The Uber victim did come from the shadows. Arizona gets dark at night, especially during a new moon. The question isn't why the optical cameras didn't see her, it's why did the LIDAR failed to detect her?
AZ gets dark at night. The workaround is for cars to have headlights.
Not knowing the law as it applies for (semi-)autonomous vehicle testing in Arizona, I am very curious how this one turns out.
Based on my experience driving at night (In Norway, where it also gets pretty dark), even in rural areas with no street lights, you tend to discover pedestrians -even the ones not wearing safety reflectors- significantly more than a second and a half before you cross paths with them.
So - while the vehicle is driving autonomously, there's still a driver (or whatever we should call it, seeing as they do not drive as such) behind the wheel.
If the driver had paid attention to the road, rather than her cell phone, this accident likely wouldn't have happened.
So - who is to blame, legally speaking? The 'driver' for not paying attention? The coder implementing the control algorithms? The HW engineering team deciding which sensors to use? Someone else?
This will be a lot simpler when vehicles are 'properly' autonomous - but right now, it seems AVs simply give the 'driver' all sorts of incentive for not paying attention, while still not absolving them of responsibility. That's the worst of both worlds.
Uber is a corporation, so it as a legal entity is liable, any kind of individual liability is possible to be litigated but very difficult to win. One problem is that corporate charters once granted are never revoked so corporations just keep doing terrible things and diffusing the costs of their actions onto society with minuscule repercussions. You can bet if I was looking down at my phone and not at the road and then hit a woman in the dark (all captured on camera) I would be going to jail... But, Uber has done some really despicable things (stealing the medical records of a rape victim to use for intimidation), and now they've likely killed someone and no one will ever see cell time.
Notwithstanding the fact that a typical visible light imager in a driver assistance system alternates exposure times between frames, especially at night: one over-exposed image to discern dark texture, and one under-exposed image to find the color of point light sources, such as traffic or brake lights.
The released video looks like the latter, though it could also just be standalone dashcam footage. It is also pretty mangled by compression, especially the dark parts are just completely blocky, so it is probably not a good indicator of what the vehicle saw.
OTOH, bikes can be notoriously hard on image recognition, for example it is not unheard of that reflective rims are interpreted as continuation of a curb or lane marking. LIDAR and RADAR should see them, though, and they probably did, so it is indeed unclear where it all went wrong.
The first thing I noticed in watching the video is that the Uber car's lights seem to be ridiculously focused on a very small patch in front. The dark spot the cyclist appears in remains so until little more than a second before she is tragically hit.
The question is indeed why the electronics failed to detect her, but there's something really wrong with car lights like that.
I did, including the comments to the article where they clarify that the follow up videos were taken with high-beams on, i.e., in dissimilar circumstances. And the picture in the comments showing the location of the crash, where there were no streetlights in the immediate vicinity, and a large bush shielding the median, from which the victim is alleged to have come from.
I think there are serious questions about why the Uber car didn't detect the victim, given all the sensors it was supposed to be using, but "came from the shadows" is an accurate description based on the Google Street View and overhead views of the crash site.
It helps to read the entire article before commenting.
Not knowing the law as it applies for (semi-)autonomous vehicle testing in Arizona, I am very curious how this one turns out.
Based on my experience driving at night (In Norway, where it also gets pretty dark), even in rural areas with no street lights, you tend to discover pedestrians -even the ones not wearing safety reflectors- significantly more than a second and a half before you cross paths with them.
So - while the vehicle is driving autonomously, there's still a driver (or whatever we should call it, seeing as they do not drive as such) behind the wheel.
If the driver had paid attention to the road, rather than her cell phone, this accident likely wouldn't have happened.
So - who is to blame, legally speaking? The 'driver' for not paying attention? The coder implementing the control algorithms? The HW engineering team deciding which sensors to use? Someone else?
This will be a lot simpler when vehicles are 'properly' autonomous - but right now, it seems AVs simply give the 'driver' all sorts of incentive for not paying attention, while still not absolving them of responsibility. That's the worst of both worlds.