Self Driving Uber Fatality - Thread I 17

[drawoh]

San Francisco Chronicle

As noted in the article, this was inevitable. We do not yet know the cause. It raises questions.

It is claimed that 95% of accidents are caused by driver error. Are accidents spread fairly evenly across the driver community, or are a few drivers responsible for most accidents? If the latter is true, it creates the possibility that there is a large group of human drivers who are better than a robot can ever be. If you see a pedestrian or cyclist moving erratically along the side of your road, do you slow to pass them? I am very cautious when I pass a stopped bus because I cannot see what is going on in front. We can see patterns, and anticipate outcomes.

Are we all going to have to be taught how to behave when approached by a robot car. Bright clothing at night helps human drivers. Perhaps tiny retro-reflectors sewn to our clothing will help robot LiDARs see us. Can we add codes to erratic, unpredictable things like children andd pets? Pedestrians and bicycles eliminate any possibility that the robots can operate on their own right of way.

Who is responsible if the robot car you are in causes a serious accident? If the robot car manufacturer is responsible, you will not be permitted to own or maintain the car. This is a very different eco-system from what we have now, which is not necessarily a bad thing. Personal automobiles spend about 95% (quick guesstimate on my part) parked. This is not a good use of thousands of dollars of capital.

--
JHG

 

[SnTMan]

Well, somebody had to be first

The problem with sloppy work is that the supply FAR EXCEEDS the demand

 

[GregLocock]

"Are accidents spread fairly evenly across the driver community, or are a few drivers responsible for most accidents?"

Accident rates per mile driven are biased highly towards new drivers (the stats are complex), young drivers, and old drivers. Men are about 50% more likely to crash than women.

"Perhaps tiny retro-reflectors sewn to our clothing will help robot LiDARs see us."

Perhaps, a more promising path is to detect your cell phone.

"Who is responsible if the robot car you are in causes a serious accident? If the robot car manufacturer is responsible, you will not be permitted to own or maintain the car. "

All the main manufacturers for L4 cars have announced they are liable and will self insure, and that current legislation is adequate. I suspect your second sentence is wrong in detail (somebody somewhere will buy an L4 AV) but going in the right direction.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[3DDave]

Without regard to this crash, there are cases where people try to beat locomotives around crossing arms. There are cases where the driver has limited options. Unlike most accidents, autonomous vehicles are most likely to gather all critical information leading to the crash.

 

[BrianPetersen]

The graph that Greg posted doesn't even attempt to consider other factors. Within drivers of the same age group, there are some that are highly skilled and others who are scatterbrained and un-co-ordinated, and others that are risk-takers (some intentional, others just unaware of their surroundings). Some are courteous, others are not. There's probably still an order of magnitude between the best and worst even within the same age group.

We all know someone that we don't want to be in the same vehicle with ...

I suspect that automated drivers may be better than the worst drivers, possibly even at today's technology level, but are not even remotely close to the best drivers and may never be.

 

[jayrod12]

I heard, but don't have backup evidence, that the person killed stepped onto a 75 km/hr road from a median and not at a crosswalk. Self-driving car, or human-operated may not have mattered.

 

[BrianPetersen]

That's what the initial reports are saying. It also appears that both the victim, and the person that Uber had hired to operate the vehicle, had backgrounds that were ... interesting.

 

[LionelHutz]

The police comment saying the video made it clear the accident would have been hard to avoid in any kind of mode is rather false for the autonomous mode. Darkness should have little bearing on the car's ability to have detected her. A self driving car could conceivably drive in complete darkness so what a video camera saw really means little on what the car can detect and be capable of reacting to.

To me, it seems that these autonomous cars do fairly well when reacting to something that is going to collide with them or they are going to collide into. But, they seem to be missing some or all of the "this could become dangerous so I should proceed with caution" programming. You could call it lacking prudence maybe? From what the reports have said so far, it sounds like the car decided to proceed at full speed close enough to this woman that it could not react when she changed paths towards it. The Tesla death demonstrated this as one of the faults too, trying to shoot the space between the truck and trailer wheels at full speed.

The car should have slowed and/or moved further away from the woman, even if she was proceeding in a direction that did not indicate she was going to cross into the path of the car.

 

[SnTMan]

Self-driving technology may never be better than the best drivers, but...

It won't get tired, bored, distracted, angry, aggressive, in a hurry, or any of the hundred other things ordinary drivers do. Probably already better than, say, 50% of drivers out there now.

The problem with sloppy work is that the supply FAR EXCEEDS the demand

 

[GregLocock]

Yes, Brian but posting an analysis that shows how good good drivers are and how bad the bad drivers are doesn't really help, I think. Here's the best I could find, I don't know what the source data is, or even what is really being plotted, probably crashes of any type in the last year on the x axis, and the proportion of drivers on the y axis.

The Luck curve is if you just take the average crash rate (20.3%) and do the usual probability on it. Not a whole lot different to the underlying data.

Obviously many meat drivers never have an (injury/fatality) accident in their half a million miles of driving in their lifetime. So accident free meat drivers are a hard target to beat, since they have a perfect record.

So I'd rather look at averages.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[GregLocock]

https://search.proquest.com/openvie...ae065/1?pq-origsite=gscholar&cbl=18750&diss=y

also has some interesting stuff

I like this graph. What it is saying is that if you have had 0 or 1 crashes in the previous 6 years, there is a 5% chance of having a crash in the next year. If you've had 5 crashes in the previous 6 years, there's a 25% chance you'll have another crash in the next year.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[azcats]

The police comment saying the video made it clear the accident would have been hard to avoid in any kind of mode is rather false for the autonomous mode. Darkness should have little bearing on the car's ability to have detected her. A self driving car could conceivably drive in complete darkness so what a video camera saw really means little on what the car can detect and be capable of reacting to.

I know the area in question and it's quite possible that she emerged from the median from behind brush or trees leaving little or no time for a driver (human or not) to react.

Link

 

[Archie264]

I'd ride a bicycle in a hailstorm before I'd ever get in of those things. Can you imagine? Haven't paid your property taxes? The doors would lock and straight to City Hall you'd go. Say something politically incorrect inside the cab? It'd Mary Jo Kopechne you off the nearest bridge. No thanks.

 

[BrianPetersen]

Further to Lionel's statement, I think the autonomous systems are lacking in situational awareness in general.

Hmmm. The lane to my right is ending (or is obstructed up ahead). I should allow for vehicles in that lane to merge with the lane that I'm in, by matching speed with them and aligning myself with a space between those vehicles so that the merge can be done without conflict.

Hmmm. I'm approaching a traffic signal. It just turned yellow. My rear view mirror is filled with dump truck. I'm going through this intersection even if it's a wee smidge into the red by the time I get there.

 

[Tomfh]

a median and not at a crosswalk

I've heard this defense several times and I really really don't like it. Yes if the person jumped out in front of the car from behind a bush that's one thing, but this suggestion that autonomous cars can't be expected to stop for pedestrians who aren't at cross-walks does the case for autonomous vehicles no favours.

 

[drawoh]

Tomfh,

We do not understand the algorithms used with these autonomous cars. We anticipate that people will cross at crosswalks. We anticipate they will be less likely to cross elsewhere.

A big advantage of LiDAR is that it provides its own light source and ought to detect stuff even in complete darkness. I wonder just how rapidly LiDAR detects everything. I have over thirty years experience with LiDAR. It does a finite number of scans per second. Will it detect and make sense of an object moving somewhere other than along the anticipated track?

Dodging a LiDAR equipped vehicle is not the same thing as dodging a vehicle driven by a human.

I slow right down when I pass groups of pedestrians, especially children, when I drive down narrow lanes and in parking lots. How will a robot anticipate threats like this, and how well behaved will humans be when they are following this robot as it tries to interpret threats?



--
JHG

 

[SnTMan]

I slow right down when I pass groups of pedestrians, especially children, when I drive down narrow lanes and in parking lots.

Yeah, you do, I do, but plenty of people don't. Don't get me wrong, the tech has got a ways to go. It'll get better, human drivers will not. My experience is, the cars get better and better, the drivers get worse and worse.

The problem with sloppy work is that the supply FAR EXCEEDS the demand

 

[3DDave]

Vehicle lidar can have 64 lasers, with more recent models using 128 of them.

https://www.technologyreview.com/s/...ter-but-its-still-too-expensive-for-your-car/

 

[JStephen]

For those who are statistically minded-
There are a certain number of pedestrians killed each year.
And various autonomous cars get in a certain number of miles each year.
Just based on the miles, how many pedestrians would expect to have been killed by autonomous cars?

By the way, how do we know this was an accident, anyway? Isn't that car "autonomous"? It could very well have been intentional.

 

[LionelHutz]

Given that link to the location, comparing video from the scene and the Google street view shows the car was on Mill Ave travelling north.

The news reports say she was walking and pushing the bike, and if that is true then she wasn't travelling particularly fast.

The right side of the SUV has the damage so the woman most likely came from the right side into the path of the SUV. If she came from the median on the left then she crossed almost 4 traffic lanes before the SUV hit her which certainly goes against any claims that she suddenly stepped in front of the SUV.

It also appears the accident occurred across from the second leg of the X in the median just past the road drains in the right side curb. Since the ground is rough and hilly past the sidewalk in that area, I find it hard to believe the woman was travelling perpendicular to the roadway or outside the sidewalk before entering the roadway. That location also has a sidewalk and a bike lane between any vegetation and any excuse saying she jumped out from behind a bush doesn't hold much water.

My expectation is that she was travelling southbound on the sidewalk or in the bike lane and then turned to her right to cross to that X path going across the median. Crossing Mill Ave completely and heading into the parking lots or below the overpass on the west side would be as likely a general direction of travel as any.

An alert and aware driver would know the path of someone coming southbound on that sidewalk might turn towards that path through the median.

How much foot and bike traffic is in that area? Would regular drivers of that section of roadway specifically watch for bikes or people on foot crossing at the ends of those X paths?

On another note, that X path "to nowhere" through the median appears to be one dumb road feature. It has no useful function except to dangerously tempt people to shortcut across Mill Ave.