This is simply not true. The dashcam footage released by Tempe police does not depict the driving conditions at the site of the accident accurately at all. Check out [1] for the actual driving conditions at night.
I'm not usually the one to talk about conspiracy, but Tempe police had no legal responsibility to release the footage since the investigation is still ongoing. Couple this with the statement released by the Tempe chief of police defending Uber, I'm inclined to believe Uber and some state government officials are working together to sway the public opinion and minimize damage. At the end of the day, the victim is painted as an idiot who practically committed suicide. This is gross injustice.
The thing is that it doesn't really matter. There are only 2 options here. Either, the car should have seen the woman and taken corrective action or it was driving faster than it can see. Just because the speed limit says X doesn't mean that's the speed you should be going. Often, actual conditions require driving slower.
Well, I do feel like I've been had after that and a few other videos, so thank you for that.
My point though, is that I don't want self driving cars to defeat simulations or tests, I want self driving cars that can react to real driving conditions.
But how else do you imagine scenarios like these are tested? Normally there aren't a lot of people willing to jump randomly out in front of cars to test their reaction.
I really don't know. Even with all the hours of recorded road time by Google/Uber used as a testbed for proper driving conditions, most machine learning systems still suffer from figuring out ways of cheating[1], and in those cases it could be fatal.
[1] "The program even learns how to take advantage of bugs and glitches, like timing jumps so that Mario begins falling again at the exact time that he makes contact with a Goomba. Mario's invincible when he's falling, so the touch kills the Goomba, not Mario, and it gives him a further jump boost." http://www.wired.co.uk/article/super-mario-solved
I think comparing this to human drug trials is an appeal to emotion.
I also think you're saying that bad tests are better than no tests, which I don't agree with. These would be bad tests because they wouldn't constitute the complicated reality of driving, they inherently only summarize the output of such a complex system.
You seem to have confused the concept of bad tests with incomplete tests. Asking to stop the car when pedestrian shaped obstacles appear seems like a bare minimum that we should expect from an AI before letting it drive on real roads. We can't test every driving situation in a lab, but that doesn't mean we can't test some of them.
It would be morally bankrupt to rely entirely on a quantitative test for something that is qualitative, I don't know how much clearer with the previous comment I can get?
You're missing the fact that any time someone takes a drug, they're essentially testing it against their unique body chemistry. The same as self-driving cars in unique situations.
The only difference is the car feedback loop is way better instrumented, and doesn't have to rely on mass statistics to inform recalls.
> Half of all humans who have ever lived, died from mosquito-borne illnes[s], about 5,000,000,000 human lives.
Your math doesn't work.
The current population of the world is between 7 and 8 billion.
If half of all humans who have ever lived were 5 billion, the total number of people who have ever lived would be 10 billion, of which at least 7 billion are still alive.
That leaves less than 5 billion who could possibly have ever died for any reason, much less mosquito-borne illnesses.
“Modern” Homo sapiens (that is, people who were roughly like we are now) first walked the Earth about 50,000 years ago. Since then, more than 108 billion members of our species have ever been born, according to estimates by Population Reference Bureau (PRB).
But what if the self driving cars fail the simulation? The simulation would be the first thing the car needs to pass, if car passes the simulated, then special tests, then you get approved to test on public roads with a security driver, I have no idea when is the time to test without a driver.
I know this is going to sound uncharitable and I acknowledge that any death is tragic but...
This person carelessly crossed a busy road at night! The victim practically did commit suicide. She could have just as easily been killed by a human-driving car, just like thousands of pedestrians do every year. Why are we so quick to try to blame the car? At least the car was in the right place.
The car was not in the right place, or it would not have hit her. In the state of Arizona a pedestrian has right of way in front of any vehicle, and the driver must yield to any pedestrian on the same side of the road.
Edited to add: We don't know anything about her physical or mental condition that might have made it difficult for her to perceive the car or the distance when she was crossing, so to assume that she had perfect perception and chose to cross in front anyway assumes facts not in evidence.
Coming back to this, I was wrong about the details of Arizona law. They require pedestrians to yield to drivers outside of crosswalks, marked or unmarked.
However, I still think Uber violated 28-794.1 and .3.
Notwithstanding the provisions of this chapter every driver of a vehicle shall:
1. Exercise due care to avoid colliding with any pedestrian on any roadway.
2. Give warning by sounding the horn when necessary.
3. Exercise proper precaution on observing a child or a confused or incapacitated person on a roadway.
I also think most current human drivers would also expect to spend significant time in jail for causing death by driving so careless they didn't even try to brake when confronted by an pedestrian in the roadway (even if she shouldn't have been there and an accident may not have been entirely avoidable)
I'm fine with Uber's executives and programmers accepting the same level of culpability as an alternative to meeting higher standards than humans to get on the road. Ditto for their culpability for potential bugs which would be considered [attempted] vehicular homicide in the absence of a more logical explanation for the vehicle's actions if driven by humans. Abdicate that responsibility and you'd better be prepared for more stringent tests.
Tests allow irresponsible people to blame the test when their system fails, I worry that a system like that will cause the opposite of your intentions.
Human drivers don't get to cite their test passes when their driving fails either. The point isn't that [overoptimisation to pass] tests should be the be only thing on which they are judged, the point is that if autonomous driving system developers want the special privilege of not being held as strictly liable as humans for erratic driving resulting from decisions they made, they really ought to expect more stringent testing regimes.
That video is hardly representative, you don't need street lights to see people at night. Simple car headlights are plenty to see someone in time to stop at 45 MPH.
The 'driver' was not paying attention and the software failed. An attentive human driver acting alone would have easily stopped.
> Simple car headlights are plenty to see someone in time to stop at 45 MPH.
That's a common misconception. You will see the pedestrian in time if they are near a street light or wearing something reflective. In complete darkness, if the pedestrian is wearing bright, colorful clothing (say, a yellow jacket), you will see them from 120 feet or so; often enough to react and come to a full stop, but not if the road is slippery or the driver less than fully attentive. And a pedestrian wearing black in the darkness is a dead pedestrian, basically invisible until the vehicle is on them.
I have stopped at 45MPH to a black dude crossing the street in dark blue clothing viable only in my headlights. Night vision degrades with age, but it's simply unsafe to out-speed your headlights and if you can't see you need to slow down.
PS: Just Starlight aka no moon and no clouds is plenty of light to see and walk around. I can't read small text, but you really should not be blind at night.
They claim a pedestrian without reflectors can be seen from 50 m with low beams. Other source I've seen break it down by the color of clothing; white might be seen from 50 m (160 feet), yellow from 37 m (120 feet), and darker colors even worse, down to black which is visible from a distance too short to react.
High beams are a much different story, of course. And perhaps the numbers for HID lamps on a luxury car are better - at the expense of blinding and distracting the other drivers on the road.
Good lowbeams don't blind other cars. They create a curtain of light that's fairly uniform as bright spots hurt night vision.
At 45 MPH you should stop in under 50 meters in good road conditions. 50 MPH is around 53 meters +/- depending on the vehicle and road conditions.
Reflectors are very useful because they add cushion to that. And believe me that was heavy breaking situation. But, the Uber car did not even start to break.
I just drove that stretch of road twice. I had no trouble seeing any portion of the road where the pedestrian was crossing, from easily 100 yards away. This human could have stopped. My assessment is that the majority of human drivers in AZ could have also.
Most humans can't do large square roots in their head either. I still expect my calculator to do it properly. What's the point of adding lidar to cars if they aren't going to use it?
You don't think most humans would see someone crossing a flat straight road in an area with mediocre lighting? If that's true we should have thousands more deaths a year like this.
Most current human drivers wouldn't be able to pass a test with the conditions that resulted in the fatal crash.