I see a pedestrian in the middle of the road at a random spot, especially at night, I'm slowing down since I don't know WTF they're thinking. Or if I'm in a neighborhood with regular street crossings carved out of the sidewalk and someone's coming up to one of those - I don't know how well they're paying attention to their surroundings.
I can tell pretty quickly if something is pedestrian or a bicycle and plan accordingly. In addition I can tell where the pedestrian is looking. Sometimes I slow down, sometimes I don't depending on how I assess the situation.
I think it comes down to the fact that the classification algorithms are not ready for primetime.
with significant fuzz factor, agreed. If i'm on the sidewalk and take a step toward the road, should it make the car jerk? Probably not, it's a hard call for passenger comfort. From another angle, think of subway tracks - the algo your describing would slow to a crawl as it crosses every station.
> From another angle, think of subway tracks - the algo your describing would slow to a crawl as it crosses every station.
As far as trains go, they do slow down when passing trough a track adjacent to a platform. There are some non-platform adjacent tracks the train companies use to avoid slowing down, however they will slow down or even stop if something is going on.
Similarly, high speed rail doesn't have level crossings due to safety considerations. Overall trains are very safe and they are _designed_ for safety. It is highly irresponsible and immoral to just wing it with people's life/safety.
> It is highly irresponsible and immoral to just wing it with people's life/safety
100% agree.
> As far as trains go, they do slow down when passing trough a track adjacent to a platform. There are some non-platform adjacent tracks the train companies use to avoid slowing down, however they will slow down or even stop if something is going on.
The equivalency isn't 'trains slow down through stations' (That would be cars having lower speedlimit in pedestrian areas - they do and the ubers honor) , it would be 'train spikes breaks if someone takes a step toward the edge' (Which they don't, even though it would potentially save lives).
There's always a tradeoff between usability and absolute safety. I'm not saying the uber did nothing wrong, at a minimum it should have spiked it's breaks. The 'perfect world' solution would be the uber knowing mass and momentum of approaching objects, and whether they could stop in time. But honestly, here would that have helped? We'll never get rid of people walking in front of moving cars, just have to be find the happy balance (which we clearly haven't)
> 'train spikes breaks if someone takes a step toward the edge' (Which they don't, even though it would potentially save lives).
A train's deceleration under maximum braking is far, far lower than a car. [1] suggests 1.2m/s² (paragraph 8).
[2] says the deceleration of a low-speed train crashing into the buffers at the end of the line in a station should not be more than 2.45m/s² (paragraph 35). That caused "minor injuries" to some passengers.
Trains do slow down earlier if the platform they are approaching is very crowded, but there's not really anything else they can do.
> If I'm on the sidewalk and take a step toward the road, should it make the car jerk?
No. That's my point. Take less drastic measures earlier, and only escalate when you have to. That's how I drive, and a self-driving car can do the same.
