sort of OT autonomous car fatality

[tuscanyswe]

Originally Posted by David Tollefson

Why does the lighting in the videos even matter? Wasn't the car equipped with detection systems that don't rely on the visible light spectrum?

Yes for the accident its does not matter like you say.
But what it implies if tampered is that someone is allrdy trying to push public opinion by making it look like a very possible accident for any driver while infact it could look very different in reality. Perhaps even like an accident that would not have happened to a human driver?

[MattTuck]

Originally Posted by David Tollefson

Why does the lighting in the videos even matter? Wasn't the car equipped with detection systems that don't rely on the visible light spectrum?

It depends on what your thinking is. I think most people are trying to suggest that the autonomous driving system made a mistake, and the person was not recognized as an object in the road, and thus no actions were taken to avoid her.

And they're using the difference in video lighting to indicate that Uber either 1) deliberately made the video darker so as not to make it look so bad as hitting someone on a well lit road or 2) to indicate that the road was actually quite well lit, regardless of what you see in the video.

[ripvanrando]

Originally Posted by David Tollefson

Why does the lighting in the videos even matter? Wasn't the car equipped with detection systems that don't rely on the visible light spectrum?

LIDAR and Radar can identify an object but they can't see a red light or identify objects with precision. It would seem to me that LIDAR should have alerted and slowed the vehicle and the optical vision system should have confirmed that a human being was crossing the street and avoidance and braking algorithms engaged. Optical systems need visible light, which is clearly missing from the dashcam rendition. That is why light is important or at least in my opinion. If Uber doctored the dashcam video to make Volvo's lights look crappier than they already are, shame on them.

[ultraman6970]

I have rear almost nothing about the accident in question and the little i read gave me the impression that besides having a safety driver inside of the car that is/was a felon, the driver was distracted but at the same time if you see the video it was super dark and the lady pretty much was nowhere to be seen (need to check details of the area, speed of the car and stuff) aswell... dont take me wrong ok but im under the impression that uber wants to throw the felon under the bus for non stopping or looking at the road????

I have to drive a bus at night a couple of days a week and depending on the conditions you can;t see jack; in the video im under the impression the car is going way too fast for my taste and looked like it was a foggy night? Dunno man... one thing is to have accidents like this that could have happened to anybody but when you have like zillions of sensors and craaaaaaaaapp around the car that did not work how you can throw the driver (felon or not) under the bus?

The idea of autonomous cars sucks in practice IMO.

[Mark McM]

Originally Posted by ripvanrando

LIDAR and Radar can identify an object but they can't see a red light or identify objects with precision. It would seem to me that LIDAR should have alerted and slowed the vehicle and the optical vision system should have confirmed that a human being was crossing the street and avoidance and braking algorithms engaged. Optical systems need visible light, which is clearly missing from the dashcam rendition. That is why light is important or at least in my opinion. If Uber doctored the dashcam video to make Volvo's lights look crappier than they already are, shame on them.

I won't argue that the optical system iss needed to actually identify the object in the middle of the road - but shouldn't the LIDAR and/or Radar detecting that there is something in the road be enough to stop or go around the object? Are there any types of objects (of the general size of a human) that there is no need to avoid? As stated before, lighting conditions are no excuse for this crash.

[MattTuck]

Originally Posted by Mark McM

I won't argue that the optical system iss needed to actually identify the object in the middle of the road - but shouldn't the LIDAR and/or Radar detecting that there is something in the road be enough to stop or go around the object? Are there any types of objects (of the general size of a human) that there is no need to avoid? As stated before, lighting conditions are no excuse for this crash.

I think you're saying two different things. To go around an object requires complex algorithms and intelligence to actually do. To just detect something in front of the car and just 'dumb' apply the brakes, that is much easier, and is probably something the Volvo should have done automatically... but perhaps the Volvo system is over-ridden by the autonomous driving system that Uber uses.

[72gmc]

Today's NYT looks into the state of this project prior to the fatality.

Everything uber does should be questioned.

[ripvanrando]

Originally Posted by Mark McM

I won't argue that the optical system iss needed to actually identify the object in the middle of the road - but shouldn't the LIDAR and/or Radar detecting that there is something in the road be enough to stop or go around the object? Are there any types of objects (of the general size of a human) that there is no need to avoid? As stated before, lighting conditions are no excuse for this crash.

Motor cycle in opposite lane.

I can only guess that so much information needs to be processed and prioritized that like all/most failures, they are multifactorial and poor lighting could have been one chain in the failure link. Truly speculation on my part. I used to know something about imaging systems in the stone age.

[Kontact]

Originally Posted by ripvanrando

Motor cycle in opposite lane.

I can only guess that so much information needs to be processed and prioritized that like all/most failures, they are multifactorial and poor lighting could have been one chain in the failure link. Truly speculation on my part. I used to know something about imaging systems in the stone age.

Something in the opposite lane is already avoided.

The point here is that there are no man size objects that a car could drive into without concern. No ghosts, no radar reflecting columns of fog, no free roaming towers of whipped cream. Anything in the lane that big is something that needs to be avoided. The AI doesn't need to know whether it is a tree limb, a refrigerator or a person to know that hitting it isn't acceptable.

[ripvanrando]

Originally Posted by Kontact

Something in the opposite lane is already avoided.

The point here is that there are no man size objects that a car could drive into without concern. No ghosts, no radar reflecting columns of fog, no free roaming towers of whipped cream. Anything in the lane that big is something that needs to be avoided. The AI doesn't need to know whether it is a tree limb, a refrigerator or a person to know that hitting it isn't acceptable.

Until it isn't. How are objects prioritized. What is the computational delay especially if the optical system is compromised. Another example would be a car door opening. Parked cars would have to be de-prioritized.

That is the point

[Kontact]

Originally Posted by ripvanrando

Until it isn't. How are objects prioritized. What is the computational delay especially if the optical system is compromised. Another example would be a car door opening. Parked cars would have to be de-prioritized.

That is the point

Parked cars in the path of the vehicle would totally be a priority. I don't follow the point you're making - a correctly designed system isn't going to choose to drive into something. The pedestrian was moving slowly enough that there is no reason the Uber shouldn't have regarded it as an object in its path without even needing to guess where she was going.

[FlashUNC]

Interesting fact for comparison I saw on Jalopnik: Waymo, Google's autonomous car effort, has traveled 5,600 miles and counting without human intervention needed. Ubers program has failed to meet it's own stated goal of traveling more than 13 miles without requiring human intervention.

Their system is a joke and shouldn't be on public roads.

[HenryA]

An autonomous vehicle is going to do things a human won’t do, yet it is expected to interact with humans on and around the road. Drive the posted speed limit using various sensors that allow the car to “see” better than human? Why, yes. But the humans around it still can’t see any better than before such a device hit the road. Is it legal to drive the posted limit when conditions might dictate to a human to slow down? Questionable, perhaps. Will the machine make such a decision?

Until roads are redesigned to accommodate autonomous vehicles and eliminate human interaction with said AVs I don’t want the damned things around me. Not at all.

As for liability for an AV, I look at them like a “set-gun”. The owner/user should be strictly liable for any and all harm done by the device they operate in public.

Someone will no doubt gain great wealth from the production of AVs but the general public will suffer and suffer badly. They’re coming, but you won’t like it.

[pasadena]

Originally Posted by ripvanrando

poor lighting could have been one chain in the failure link.

Uber's uses Lidar

The president of Velodyne Lidar (makes the sensors in the Uber car) already said the system should have 'seen' the victim crossing the street and avoided them.
The system is designed specifically to avoid this exact type of situation.
They do not understand why the Uber car did not stop.

I have no faith in current technology. They fail even in cars now, that are not even close to autonomous... and they use those same sensors for autonomous cars.
Snow, freezing rain, debris, dirt that covers sensors, objects that "confuse" systems...
I can point to two specific
1. The Mercedes caravan cars that self-slammed on the brakes riding near the peloton, and Mark Cavendish slammed into the back of (forget the race, a few weeks back)

2. auto-braking sensors that slam on the brakes in curving, tight roads when it 'sees' the guardrail or lumpy road and tight turn (Mulholland Drive)

those are just ones I know about but i have read of cars slamming on brakes on the freeway for no reason among a host of other issues.

[Louis]

As drivers and cyclists, how high do we think the bar should be set for this type of vehicle?

1) Very close to perfection? (Assumes that 100% perfect isn't possible)
2) Maybe not 100% perfect, but still pretty darn good?
3) At least as good as the average human?
4) As good as a so-so human?
5) Same bar as humans, so as good as the very worst humans who are currently allowed to drive?