"Returning to the files, I soon found the puzzle piece. It was in a document Otto had submitted in support of its application to operate the state’s first Autonomous Technology Certification Facility (ATCF). This is an independent organization that will assess all autonomous vehicles that want to operate (rather than just test) in Nevada, checking that they comply with the state’s safety requirements and traffic laws. The application was a success, and construction of the facility is now underway in a suburb of Las Vegas. (Yes, you read that right — Uber will soon be assessing other companies’ compliance with safety rules.)"
https://www.wired.com/2017/02/how-my-public-records-request-...
Certify what, that the car passed the test track? That’s not the hard part.
The hard part is taking responsibility for the performance of the vehicle including the liability for accidents and malfunctions.
When the autonomous vehicles become a comercial reality they will be need to be treated like aircraft: regular maintenance by certified engineers and NO unapproved modifications. Gone will be the ability to pimp your ride, gone will be back yard mechanics and certainly no “jail broken” cars running modified software on the road.
I don't see why an experimental self driving taxi should be treated more leniently than that.
Many people would travel in a time of their lives like that, and I see this as much like that. Its just that this is depth over breadth, that is, exploring my beloved NYC in very fine detail excites me more than exploring the world in almost pointlessly coarse detail...
Making a few extra bucks and making myself useful rather than purely self indulgent (unlike traditional tourism) is a plus also. It's also something that comes with zero schedule commitment and no boss. It also dovetails with a longtime obsession with transportation.
That wasn't the point of my comment of course, just stating where I got this info, but I guess I piqued your curiosity.
> and an algorithm which encouraged the car to speed 3 mph over the limit when it was dark conditions
But the linked NYT article said 40 in a 45 zone.
After the bridge it changes to 45mph so it could be the Uber was also in a 10mph speed up phase in the hundreds of feet from the bridge right where the pedestrian was.
I’d expect a self driving car to be more aware of the rules and drive according to the law.
What are “reasonable” rules for autonomous vehicles? Should they stick to the limit literally, so only start accelerating after the increased limit sign? (There will need to be warning signs when a speed limit reduction is ahead to give the vehicle time to comply.)
Would a vehicle speeding up before the sign indicate a failure or error in the system?
It literally is the status quo. If I were still working on autonomous cars, I'd be much more inclined to work into the true status quo rather than imagine the status quo ought to be different and build towards that.
On the northbound side it changes from 35 mph to 45 mph right after the bridge just before the freeway. Southbound, it changes from 45 mph to 35 mph about 140 feet after Washington St, well before the freeway and bridge, and about 150 ft north of where the accident occurred.
If someone near the scene wanted to find the speed limit, and failed to consider that the two directions might have different speeds, I'd expect them to be more likely to find the southbound speed. That's because it is common to place speed limit signs shortly after intersections so that people turning onto the road won't have to go long to find out the correct speed limit. So I'd expect them to head toward the Washington/Mill intersection to start searching for a speed limit sign.
For the curious, here is the accident location on Google street view [1]. Back up 17 steps from that, continuing to face north, if you want to see the applicable speed limit sign. Or cross to the other side from the accident location and go north 8 steps to find the speed limit for the southbound road (it's on the last street light before you reach the intersection).
[1] https://www.google.com/maps/@33.4363673,-111.9425082,3a,75y,...
A test-road intersected by a tram-track with a motorized trolly on which various test-subjects are mounted; the car drives the road (at night under controlled lighting conditions) and the tolley drives the test-subject across the car's path at various speeds and times. The car must perform an emergency-brake, swerve or other safety move depending on the test-subject (a crash-test dummy, cardboard-box, plastic-bag, fake-bird etc).
The test result data could be fed back into the machine-learning to inform simulations for further testing in virtual environments. A company with Uber's resources can afford to build such a test-track and should have. There's a gap in the market for a company to provide such testing facilities for car makers.
So, which is it? A controlled environment, or a real city? I think it is a bit naive to expect /every/ edge-case to be worked out prior to even a testing permit being given out. If you could pass a test that faithfully simulates ALL the intricacies of driving in a real city, you're well past testing phase anyway.
Further, this ignores that we may be able to incentivize this behavior (learn all the lessons you can in controlled environments before putting people at danger) in other ways, e.g. financial penalties.
I guess that description doesn't make much sense, but it would be interesting to develop a virtual testing environment for testing the algorithm package even though that doesn't address hardware and numerous other concerns.
Looks like Nvidia has turned it into a self-driving car on chip, due for release later in 2018:
https://www.engadget.com/2017/10/10/nvidia-introduces-a-comp...
The NTSB investigation should be interesting.
This would incentivise the companies (and force the drivers) to not disengage in borderline dangerous situations.
They need to go beyond periodic checks. Simply have constant video telepresence any time they are operating. Another operator monitors them to ensure they are attentive.
Most current human drivers wouldn't be able to pass a test with the conditions that resulted in the fatal crash.
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.
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.
[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
But I dont see how you can avoid having a simulation. These events are just too rare to wait until they occur in real life.
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.
By the way, you may be misunderstanding “appeal to emotion” as “having emotions” and that’s worrying. I can only hope your attitude is literally pathological, and not just your personality. Either way, please never work near people, or I’m a critical system effecting people.
The only difference is the car feedback loop is way better instrumented, and doesn't have to rely on mass statistics to inform recalls.
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.
Source: https://www.prb.org/howmanypeoplehaveeverlivedonearth/
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.
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.
https://www.azleg.gov/ars/28/00793.htm
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'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.
The 'driver' was not paying attention and the software failed. An attentive human driver acting alone would have easily stopped.
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.
Now, some mostly older cars have terrible headlights, but that's another story. http://www.iihs.org/iihs/sr/statusreport/article/51/3/1 Note Uber hit someone in bluejeans and a pink bike which is not that dark: https://www.theguardian.com/technology/2018/mar/22/video-rel...
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.
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.
>The Phoenix area was added a year ago, and quickly became the company’s main testing ground, with 400 employees and more than 150 autonomous cars driving local roads because of "favorable regulatory environment, favorable weather conditions,” according to a company document.
The sorts of regulations you're suggesting cost California the privilege of hosting Uber's self driving program.
Funny how the pile on is targeted at Uber and the tech, when this is a test vehicle with a human driver who was hired to be responsible to monitor, respond, and override the computer at all times.
The human failed. Blame the computer. Makes sense.
But it's obvious that riding long stretches in an automatic car is going to lead to people getting bored and distracting themselves and not paying attention. It's a HUGE topic in every discussion on self-driving cars. It's why Teslas require you to grab the wheel every 15 seconds.
Uber should have short shifts, bans on personal cell phones, attention and tiredness tracking (which IIRC Volvos already have!) etc, etc to combat this. But they don't actually care, it's just a regulatory checkmark, so they stick some minimum wage warm body in there on a full-time shift and this is what happens when it gets late.
Other safety-conscious know that humans will cut corners and hence have tricks like safety interlocks, double-checks, multiple people involved etc to combat that. In aerospace you can't just say "well the maintenance guy shouldn't have done that, 300 people died and it was all his fault" - you have redundant systems, verifications, checks, etc.
edit: from the article, Uber used to have two safety drivers in each car, but cut it down to one for whatever reason
> When Uber moved to a single operator, some employees expressed safety concerns to managers, according to the two people familiar with Uber’s operations. They were worried that going solo would make it harder to remain alert during hours of monotonous driving.
They knew this was going to be unsafe but did it anyway
* Construct a test to be able to operate under day and night conditions. Periodically audit each operator to confirm they can pass this. This is essentially equivalent of a driving test but would be a controlled environment that simulates real cities. Maybe have robot humans to simulate pedestrians and obviously have other cars on the road. * Require that operators test X miles of specific scenarios (city driving, highway, etc.) each month and audit their # of disengagements in each of these scenarios. When it his a threshold, their license is revoked and they need to go back to the lab. They should also submit videos of clips for all disengagements. * Require all operators to have video on their safety operators and submit records of % of the time the operator has eyes on the road along with videos. Regulators can revoke license if they see issues with the the operator not paying attention.
We can't rely on a corporation to safely controlling their experiments, the regulators need to step up here. was uber doing anything for example, to make sure the operator was watching the road? The operator is looking down for most of the time during the crash video.