Robocars - Autonomous Drive, Self-driving, Driver-less cars

Tempe Police Chief Sylvia Moir has told the San Francisco Chronicle that the SUV likely wouldn’t be found at fault. But two experts who viewed the video told The Associated Press that the SUV’s laser and radar sensors should have spotted Herzberg and her bicycle in time to brake.
“The victim did not come out of nowhere. She’s moving on a dark road, but it’s an open road, so Lidar (laser) and radar should have detected and classified her” as a human, said Bryant Walker Smith, a University of South Carolina law professor who studies autonomous vehicles.

http://nbc4i.com/2018/03/22/deadly-crash-raises-questions-about-uber-self-driving-system/

"Tempe police have identified the driver as 44-year-old Rafael Vasquez. Court records show someone with the same name and birthdate as Vasquez spent more than four years in prison for two felony convictions — for making false statements when obtaining unemployment benefits and attempted armed robbery — before starting work as an Uber driver."

Uber said Vasquez met the company’s vetting requirements.

If that's how they do their staff checks, how thoroughly do they do their equipment checks?

(offences >7 years ago so ok by them)

Amnon Shashua, head of Intel Corp's Mobileye vision systems unit, said the industry must do more. He has called for the self-driving vehicle industry to develop "provable safety assurances".

"We need to prove that these vehicles are much, much safer than humans," Shashua told Reuters. "How do you go and guarantee that you have a technology that the probability of a fatality per one hour of driving is 1,000 times better than a human? Nobody talks about that because nobody knows what to do."

https://finance.yahoo.com/news/self-driving-car-industry-confronts-trust-issues-uber-114112829--finance.html

so Lidar (laser) and radar should have detected and classified her” as a human

Human or not, prob a good idea to break for any kind of object coming your way...

Yeah, it's almost like their system doesn't work in the way it's supposed to...

Or that the LIDAR/RADAR array wasn't enabled at the time.

It would be considerably cheaper if a company could come up with an autonomous car that didn't rely on an expensive LIDAR/RADAR array.

Well not having it enabled is pretty feckin' broken if you ask me.

I doubt the car would have been out LIDAR/RADAR enabled. Obviously it's failed as it's intended to work in low light (or rather the algorithm that interprets the data has failed).
The bloke sitting in the driver's seat was useless certainly.

I'd be interested to know what the car decided it saw. I doubt it saw nothing, but there's obviously some determination it's made that meant it didn't judge Herzberg as an obstacle. Whether it's because she wasn't quite person shaped, or bike shaped. No idea. I also don't think we'll find out what the problem was. Unless it makes it to court, and unless as part of that Uber are required to explain why the car didn't stop. And if they're required to do that then the data it had available at the time and the algorithms used to make a decision would hopefully be available so that it doesn't just end up as talking heads and Twitter arguments saying "Well Uber has a bit of a Bro culture so they're probably happy to send out faulty cars and run over homeless people", which doesn't get anywhere.

Since when did lidar/radar care about ambient light?

Woman safety driver, no? I don't know, there's been male and female names mentioned in the press.

Since when did lidar/radar care about ambient light?

It shouldn't care. Sorry, that could be ambiguous without punctuation.

Obviously the RADAR/LIDAR has failed or the system that interprets the data has failed.
It is intended to work regardless of daylight/low light.

Wasn't sure about the safety driver. Quite possibly wrong, only saw a grainy vid. Either way, they were useless.

I doubt the car would have been out LIDAR/RADAR enabled. Obviously it's failed as it's intended to work in low light (or rather the algorithm that interprets the data has failed).

As mentioned LIDAR/RADAR is active (not passive) so the lighting conditions do not affect it.

To spell it out, the rumour is that LIDAR was disabled on purpose as they were conducting testing using passive cameras only (since a passive setup would be considerably cheaper than active LIDAR/RADAR). The test failed with the cost a human life.

If this turns out to be true it's going to be an even bigger shitstorm for Uber although I guess they'll just be throwing the safety driver under the bus.

It’s unrelated to this but something I’ve wondered recently is whether the LIDAR/RADAR systems of different cars could interfere with each other, or whether they’re somehow coded/encrypted so only the car that transmits a signal can read the scattered return.

To spell it out, the rumour is that LIDAR was disabled on purpose as they were conducting testing using passive cameras only (since a passive setup would be considerably cheaper than active LIDAR/RADAR). The test failed with the cost a human life.

If this turns out to be true it's going to be an even bigger shitstorm for Uber.

Ah, I hadn't read the full thread so had thought people were assuming it just hadn't been working that day and someone had gone 'ah fuck it, just send it out anyway', which I thought unlikely.
If it transpires that they were testing it using just cameras then yes, that's a different thing entirely.

That, you would think, assumes that they've done extensive testing with both enabled and found no previous instances where the LIDAR has provided information that would have averted a crash that the passive cameras had not seen.

Amnon Shashua, head of Intel Corp's Mobileye vision systems unit, said the industry must do more. He has called for the self-driving vehicle industry to develop "provable safety assurances".

Take his comments with a dash of competetive piss.. Uber uses a platform based around NVIDIA's Drive (formerly known as the Drive-PX2).. A lot of MobilEye is old school computer vision which amends itself better to these kinds of tests..

the rumour is

Where?

Twitter, so nothing substantial, which is why I stressed that it is a rumour.

“The problem of complacent safety drivers is going to be a problem for every company.”

This is an issue not just for SDCs and AVs but increasingly for all advanced motorcars given the power of their ADAS. The solution that I have been advocating is to sense the affective state of the driver. This can be done contactless by measuring the micro movements of the driver's body.

Indeed, there's plenty of eye-tracking technology out there.

Just need the equivalent of a dead man's switch in the car. If the safety driver doesn't pay enough attention then warnings sound, still not looking and the car pulls over and stops. Too many warnings sound on a journey and it's dealt with as a disciplinary matter.

Should have that in every car.

Good point, no warnings though, just AutoCockPunch(TM).

Only on the road would we allow an autonomous vehicles that isn't 100% safe, and when one does take a life continue to develop the technology anyway.

Is this the first death by robot?

Only on the road would we allow an autonomous vehicles that isn't 100% safe

No vehicles are 100% safe (irrespective of the driver). Even the technology of normal cars fail and kill people (where it is no fault of the driver).

Saying that, Uber's technology as it stands has demonstrated that it is not safe enough for use on the roads at this point in time. But that doesn't mean we should stop developing it, or not let it back on the roads in the future.

It needs to have much more stringent controls before it is let back on the road, and I think all of the cities that have granted companies permission to do AV testing on the roads will be reviewing things in the light of this accident.

and when one does take a life continue to develop the technology anyway.

You don't just stop developing it because someone was killed. Almost every bit of technology has resulted in a death at some point (electricity, medicine, cars/trains/planes, etc).

You stop developing it when you get to the point that you know there are problems that you simply can't solve (or when you burn through all of your VC cash).

Plenty of people have been killed by robots - but this is probably the first case of a random death of a random person in public by a semi autonomous robot, as opposed to death of an employee in a workplace. Or a combatant. Etc.

There are issues of so-called cognitive loading.. Many studies have clearly indicated that low driver workload is likely to induce drowsiness in drivers.

Indeed, there's plenty of eye-tracking technology out there.

Actually gaze tracking is not a good solution. It is not only extremely difficult to track gaze in a "any driver", "any lighting condition" setting-- pupil reflections etc. are non-trivial-- but people can track objects despite being cognitively "elsewhere". Hand on heart.. How many of us have not slept in a lecture with our eyes open trying to look awake?

I know a few companies trying to do facial expression recognition-- Affectiva is probably the most famous of the lot-- but it too is, in my opinion, sub-optimal as, despite Eckman's belief, it is quite culturally biased (Affectiva addresses the bias by using modified models for different countries but this is not terribly good. They assume that immigrant communities are never isolated and rapidly adopt local cues)

I have been mentoring a start-up to sense driver state by measuring their vibrations-- similar to many ways to Ballistocardiography. These movements can be, for example, contactless sensed using accelerometers, sub-THz radar or even cameras.

Beyond drowiness I'm also interested in states such as fear and anger which might result in sub-optimal disengagements.