Uber self-driving test car ...pedestrian death

[93Regina]

In the video, observe where his head/eyes are focused...which is not in sync with external video.

Driver sees accident coming from his left view, but external camera shows a different view.

Guilty...another SmartPhone Accident 30 days in Jail, and one year without a cellphone.

[mrbrian200]

Angle of the headlamps looks about right for low beams. The point about these cameras not having the same dynamic range as the human eye is 100% spot on. In the video it looks like the pedestrian pops out of nowhere. But that's because the limited dynamic range of these cameras hides the light scattered outside the primary forward focus point. The human eye can see objects in these peripheral lower lit areas. If the driver had been looking up instead of down it is very likely they would have seen and reacted to the pedestrian before it was too late.

Now, these views are from low res monitoring 'dashcam' type cameras that probably aren't tied into the autonomous system. The problem engineers need to work on is why the autonomous system which would use a more sensitive array of specialized cameras/IR imaging and possibly some form of radar didn't identify a 'large object' moving into the vehicle's path and take measures to avoid a collision regardless whether the system was able to differentiate that it was a pedestrian, large animal, or a plastic garbage bin blowing across the road (if there were a heavy wind).

So.. before I asked for the (dashcam) video. But because this happened at night, the dashcam video raises as many questions as it answers. The pedestrian did not pop out from between 2 parked tall SUVs or was standing on the right hand side to then suddenly venture into the vehicle's path in the last instant. The data captured within the autonomous system (not these dash cams) is what needs to be examined. The driver could probably be held at least partially at fault IMO.

Had the driver been looking forward instead of down, the collision would have been much less likely. I would imagine at the point when a driver overrides the autonomous system for any reason/corrective action, an event with the previous 10-20 seconds of system data would be stored for the engineers to examine. I said the driver could be held partially responsible. Primary faults appear to be with the pedestrian (appearing to be unaware/walking right in front of a moving vehicle) and the autonomous system, in the end, was in control of the vehicle in the critical moments where measures to avoid a collision needed to be applied.

I'd bet $ the drivers job was to keep their eyes on the road at all times as if the autonomous system didn't exist and they were actually controlling the vehicle. If this is the case it is another example of problems we've already seen with people's tendency to become over confident in systems that aren't, and probably never will be perfect. Remember the guy killed in his Tesla that drove into the side of a clearly visible truck trailer and police determined the driver appeared to be watching a DVD movie on a handheld device instead of paying attention to the road?
However, if this driver was looking down at something at UBERs instruction (such as being told to closely watch some vehicle/system status monitor) then the driver would be absolved. In that case it would be UBER engineers/management who became over confident.

[93Regina]

pedestrian

I can't find the cite, but I think GM was having an issue with pedestrians, but I recall another news tidbit about autonomous vehicle having issues with bicycles. This GM press release below suggests a pedestrian problem.

GM Files Patent Application for Vehicle-to-Pedestrian Communication
by Michael Accardi January 31, 2018
==============================================

[93Regina]

Self-driving cars aren’t good at detecting cyclists. The latest proposed fix is a cop-out.

Autonomous cars have a potentially fatal flaw: They struggle to detect
and react to cyclists on the road. According to a January 2017 report
by IEEE Spectrum, bicycles are generally considered "the most
difficult detection problem that autonomous vehicle systems face."
...
...
One solution presented by Ford, Tome Software, and Trek Bicycle at CES
last month is a concept known as bicycle-to-vehicle communications.
Instead of just autonomous vehicles (or all motorized vehicles) on the
road being able to wirelessly communicate their position and
intentions with one another, bikes would be able to join the party.
...
...
There’s one problem: This is cheating. Autonomous cars, out there beta
testing on U.S. roads today, can accurately detect other vehicles,
pedestrians, even big game charging suddenly across a street. Forcing
cyclists alone to strap a sensor onto their backs feels like a crutch,
a cop-out.

[mrbrian200]

The interesting details about this case is that from the video it appears the person wasn't riding the bicycle, but walking it across the street (as is legally required to traverse a pedestrian crosswalk or sidewalk in many areas). Considered a pedestrian not a cyclist while doing this, hence police referencing a ped, not a cyclist. **At a walking pace** I'm puzzled why the autonomous system didn't recognize a slow moving object the size of a person/bicycle moving into the vehicle's path. Tells me autonomous systems are way not ready for prime time yet, or at least not the one UBER is using.

I recall a write-up about Volvo semi-autonomous systems having trouble with kangaroos. I recall it had something to do with distance calculations in the system being thrown off while the kangaroo was airborne/mid-hop.

I agree, the proposal to strap communication devices to bicycles is a cop out. Inevitably some incident will happen where people will get the idea that pedestrians might not be safe without one too. Which I could see turning into a PR disaster. If the technology isn't there or an adequate level of sophistication isn't currently feasible due to cost or aesthetics, OEs and companies like UBER need to step it back to R&D and not try to push these systems up to mass market scale just yet.

At the moment I think the technology is viable enough to provide some basic collision mitigation and assistive functionality. But the human driver must at least have the impression they retain control of the vehicle at all times thus forcing people to remain alert and focused. AI systems can intervene if and when the system manages to properly identify a danger and the human driver isn't making ideal corrective actions. This business of giving people the impression that autonomous systems are 'doing the driving'. It's not there yet. Driving a car in the real world requires genuine subjective intelligence. Tech people behind these AI/autonomous systems been watching Star Trek a little too much and practically confusing current AI technology that can't adapt outside a 'predefined envelope of foreshadowing' with infinitely self adapting 'sentient' intelligence like Commander Data or the Holographic Doctor.
ahem...Probably not in my lifetime.

[93Regina]

> person wasn't riding the bicycle

Being homeless (?) might explain why. Its apparent she wasn't looking either, so a suicide or bad judgment (limited mental ability).

I don't know that area....and what might be expected to happen.

>basic collision mitigation and assistive functionality

I have no issues with cruise control on highways, nor collision mitigation (deer too)....autonomous driving requires current maps, which must be updated, over and over.

Several months ago, I "kissed" a deer, on two lane back-county road...due to oncoming driver's brights being on too long...I suspect he saw deer, and forgot to dim his lights. When he flipped off brights...there it was...Shiiiit...

[tryingbe]

The driver is a her.

Even with the Pedestrian detection system, the car would have not have stopped.

The camera's function is deactivated and will not detect a pedestrian in darkness or in tunnels, even if there is street lighting in the area.

https://support.volvocars.com/en-CA/car ... e8004927e7

[mrbrian200]

The driver is a her.

Even with the Pedestrian detection system, the car would have not have stopped.

The camera's function is deactivated and will not detect a pedestrian in darkness or in tunnels, even if there is street lighting in the area.

https://support.volvocars.com/en-CA/car ... e8004927e7

I thought UBERs autonomous system was a completely separate system from Volvo's city safe technology which would need to be deactivated/not present on the UBER autonomous cars ???

UBER is rolling vehicles in fully autonomous mode after dark, they would absolutely have to incorporate ways to detect pedestrians at night. The google street view camera vehicles I believe only operate during the daytime..as if Google has a better handle on current limitations (and for what they're doing it doesn't make sense to run at night anyway).

[93Regina]

The driver is a her.

What driver? Smartphone addiction

UK: 9 November 2015 - Rise of smartphone injuries: 43% of people have walked into something while glued to their screen, while 60% have dropped their phone onto their face while reading

March 30, 2017 Smartphones may be to blame for unprecedented spike in pedestrian deaths

April 4, 2017 More evidence that smartphones and driving don't mix - Cambridge Mobil Telematics (CMT) has released new data showing that phone distraction occurred during 52 percent of trips that resulted in a crash

PS: Thx for correction...and Volvo's information

[93Regina]

I thought UBERs autonomous system was a completely separate system from Volvo's city safe technology which would need to be deactivated/not present on the UBER autonomous cars ??

Combining Volvo’s cars with Uber’s self-driving system