Any details on headline - Arizona pedestrian is killed by Uber self-driving car

[d21mike]

Well, considering that AP1 can read speed limit signs, it seems like something the car could adjust for on the fly. Otherwise it would always be putting construction workers in danger when they put up temporary lower speed limit signs. No mapping solution will ever be able to keep up with that.

There is a safety driver behind the wheel. Would you not think one of their responsibilities is to make sure the car is not speeding? We Tesla Drivers are required to be in control even when on AP where AP seems to be performing perfectly. Like a well marked highway. Example, when going from freeway to freeway you might need to manually reduce the TACC Speed.

 

[scottm]

Dash cam video might tell how much reaction time was available for either car or human driver to apply brakes.

 

[Tam]

...It was reported the car did not apply the breaks "to any degree"...

There's no excuse that the LIDAR could not detect 3 football field away as the pedestrian was walking from in the darkness of shadows from the center median toward the side walk.

Lack of sunlight is not a problem for LIDAR because it shines laser beams even in complete darkness.

May be Uber is using an inferior LIDAR that does not have a good range.

And what's a good range to Uber LIDAR?

The system should have detected the pedestrian as she walked from the car's front right and didn't got hit till she reached the car's front left when the car hit her at last.

The same goes with the human driver: the driver should have applied the brake when the pedestrian was at car's front right and shouldn't wait for the car to collide the pedestrian at the car's front left.

It's a machine's failure as well as the human driver's failure.

This is not rocket science: Waymo LIDAR has been able to do it even with blinded corner and a group of crossing chickens who did not use a crosswalk!

 

[brkaus]

Newest I found -

(Bloomberg) — Police say a video from the Uber self-driving car that struck and killed a woman Sunday shows her moving in front of it suddenly, a factor that investigators are likely to focus on as they assess the performance of the technology in the first pedestrian fatality involving an autonomous vehicle.

The Uber had a forward-facing video recorder, which showed the woman was walking a bike at about 10 p.m. and moved into traffic from a dark center median. “It’s very clear it would have been difficult to avoid this collision in any kind of mode,” Sylvia Moir, police chief in Tempe, Arizona, told the San Francisco Chronicle.

“The driver said it was like a flash, the person walked out in front of them,” Moir said, referring to the backup driver who was behind the wheel but not operating the vehicle. “His first alert to the collision was the sound of the collision.”

 

[SMAlset]

....The system should have detected the pedestrian as she walked from the car's front right and didn't got hit till she reached the car's front left when the car hit her at last.

The same goes with the human driver: the driver should have applied the brake when the pedestrian was at car's front right and shouldn't wait for the car to collide the pedestrian at the car's front left.....

I think you meant to say the pedestrian with bike entered the roadway from the left center median, walked her bike over 2 left turn lanes, then across two regular car lanes, then into the combo car/bike lane and past the left drivers side of the Uber vehicle until getting struck by the right passenger side of the Uber. To me that's quite a bit of time to be observed in the open roadway and as they aproached the vehicle from the left side. How fast could she have been walking or running with the bike? And if she saw the car with head lights on (and we've been told neither person appeared to be under any drugs or alcohol) why would she try to out run the car and not wait for it to pass?

I really hope they show the video to the public. If I were driving and not paying attention to the road view in front of me maybe I could see failing to notice something entering a few lanes of traffic ahead of me and possibly not see the person until they were walking in front of my lane until it was too late. As for the car even with our current AP 2.0 software it doesn't pick up too much on the sides as you are driving in your lane. But as a driver you do see other cars on the road in other lanes.

 

[S4WRXTTCS]

The system should have detected the pedestrian as she walked from the car's front right and didn't got hit till she reached the car's front left when the car hit her at last.

The damage to the car is on the right side.

Do we even know which direction she was headed?

 

[S4WRXTTCS]

This article in from azcentral (part of USA Today) gives some more info on the two people involved. It does state again that the lady who died was crossing from west to east (from Center Median toward park area). Still would assume a person walking a bike loaded down with stuff and passing from left to right of you across several lanes of traffic should have been detected by some form of the equipment on the Uber, and still hard to understand a driver who was paying attention to the road in front not seeing that person either given the car had headlights and there were two street lights on each side of the NB traffic lane there as well.

The article doesn't really state whether she started at the medium (to the cars left) or to the cars right. It does state west to east, but doesn't say which direction the car was going.

If she was coming from the medium then I find it hard to not lay at least partial blame on Uber, and the safety driver. If I hit someone on the right side of my car and they were coming from the left then I'd be really pissed at myself for failing to slow down. But, I wouldn't feel nearly as responsible if they darted into the road from the right.

 

[Olds442]

What? No. Come on, this is Uber we're talking about. The only people driving Ubers these days are either molesters or felons. In this case, I believe it was a felon at the wheel: Operator of self-driving Uber vehicle that killed pedestrian was felon

surprised level is zero.

 

[SMAlset]

The article doesn't really state whether she started at the medium (to the cars left) or to the cars right. It does state west to east, but doesn't say which direction the car was going.

If she was coming from the medium then I find it hard to not lay at least partial blame on Uber, and the safety driver. If I hit someone on the right side of my car and they were coming from the left then I'd be really pissed at myself for failing to slow down. But, I wouldn't feel nearly as responsible if they darted into the road from the right.

If you look at the photos and the report from ABC15 who was on site reporting, the car was northbound towards E Curry on N Mill Avenue so that makes the island center median on the left (or west). Numerous articles reported the pedestrian was traveling from the median including the SFChronicle, "Pushing a bicycle laden with plastic shopping bags, a woman abruptly walked from a center median into a lane of traffic and was struck by a self-driving Uber operating in autonomous mode." This hasn't been refuted anywhere I've seen.

I agree about a pedestrian stepping out suddenly from parked cars giving you little time to react. If we consider she came from the right side instead and it was reported wrong and not corrected over several days, looking at photos of the roadway at the curb near where this happened, there weren't any illegally parked cars there and these were traffic lanes. The landscaping and palm trees look fairly set back so that area would have been visible I assume especially with overhead street lighting. Until a lot more is made known I'm not sure you can say the driver or Uber was not at fault at all.

The news and police mention looking at the front and rear dashcams in the vehicle. Wonder how well they record in low light. The human eye and/or some sensing systems probably saw much more. From what I understand the vehicle also had a camera watching the safety driver (just like the Model 3 has I suspect). Curious what the driver was looking at as they approached this stretch of roadway.

 

[RubberToe]

(Bloomberg) — Police say...

The Uber had a forward-facing video recorder, which showed the woman was walking a bike at about 10 p.m. and moved into traffic from a dark center median. “It’s very clear it would have been difficult to avoid this collision in any kind of mode,” Sylvia Moir, police chief in Tempe, Arizona, told the San Francisco Chronicle.

“The driver said it was like a flash, the person walked out in front of them,” Moir said, referring to the backup driver who was behind the wheel but not operating the vehicle. “His first alert to the collision was the sound of the collision.”

The police chief, Sylvia Moir, I'm assuming isn't familiar with how lidar works. The lidar is constructing a world view hundreds of time per second. A working lidar would be able to follow the person from wherever they were coming towards the car. If the person started 2 feet from the car or 10 feet from the car when they started the "quick movement", the lidar should be able to see it. Whether the Uber software is using the lidar, and how fast they are processing the scene, are entirely other questions. Waymo videos clearly show code that classifies detected objects and reacts to them in real time.

Getting back to the police chiefs comment... While a human being may not have been able to see a person jump from the shadows into the path of the car, there are certainly sensor systems that would have seen the person coming and could have reacted in time.

Kinda gets back to the question of whether the autonomous car is held to a higher standard than the human, and if so, how much higher, and what regimen of testing would be required to verify that. Because different manufacturers are using different sensor suites, you can't just run simulations of all possible scenarios, due to the different hardware configurations. Waymo does run many more hours of simulations than they do real driving.

RT

 

[SMAlset]

I also found this statement by the chief a bit strange (as reported in the SFChronicle), "The incident happened within perhaps 100 yards of a crosswalk, Moir said. “It is dangerous to cross roadways in the evening hour when well-illuminated, managed crosswalks are available,” she said. The emphasis is mine, just wondering why she would bring this up, why not just say pedestrians should always use available crosswalks or is it because the area where it happened was well illuminated and maybe the safety driver should have noticed her in advance.

 

[Tam]

The damage to the car is on the right side.

Do we even know which direction she was headed?

Sorry for my confusion and I wrote the direction wrong.

The correct one is:

Center Median to sidewalk: Car's front left to car's front right.

"Pushing a bicycle laden with plastic shopping bags, a woman abruptly walked from a center median into a lane of traffic and was struck by a self-driving Uber operating in autonomous mode."

 

[Tam]

...whether the autonomous car is held to a higher standard than the human...

If a robot dog specification says it barks and it doesn't, then its specification needs to be revised to reflect the reality.

Machines are held to its own specifications. If LIDAR can't even see a walking pedestrian on the car's front left and it let the system hit her once she got to the car's front right, then its specifications should be revised as NOT 3-football length away and NOT 300 meter range (>900 foot range)!

 

[S4WRXTTCS]

surprised level is zero.

Does it matter that safety driver was a felon from something he did around 18 years ago? Does that somehow disqualify him from being a safety driver?

The whole safety driver thing is a bit bizarre anyways.

You know how hard it is to concentrate on a task that's completely automated where only once every X number of days do you have to do anything. From what we know so far that is looking like a major contributor. Where the pedestrian was coming across from the left, and across the lane before the car hit her.

I don't fault the safety driver (unless he was texting, browsing facebook, etc), but I do have some major concerns whether a safety driver is an adequate backup. People driving with AP have a had a fair amount of "I thought the car was going to ..." accidents along with the "It happened suddenly" excuse. There are also people like myself who purposely don't use AP in a lot of situations because it diminishes my situational awareness.

From a technology standpoint I'm not sure how the Lidar missed it. The car wasn't traveling particularly fast, and we're talking about a fairly large obstruction entering the path of the vehicle.

The radar missed it as well, and it seems like a fairly big combination of things for it to miss.

The vision system likely missed it because it's a person pushing a bike with a bunch of crap on it. I'm pretty sure that wasn't part of any of the training data for the neural network.

 

[daktari]

I would have hoped the technology reduces reaction times when braking. And maybe also apply brakes harder. These numbers seem right:
38 Mph Braking Distance
Thinking Distance: 37 ft (11 m)
Braking Distance: 72 ft (22 m)
Stopping Distance: 109 ft (33 m)
Google Street view shows pretty decent visibility, at least a hundred meters.

But look at this video for reaction time:

 

[Evoforce]

Not one person has talked about possible other traffic going in the same direction, that may have somehow obscured her from view, until she popped out at the last minute.

I don't know what happened, but many other variables could be involved. Her view could have been blocked also by cars in the lane closest to her. She may have jetted out after waiting on the near lane to clear and didn't see the Uber in the far lane that was lagging a few paces back.

So possibly, in this scenario, she didn't quite "thread the needle" so to speak.

 

[scaesare]

There's no excuse that the LIDAR could not detect 3 football field away as the pedestrian was walking from in the darkness of shadows from the center median toward the side walk.

Lack of sunlight is not a problem for LIDAR because it shines laser beams even in complete darkness.

May be Uber is using an inferior LIDAR that does not have a good range.

And what's a good range to Uber LIDAR?

The system should have detected the pedestrian as she walked from the car's front right and didn't got hit till she reached the car's front left when the car hit her at last.

The same goes with the human driver: the driver should have applied the brake when the pedestrian was at car's front right and shouldn't wait for the car to collide the pedestrian at the car's front left.

It's a machine's failure as well as the human driver's failure.

This is not rocket science: Waymo LIDAR has been able to do it even with blinded corner and a group of crossing chickens who did not use a crosswalk!

A pedestrian can, however, be concealed from Lidar too. Think darting out from between parked cars.

 

[Tam]

...obscured her from view, until she popped out at the last minute...

That's possible but how could other cars in other lanes in the same direction (as well as in other opposite direction) could dodge her while Uber could not?

She walked across from the opposite traffic flow first then reached a Center Median with lots of trees and shadows.

She then proceeded and walked out from the shadows of trees in the Center Median to the open to:

cross lane#1 (left turn lane),

cross lane#2 (another left turn lane),

cross lane#3 (not Uber's lane),

cross lane#4 (not Uber's lane),

then:

she finally reached lane#5 which was Uber's lane and she was cleared from Uber's front left bumper but she got hit at Uber's front right bumper at lane#5.

It's doubtful that there was any car blocking Uber's front left bumper except for the pedestrian herself!

There were no reports of other cars even after the police watched the video.

The condition was reported as clear and dry but she did step out of the shadows from the Center Median trees and she almost made it as her body bounced onto the sidewalk together with her bicycle.

A diagram from The New York Times:

NTSB picture of the Uber with the right front damage:

 

[McRat]

That looks like one of those areas where you are so far from any intersection, that you are allowed to cross the road on foot.

There is a street light in front of the impact site, behind, and on the opposite side. Vision from the driver's seat would be very good. Wide sidewalk to clear visibility. If lanes are 12' wide, I'm seeing 12x9 or 108' of visibility at a minimum.
Note that when a vehicle brakes, it slows immediately which increases the time to impact, or the effective distance. I would think the lack of braking was the culprit.
I wonder what the Reaction Time of a Uber AV is to threats? If it was human-speed RT (.25s for an alert driver), I can't understand how the impact was at full speed.

I have my doubts whether the Uber 'saw' the bicycle at all even when it hit it.

 

[McRat]

This is an odd road layout.
Green lines are a bicycle lane, compare to picture in above post.
Blue circles are street lights I can easily ID.