Rollfast

Are you refering to
This shows the 'operator's' activities at the time of the fatality as well.

I don't see where this technology has proven itself capable of anything yet. Objects crossing over from the sides are not being sensed well enough. The actions of the operator show to me that it was more important to make notes or read things that monitor the actual effectiveness and safety of the technology they were sitting in.

Irresponsible and negligent behavior for a test pilot.

Leisesturm

It has crossed a line this incessant discussion of autonomous motor vehicles in a sub-forum of Bicycle Forums. Those who were expecting AV's to overcompensate for human inattention now have an object lesson that that may not always be the case. I plan to make a strenuous case for the creation of a sub-forum of Bike Forums and not in A&S, where those obsessed with driving in general, and autonomous vehicles in particular, can engage one another in pointless speculation and outrage away from the rest of us cyclists and cycling oriented forum users.

CliffordK

The "Shadows" were two fold.

Sparse street lamps created shadows, perhaps worse than not having any lamps.

But, more importantly, the Uber car was driving with the lights dimmed and dipped to the right, with barely 2 seconds of ahead view. This meant that there is a significantly worse view of objects entering the field of view from the left than entering from the right.

On a one-way road, the following distance for the vehicle ahead would have been on the cusp of legal use of high beams which would have provided much better illumination.

However, the car could likely also use fairly high intensity IR and UV to provide better illumination (or the IR LIDAR), without impacting human vision. The pedestrian's white shoes would have glowed in "black light".

Even if the car had only "seen" the pedestrian 2 seconds in advance. Say glacial slow 0.5 second processing time. That would still have left 1.5 seconds to react. Which the car didn't. And the "driver" wasn't watching the road.

Braking time would have been tight, but a quick estimate of the pedestrian's trajectory, and the car could have executed a hard left. I don't see any obvious shadows of other cars around the Uber car preventing it from reacting.

Is the Uber system incapable of reacting before it hits stuff? Fairgrounds bumper cars?

Leisesturm

There is a camera that can create an image using a SINGLE PHOTON of light. A human eye needs BILLIONS of photons to just perceive an image at all, and BILLIONS more to perceive it well. What are you talking about********** IF the results of this sad mishap truly show that "testers should pay full attention to the road at all times" then there is no future. You might as well say, only humans are gifted enough to safely (5,000 pedestrian fatalities 2017) pilot cars because (38,000 driver fatalities in 2017) cameras, LIDAR, RADAR ... these technologies are not superior to human senses.

What this sad mishap actually shows is that humans are bad enough drivers when required to give full attention because there is no computer back-up. They are even less reliable when put in the role of back-up to a computer. Long idleness is corrosive to the human spirit. Boredom ensues. Deadly. It is a MISTAKE to continue to require human back-ups in AV testing. Either scrap AV testing altogether (I'm fine with that actually) or do it right. No human interventions, the computer succeeds or it fails. If it fails people may die. If that is unacceptable... well... I don't know... there was a human involved in this accident. Someone died anyway.

How many have died since this thread began? The old fashioned way? At the hands of distracted or inept vehicle operators? Human operators? There is a point in there somewhere. It might be that the answer to the rising pedestrian, cyclist and motorist deaths taking place in America is NOT autonomous vehicles. This accident makes that clear. But humans are great at non-solutions. They will be fine with saying AV's aren't the answer and halting further progress towards their development and watching helplessly as the death toll rises... and rises... and rises... I'm not one of them.

SHBR

After a few weeks have passed, I doubt it.

This incident should be a show stopper, but it won't be, even if it continues to happen, it will be considered an acceptable risk.

I can only imagine how the AV hucksters will try to spin this P.R. disaster.

CliffordK

If only HAL had carried on an intelligent conversation, perhaps the backup driver wouldn't have nodded off.

Was the backup driver texting?

Even in the best situation with two hands on the wheel and the foot on the brake, the backup likely would have had a 3 or 4 second reaction time. Just hard to react in a split second when one is completely not expecting to be called to duty.

CliffordK

One can't create an image with a single photon... if one is lucky, one can simply detect a single photon. Two very different things.

In the past I've had issues with digital cameras being very poor at working in low light. One would have to increase the exposure time or do other adjustments. And, increasing the exposure time would mean too blurry for a hand-held camera, and probably a horrible image for a camera moving at 38 MPH.

I think the sensors are better, but still have limitations.

I tried to pick a still image from the YouTube video of the pedestrian just before becoming visible in the headlights, and tried adjusting the brightness of the Uber image, and absolutely nothing.

I tried taking my cell phone, and over-saturating the image with shadows in my house. Mixed results. But, it will warrant more testing.

Of course, the human eye is both good and bad at this. Pupils dialate in the dark for good night vision, and contract in the daytime for good daytime vision. However, the eye battles in a mixed light/dark environment.

Anyway, it is quite possible that Uber could have run 2 cameras. One under-saturated, and one over-saturated, and merged the data for a much better view in mixed light environments.

genec

Or, you can ignore these threads yourself. It's not as if the titles aren't quite visible.

GrainBrain

[Inflammatory comment]

I'd just like to congratulate Elaine on her selfless sacrifice to autonomous driving by providing Uber with data that would be impossible to gather any other way. Wearing a dark black shirt and no reflective gear, jay walking 100ft down from a lighted intersection, and literally stepping into the path of an oncoming car without even looking to her right.
[/Inflammatory comment]

I'd hazard a guess the hardware detected this, but the software didn't make sense of it.

The first thing I'd be doing is recreating the scenario.

Maelochs

Interesting. The video shows only what the video camera was able to capture---we have no idea of the quality of the video camera.

TV cameras can use a wide enough aperture that night looks like evening--but of course any light source causes glare. I don't know the sensitivity of this camera ... better or worse than a human eye.

The human eye is not as sensitive as a camera can be---and particularly if the lights are very bright, the brain does a sharp contrast---think driving towards an approaching car with very bright headlights. The eye adjusts so that everything not brightly lit is very dark.

Based just on the video, without knowing the capability of the camera, it is hard to know what a human eye might have seen.

Of course, the video camera was only recording---the AV was relying on other sensors.

But ... even if it had been a human driver ... 10 p.m. on a Sunday night on a fairly quiet road where one wouldn't expect someone to be walking .... One wouldn't be looking side-to-side for pedestrians crossing a multi-lane, 45-mph roadway at some random point.

And all the "blame the victim" talk is trash.

if that is me pushing my bike, AVs would still have a fatality-free record.

Any of us who ride bikes ... can Any car sneak up on you at night, with full headlights and traveling 38 mph?

Doubt it. Cars are loud, and headlights are bright.

The lady Obviously was oblivious to her surroundings--and I don't car if you are crossing in a crosswalk, you have to be aware that you are crossing the road where cars travel.

I will blame the victim, and if that makes the irrational people trying to "win" the debate in their own minds angry, so be it. Crazy people need treatment, not others to buy into their craziness.

This lady walked right in front of a car traveling at 38 mph, with full lights, and which would have been audible and visible for quite a ways.

An alert driver might have noticed her ... can't tell from the video. I would have expected the AV to have noticed her. But another simple, obvious fact is that she should have seen the car and not walked right in front of it.

As the Slate article notes---the big takeaway is that Uber's sensors need some boosting.

We still don't know what Waymo cars can do.

Maelochs

Again, more people use this forum as a platform to try to press others to accept their world views, rather than for rational discussion of factual things.

It is people who take a mindset and inflexibly defend it that are the issue SHBR. People who put their ideologies ahead of reality.

Glad you can so unerringly predict the future. Why didn't you predict this crash and prevent it?

I was surprised at how quickly the crash video was released.

jefnvk

Probably the same way those in favor of human drivers will continue to proclaim that the humans who kill 40000ish folks a year on the road are obviously a better option and have no issue not stopping the show because of human ineptitude? I mean, I can do a quick search on local news and find a half dozen pedestrians who have been killed by human drivers in the past three months or so. I don't hear much rumbling about how awful human drivers are.

That said, Uber needs to figure out why the person wasn't detected, but thinking a human would have been any better after watching that is still silly.

02Giant 

I will blame the vehicle. There is no doubt, it failed.

02Giant 

There is absolutely no way of knowing that for certain. What is certain, the vehicle failed at detecting the pedestrian.

I-Like-To-Bike

Trash Talk, not very funny.

jefnvk

What is certain is that 4000 or pedestrians are killed a year by humans, and after a decade or so of testing AVs on the road, computers have finally hit 1.

By my count, that is still a massive improvement.

SHBR

This is a total failure, for everyone involved.

What is the point of a standby driver if they don't watch the road?

SHBR

Well given the current state of technology, replacing all private vehicles with flawed AV vehicles would be far worse.

SHBR

Many people have predicted this would happen. What happens from this point on will be far more telling.

I-Like-To-Bike

But wouldn't the apologists just blame the pedestrian victims in most, or even all, those incidents as coming out of nowhere, or they should have seen the vehicle, or shouldn't have been in the way or didn't belong there?

Just like the apologists who emerge and bleat these excuses by rote after every cyclist-auto collision.

Just like the apologists for the AV promotion are currently doing in this specific incident, anything to deflect blame or fault from the tar baby of testing shoddy/inadequate/unsafe AV products on public roads with little or no oversight.

BTW, the statistical comparisons and analytic methods used to conclude about "massive improvement" indicate you could use a refresher course on logic as well as statistics..

njkayaker

You suggest it's relevant here:

02Giant 

How many millions of human operated vehicles are there on the road in comparison?

rgconner

That was what I was getting at.

Something ain't right.

02Giant 

Would it make any difference to anyone, had the death been to an occupant of the vehicle?
There are many instances where an animal wonders across a roadway, causing a fatal accident.

njkayaker

That doesn't cotradict what I said. I believe the "wysiwyg" versions of these came out first on the Mac.


Keep in mind that what I said was a bit of an over simplification.