Self-driving Uber kills Arizona woman in first fatal crash involving pedestrian

Gene Wirchenko wrote:
On Wed, 21 Mar 2018 23:40:04 -0500, Char Jackson
wrote:

On Wed, 21 Mar 2018 16:26:09 -0700, Gene Wirchenko
wrote:

On Wed, 21 Mar 2018 14:03:22 -0400, Keith Nuttle
wrote:

[snip]

This will take years, and a lot of money which many governments
organizations do not have. Even then what will happen when the Live
speed sign is vandalized, hit by a car, or a deer is standing where the
car can not communicate with the sign
True. It is bad enough when snow covers the sign so it can not
be read. I have encountered this many times in British Columbia,
Canada.
And how did you deal with it, and why do you think the car won't deal
with it in a similar fashion?

I knew the route and was expecting the sign. The road was also
steepening and getting curvy. (In case it matters, coming from
Keremeos, BC to the junction of highways 3 and 97 south of Penticton.)

I can not know that the car will have been there before or
otherwise know about it.

Had I not been there before, I might have had an accident. So
could a self-driving car.

I've been in the same situation many times and I can't say that it's
ever been a serious problem. Signs can be covered by snow, obscured by
trees or other vehicles, or the sun can be directly behind the sign,
etc. We all deal with similar situations all the time. Software is being
adapted to do likewise.

We do. We can not rely on what we can not perceive so we take
other factors into account. It is not as safe -- after all, if it
were, the sign would not be necessary -- but we do what we can.

What will a self-driving car do?

Sincerely,

Gene Wirchenko

A self driving car used GPS and Google Maps.

It will have some ideas of the topography of the road
it will be driving on. It will continue to need to
synchronize the sensor data, with the map data
(for "best results"). So it "knows where it is".

It isn't a Roomba. No matter what isolated chunks of
web articles might make it look like. It doesn't bumble
along the road without a plan. It has to do route planning
to start with. It has to notice unaccounted-for obstacles
("Road Closure" sign) and re-plan the route and try again.
The GPS and maps won't be perfect. There will be tales of
driving half-way round the world to get somewhere, because
occasionally these route planning episodes will not work
right.

Just like that Apple Maps thing had a bumble or two at
launch.

The car is going to know that human drivers caused 12
accidents at the corner of 23rd and main. And it's going
to route-plan around that. Even if the screen doesn't
say why it's taking a detour, the car will have its
reasons.

The "reaction in real time" is needed to keep the
accident rate to zero. It needs to not run over pedestrians.
It has a route plan. Now, it just needs to drive there
without breaking anything.

The car needs enough sensors, so it can deal with sensor
failures. For example, if you get near an airport or
a military base, there's a danger a radar on the base
will "wipe out" the car radar sensor while the car is on
that stretch of highway (RF overload). That means you need
sufficient redundant methods, to make up for the lost of
one sensor type. The Lidar could be affected by a passing Lidar.
(They're already using coding techniques to cut down on
that.) The ultrasonic collision sensors could interact
with a similarly equipped car on the other side of the
road. It's hard to "guarantee" a result with this stuff.

If you look at how the computers on the Space Shuttle were
designed, you can see that if you add enough of them,
you can make a pretty decent system. Will they use a Space
Shuttle computing system on an autonomous car ? Not
until a lawsuit convinces them "it's a good idea".

Paul

Mark F wrote:
On Thu, 22 Mar 2018 09:49:54 -0400, Paul
wrote:

Rodney Pont wrote:
On Wed, 21 Mar 2018 11:39:35 -0600, Jeff Barnett wrote:

From what I read, she was hit just after stepping in the street by a
car gin a legal 45mph. That doesn't sound like lack of driver or
computer attention or lack of reflexes.
I wonder if she saw the lidar on top of it and thought 'it's one of
those computer cars, it'll stop if I step out in front of it'.
Unfortunately since she didn't survive we will never know.

There is video available now.

https://static01.nyt.com/images/2018...-master495.jpg

https://www.theverge.com/2018/3/21/1...video-released
(video at https://twitter.com/TempePolice/stat...85098542833664)
Either the headlights were no good or misaimed or the video was too
dark. If the video is too dark in means that a human would have
actually hard more time to react than indicated by the video.

Headlights should work about 150 feet. I barely saw the reflectors
on the driver's side of the road for 1.5 seconds. 45 MPH is the
highest speed I saw for the car in news reports. (Most reports
say 38MPH in a 35MPH zone.)

30MPH is 44 feet per second, so 45MPH would be 66 feet/second, so
the light reflected off of the reflectors was visible
less than 100 feet away, when the woman should have been visible
more than 150 feet away.

Also:
1. I thought that the car was able to "see" driver's side of the
road further than humans can see with low beam headlights.
2. I also thought the car was able to see the opposing side
of the road at night much further than a human could since
no care need be taken to avoid shining a "light" into the
eyes of the human or robot on the other side of the road.
(i.e., LIDAR or whatever should have "seen" further than
a human at night, except possibly for reflectors)

It'll probably be months before we get a sensor dump.
I'm hoping one of those government bodies doing an investigation,
will make the details public. As they affect public policy.

There are already videos around of what Lidar+Classifier
looks like in real time. I just can't find the one I saw
less than six months ago.

I think the Lidar dump would look pretty impressive.
The question is, what other sensor was it being correlated
with. What other sensors were working.

The video we have, comes from a recording camera, a
"flight recorder". The camera does not necessarily reflect
on the capabilities of the car. It gives us a rough idea
of what a cheap flight recorder ("Dash Cam") can do without
a Sony HAD or an image intensifier strapped to it.

Treat it as a "recording of the accident".

It remains to be seen, how sparkling clean the sensor
streams are. They should have recorded great data.
Enough data to make the car swerve and brake. Some part
of that blob of hardware, didn't work right for some
period of milliseconds. Just like a Windows 10 computer
(after you open a 36,300 page PDF in MSEdge). The Windows 10
computer does memory garbage collection for 20 seconds
straight, causing the screen to stop responding to
clicks. The classifier in the car doesn't work that way.
Part of the solution will use an embedded OS with "real time"
priority available to the applications on it. Not some
"best effort" "I'm doing garbage collection right now
don't bother me OS".

What I thought was important, is the woman was in the
middle of the road. She didn't fall off the curb while
standing on the median and get clipped. This wasn't
a "zero millisecond response" event, where nobody,
computer or human, could react in time. She's in
the middle of the road! The car made no response!

Paul

On Thu, 22 Mar 2018 08:56:16 -0400, Paul wrote:

Char Jackson wrote:
On Wed, 21 Mar 2018 16:27:07 -0700, Gene Wirchenko
wrote:

On Wed, 21 Mar 2018 14:03:43 -0500, Char Jackson
wrote:

[snip]

Exactly the same thing that happens when a human driver encounters such
a thing, except that the computer will analyze the situation and make a
decision much faster than humans can.
Will the decision be correct?

You can ask that question regardless of who or what is "behind the
wheel". Here in early 2018, I think the balance tips in favor of the
computer. As the months and next couple of years go by, I expect it to
tip overwhelmingly in that direction.


The street layout in the accident scene.

https://static01.nyt.com/images/2018...-master495.jpg

This is video from in-car. Released by the police. The player wrapper
on the Verge seems to work, while trying the twitter one using that
URL, didn't.

https://www.theverge.com/2018/3/21/1...video-released

pic.twitter.com/2dVP72TziQ

The woman hadn't just stepped off the media.

She was out in the middle of the ****ing street when hit.

This wasn't "one step off median, clipped by car in left lane".

She was moving.

She has no retro-reflectors on the two bike wheels.

She has white sneakers.

The "Safety Driver" is looking down at something
in his lap, at the time of collision. Playing
with a Smart Phone ?

*******

Sorry, but in this case, the computer loses.

I'm trying to figure out how you arrived at that (faulty) conclusion.

Did someone represent to you that the current state of the technology is
fully baked, that it's flawless under every condition? If so, you've
been misled. The technology is still very young. Testing is still in its
early stages, only recently having been moved off of test tracks onto
city streets and highways. It's not perfect, and this is why the testing
is so important.

Some of the youtube video comments claim that if a human were driving,
(or more specifically, some people are making the claim about
themselves), the accident could have, and would have, been avoided.
I call BS.

Even knowing in advance what's going to happen, which is a luxury no one
has out on the road, the person still comes out of the darkness far too
quickly to be avoided. Anyone who says they would have been able to stop
in time or otherwise avoid the accident is full of crap.

On Thu, 22 Mar 2018 14:29:12 -0400, Paul wrote:

Char Jackson wrote:
On Thu, 22 Mar 2018 13:51:22 +0000, "J. P. Gilliver (John)"
wrote:

so additional adaptation for autonomous cars is a minor problem IMO.
... but not agreed necessarily. I'd accept that any _new_ road being
built could be made autonomous-compatible (for want of a better
word/phrase) at minimal extra cost, but modifying _existing_ ones (or
de-modifying, if you like) would cost something. (_How_ much is
arguable; might be very little.)

I'm not sure why you're pressing the point of expenditures being
required when the folks who are actively doing testing repeatedly say
they neither need nor expect anything different from what we have now.

These things have features such as GPS and Google Maps.

If so, it's only as a form of secondary backup, and none of the big
players are talking about it anymore. GPS and preloaded maps were
abandoned pretty early on as the primary means of navigation for
self-driving cars and trucks. They quickly realized that preloaded maps
would never be accurate enough for use in the real world.

These days, they just use cameras and other sensors to look at the road,
look at and read the road signs, look at and react to surrounding
traffic, etc. In other words, pretty much the exact same things we all
do.

On Thu, 22 Mar 2018 11:57:30 -0700, Gene Wirchenko
wrote:

On Wed, 21 Mar 2018 19:54:06 -0400, Wolf K
wrote:

On 2018-03-21 19:27, Gene Wirchenko wrote:
On Wed, 21 Mar 2018 16:49:47 -0400, Wolf K
wrote:

On 2018-03-21 15:42, J. P. Gilliver (John) wrote:

[Live highway signs] would. Not just the cost of the signs, but their supporting
infrastructure, and its and their maintenance. Even just the cost of the
wire, in some areas, would not be insignificant.

Use solar power and rechargeable batteries (already used in a lot of
situations up here, mostly for temp signs and lights in construction
zones), and low power radio to communicate with the car. Easy-peasy (a
hobbyist could probably whip one up on a Raspberry Pi in a day or less).
The chips can be reprogrammed remotely as needed, lots of choice for that.

Snow-covered signs?

????

In British Columbia (and probably some other places), snow can
stick to signs to the point where the signs can not be read.

And?? If you're suggesting that snow-covered signs are a problem that
needs to be addressed, whether for humans or non-humans, we'll just have
to agree to disagree. I've seen a lot of snow-covered signs, and tree-
covered signs, and just plain missing signs, and so far none of those
situations has made it difficult to figure out what I should do. There
are always other contextual clues, if you look for them.

On Thu, 22 Mar 2018 15:04:34 -0400, Paul wrote:

Char Jackson wrote:
On Thu, 22 Mar 2018 13:57:44 +0000, "J. P. Gilliver (John)"
wrote:

Was it an electric car (and thus rather quiet)? Was she hard of hearing?
Do any of the autonomous systems currently being developed use the horn?
(I'm pretty sure the answer to that one is no, probably because the
false positives would result in its over-use and thus be used [as
another argument] against autonomous vehicles; however, once things
improve in that respect, I'd say they should - if audible warnings
_inside_ the car are a good thing, then they would be outside too.)

For a very long time now, I've thought that horns are obsolete. They may
have made sense a century ago when cows roamed freely and drivers hadn't
yet agreed on a common set of driving rules, but not so much anymore. A
non-directional blast of noise has never been very effective or very
efficient.

I think the last time I used a car horn was well over 45 years ago. The
last time I used the horn on my motorcycle was in the summer of 2012. I
made sure I was well out in the country with no other vehicles around. I
just wanted to see if it worked. It would never occur to me to use a
horn where others might hear it, which of course, rather defeats its
purpose.


You don't have to use a horn solely as an "indicator of displeasure".
It's not an "editorial comment" button.

It can also be used to send positional information to another
driver, who may be showing signs of not realizing you're present.

I don't know where you live or what the driving conditions are like, but
with 15-20+ vehicles close enough to hear a horn blast, what are people
supposed to do when someone hits the horn? They do the only thing that
makes sense: they ignore it. Some will flip the bird or honk in return,
but no one takes any actual action as a result. Nor should you expect
them to.

Maybe where you live it's just you and one other vehicle, so if you hear
a horn you know it's for you. That doesn't work so well when you have
lots of vehicles around you. Turn the tables and become the horn blower.
What is everyone around you supposed to do? As above, some will flip you
off, others will honk in return as if it's some kind of game, but no one
will take any action. A horn is simply ineffective in that situation.

I have avoided accidents by using the horn for that purpose.

I have avoided accidents by adjusting my speed, adjusting my lane
position, etc. Looking back over 50+ years of driving, I can't honestly
say I should have used the horn more. Fortunately, the vast majority of
drivers are like me, (excluding certain cities like NYC where horn
blowing seems to be the national pastime). I hear someone use their horn
maybe once or twice a year, some years zero, and I never take any action
as a result.

After you've had your first accident where you *should* have
used the horn, you'll be less timid in reaching for it the
next time. Trust me. Ask me some time, what it cost me
to learn that.

If you have time to use the horn, you have time to avoid the accident.
It's not about being timid or aggressive or any of that hogwash.

I don't get the thought process. You see a situation developing, and
rather than address it, you decide to use the horn. Let's see what
happens next. Ideally, every one of the 'wrong' people will forgive the
noise and ignore the warning because they'll assume it's not meant for
them. The 'right' person will realize the horn is for him or her and
will take corrective action. Is that about right? If so, it's
ridiculous. In the time that you boldly used your horn and waited for
the right person to figure out that it's intended for him and figured
out what he's doing wrong and corrects it, you've had time to avoid the
situation 10 times over. Become a defensive driver and your need to
reach for the horn will become a distant memory.

And for God sake, practice :-)

Agreed. See the note above about driving defensively. Someday you'll be
glad you practiced it. Learn to look for clues and anticipate what other
drivers will do. They taught us that in Driver's Ed when I was a kid.
They didn't teach us to use the horn, thankfully.

That's what the horn is for.

In the movies, the horn means the taxi is waiting outside the house.
That's fairly legitimate. Using the horn in traffic is much less so,
IMHO.

On Thu, 22 Mar 2018 12:09:47 -0700, Gene Wirchenko
wrote:

On Wed, 21 Mar 2018 23:40:04 -0500, Char Jackson
wrote:

On Wed, 21 Mar 2018 16:26:09 -0700, Gene Wirchenko
wrote:

On Wed, 21 Mar 2018 14:03:22 -0400, Keith Nuttle
wrote:

[snip]

This will take years, and a lot of money which many governments
organizations do not have. Even then what will happen when the Live
speed sign is vandalized, hit by a car, or a deer is standing where the
car can not communicate with the sign

True. It is bad enough when snow covers the sign so it can not
be read. I have encountered this many times in British Columbia,
Canada.

And how did you deal with it, and why do you think the car won't deal
with it in a similar fashion?

I knew the route and was expecting the sign. The road was also
steepening and getting curvy. (In case it matters, coming from
Keremeos, BC to the junction of highways 3 and 97 south of Penticton.)

I can not know that the car will have been there before or
otherwise know about it.

Had I not been there before, I might have had an accident. So
could a self-driving car.

I've been in the same situation many times and I can't say that it's
ever been a serious problem. Signs can be covered by snow, obscured by
trees or other vehicles, or the sun can be directly behind the sign,
etc. We all deal with similar situations all the time. Software is being
adapted to do likewise.

We do. We can not rely on what we can not perceive so we take
other factors into account. It is not as safe -- after all, if it
were, the sign would not be necessary -- but we do what we can.

What will a self-driving car do?

According to published reports and interviews with people who should
know, cars will do exactly as you described above. They will use the
information that's available to them.

"Paul" wrote

| This also highlights a problem with having a human in
| a robot car. A driver will react in a split second. A human
| in a robot car will probably take at least 1-2 seconds to
| judge whether they need to take over. By then it's too late.
|
| Yes. The "Supervision Paradox".
|
| That Safety Dude, might as well have been stretch out
| asleep, in the back seat.
|

I suspect it's actually much worse than it looks.
The road ahead is pitch black, despite streetlights
nearby. If it were really that dark the driver would
need hi-beams on. It looks like probably the camera
is limited, picking up nothing below a certain light
level. It's pitch black until the woman is actually
in the headlights. In other words, a human driver
probably could have seen her starting to cross.

And, yes, Mr. Safety Dude seems to be looking
at a CD player.

Char Jackson wrote:


If you have time to use the horn, you have time to avoid the accident.
It's not about being timid or aggressive or any of that hogwash.

What if there's a driver right on your rear bumper, there's no place to
back up to get away from a wild driver in front of you.

Yes, you can drive defensively, read situations before they happen,
and take measures. But once in a while, it's just "the enemy in
front of you" and a brick wall behind. And at that point,
it's horn time.

Paul

On Thu, 22 Mar 2018 12:54:30 +0000, "J. P. Gilliver (John)"
wrote:

In message , Char Jackson
writes:
On Wed, 21 Mar 2018 16:48:37 -0700, Gene Wirchenko
wrote:
[]
Think about it. When you see the next black-and-white speed
sign, it cancels the previous one. This is true even if there is no
"unless otherwise posted", so what does "unless otherwise posted" add?

I thought about it 50 years ago when I first encountered it as a new
driver and it immediately made sense. It relieves the municipality of
putting speed signs on each and every street. IOW, it allows a
municipality to have a speed limit everywhere without the expense of
having speed limit signs everywhere. They only need to put up speed
signs on main roads and streets where the speed limit is not the
default. How is that not obvious?

Perhaps written driving tests should address basic things such as this.

I don't know if things are different in the US (or some states), but
here (UK), _all_ speed limit signs _do_ apply until the next one is
encountered. So I too don't see how these "unless otherwise posted"
signs work, or at least save anything - unless the ones posting
otherwise say things like "for next quarter mile" or something. Do they?

You're over thinking things, but you're forgiven because you're an ocean
away and they apparently do things differently there. That's fine with
me. I've spent a lot of time driving in the various Western European
countries, Spain, Turkey, Korea, New Zealand, and others. Each area has
its quirks when it comes to driving.

Here in the US, especially in smaller towns, as you approach the town
you'll be likely to see two signs. The first might say, "Welcome to
Mayberry, birthplace of Andy Griffith. Population 367." The second sign
might say, "Speed Limit 25 MPH unless otherwise posted."

So now you know the name of the town that you're entering, and you know
the default speed limit in that town. The town doesn't need to stick a
speed limit sign on every single street that says "Speed Limit 25 MPH"
because they stuck a sign out at the city limits that applies to the
whole town. They only need to erect and maintain signs where the speed
limit is different from 25 MPH.

It's really very basic stuff, when you think about it.

Within that town, if you're driving down a street that doesn't have a
posted speed limit, can you be cited for speeding? Absolutely!
If the street isn't posted, what's the speed limit? 25 MPH, of course.

As a driver, it behooves you to pay attention as you enter the town. The
next town will have it's own default limit. You get used to it very
quickly, and I don't mean after you've been cited a time or two. You
just learn that it's important and you go on with your day.

Mayayana wrote:
"Paul" wrote

| This also highlights a problem with having a human in
| a robot car. A driver will react in a split second. A human
| in a robot car will probably take at least 1-2 seconds to
| judge whether they need to take over. By then it's too late.
|
| Yes. The "Supervision Paradox".
|
| That Safety Dude, might as well have been stretch out
| asleep, in the back seat.
|

I suspect it's actually much worse than it looks.
The road ahead is pitch black, despite streetlights
nearby. If it were really that dark the driver would
need hi-beams on. It looks like probably the camera
is limited, picking up nothing below a certain light
level. It's pitch black until the woman is actually
in the headlights. In other words, a human driver
probably could have seen her starting to cross.

And, yes, Mr. Safety Dude seems to be looking
at a CD player.

The street isn't actually that dark.

That's a dash cam video as far as I know.

If there's a camera intended for visual input,
it could use a Sony HAD with low light sensitivity.
I don't think they use good sensors for dash cams.

Sony controls where its HADs go, so nobody can make
a camcorder to compete with them on low light
sensitivity. It's possible someone else has a camera
chip now that has comparable sensitivity. I haven't checked
lately to see if anyone has managed to do it.

And the list here, isn't even a start on the subject.
There's probably 40 models by now. There is a scientific
camera company in Canada, that got bought out recently,
that uses HADs in some of their stuff. The cameras
were in the $1000+ range on price.

https://en.wikipedia.org/wiki/HAD_CCD

There's no indication in this article, of a competing
sensor.

https://en.wikipedia.org/wiki/Hole_accumulation_diode

Paul

"Wolf K" wrote

| Of course, she shouldn't have been crossing the road.
|

What if it were a moose? It might be the
moose at fault but that won't help the mashed
passengers in the robocar.

Jaywalking is not an excuse for hitting and
killing someone while not looking where you're
going.

In message , Char Jackson
writes:
[]
Here in the US, especially in smaller towns, as you approach the town
you'll be likely to see two signs. The first might say, "Welcome to
Mayberry, birthplace of Andy Griffith. Population 367." The second sign
might say, "Speed Limit 25 MPH unless otherwise posted."

So now you know the name of the town that you're entering, and you know
the default speed limit in that town. The town doesn't need to stick a
speed limit sign on every single street that says "Speed Limit 25 MPH"
because they stuck a sign out at the city limits that applies to the
whole town. They only need to erect and maintain signs where the speed
limit is different from 25 MPH.

If I passed a sign saying limit 25, I would assume the limit was 25
until I saw a sign saying otherwise. I would not _expect_ to see signs
repeating the 25. I do not see the need for the words "unless otherwise
posted".

It's really very basic stuff, when you think about it.

I am.

Within that town, if you're driving down a street that doesn't have a
posted speed limit, can you be cited for speeding? Absolutely!
If the street isn't posted, what's the speed limit? 25 MPH, of course.

As I said: if I passed a sign saying 25 on entering the town, I'd assume
the speed limit was 25 until told otherwise.

As a driver, it behooves you to pay attention as you enter the town. The
next town will have it's own default limit. You get used to it very
quickly, and I don't mean after you've been cited a time or two. You
just learn that it's important and you go on with your day.

The _only_ possible hole in my reasoning here would be that I might
continue to think the limit was 25 on leaving the town. In the UK,
there's usually a "derestriction" sign (white circle with black diagonal
bar) on leaving a town/village - minimal extra cost, as it's usually on
the back of the same pole that the limit for those coming the other way
and entering the town/village is on.

If you passed a sign telling you the limit was 25 but that _didn't_ say
"unless otherwise posted", what _different_ interpretation would you put
on it, to one that did have those words?
--
J. P. Gilliver. UMRA: 1960/1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf

Beatrix Potter was a bunny boiler.
- Patricia Routledge, on "Today" 2016-1-26

In article ,
am says...
There was a second or two that
an actual driver would have at least slammed on
the brakes. They might not have ben able to avoid
hitting the woman, but they wouldn't have hit her
at 38 mph.

Self-driving cars are a tort lawyer's dream.
Every time one of these cars kills someone, the
manufacturer of the car is going to be sued.
And that car manufacturer is going to lose.

"J. P. Gilliver (John)" wrote

| If you passed a sign telling you the limit was 25 but that _didn't_ say
| "unless otherwise posted", what _different_ interpretation would you put
| on it, to one that did have those words?

That may be a Brit vs American thing. In the US
we generally have default speed limits. 30 mph
for "thickly settled", which is an official term with
a defined meaning. Lately, where I live, that seems
to have changed to 25. Apparently it's been decided
that we just have to keep going slower until no more
phone addicts get run over while wandering aimlessly
into the street.

So "unless otherwise posted" defines a context.
25 is in effect on all streets without signs, unless
other signs say otherwise.

In VT the state roads have signs saying 50 mph unless
otherwise posted. It's 50 in wooded areas but goes down
when the road passes through town centers or past
large farms.

If it were to work as you describe then a new sign
would be needed at every point where someone can
enter a specific roadway.

Which reminds me of a great joke from Carrot Top
some years ago, comparing British and American cops.
He explained that in Britain, when police stop someone
for a speeding violation, they pass the car, guide the
driver to the side of the road, walk over politely, and
say something pleasant like, "Misbehaving a bit now,
are we?" When in the US Carrot Top saw the police
lights and waited for the police to pass him, but here
they stop behind the driver. So CT kept driving... waiting
for the cop to pass... until the cop finally pulled
alongside, rolled down his window, and shouted,
"HEY!! MARIO ANDRETTI!! PULL OVER!!"