Uber Driver Was Streaming Hulu Just Before Fatal Self-Driving Car Crash, Says Police (arstechnica.com)

An anonymous reader quotes a report from Ars Technica: Tempe, Arizona, police have released a massive report on the fatal Uber vehicle crash that killed pedestrian Elaine Herzberg in March. The report provides more evidence that driver Rafaela Vasquez was distracted in the seconds before the crash. "This crash would not have occurred if Vasquez would have been monitoring the vehicle and roadway conditions and was not distracted,'' the report concludes. Police obtained records from Hulu suggesting that Vasquez was watching "The Voice," a singing talent competition that airs on NBC, just before the crash. Hulu's records showed she began watching the program at 9:16pm. Streaming of the show ended at 9:59pm, which "coincides with the approximate time of the collision," according to the police report.

Comments

Watching or listening?

[squiggleslash]

Isn't "The Voice" a singing competition? It's not impossible to envisage someone streaming that with no intention of watching the video.

Does this matter in Uber's case?

[qzzpjs]

In this case she may have saved a life by doing her job and paying attention, but the final solution assumes that nobody is sitting behind that wheel. This is still a major fail for Uber's software.

None of the above

[PraiseBob]

How about we arrest the driver for watching TV while they were supposed to be operating a multi-ton piece of machinery?

Re:None of the above

[flug]

It's worth pointing out that this type of response by drivers is predictable. Not necessarily watching TV but zoning out in one way or another. You'll see Tesla trot out the excuse every time as well: "This system requires constant monitoring by the driver, it's not really fully self-driving, and the crash was the driver's fault for not paying attention when they should have."

But: equal--or even more--blame has to go to the designers of the system and testing protocol for not taking this obvious and well known fact about human behavior into consideration when designing their system and their testing protocol.

It's a simple fact of human behavior that once the system looks like it's working OK for a few dozen to a few hundred miles, you assume it's OK and you start to tune out.

In reality, drivers average between 90 million (auto v. auto fatalities) and 480 million (auto v. pedestrian fatalities) miles between fatal collisions. So a system that can manage to go a few dozen or a few hundred miles without anything disastrous happening is still many orders of magnitude less capable than even the worst human driver. But once the automated system has driven a certain route a few times successfully, just about any human "monitor" is going to start have confidence in the system and tune out.

There are many ways around this issue, and companies shouldn't be allowed to test self-driving systems out on the public roads without using some or all of them:

  * Far more extensive testing can be done using simulators etc before going live on public roads. They should be testing many billions of miles in this type of environment first. Some companies are putting more emphasis on this now (ie, nVidia). All should be required to do this or something similar.

  * Far more testing should be done on tracks & other non-public locations before testing proceeds on public roads.

  * Systems should not be allowed to be tested on the public roads until they have proven they are actually capable.

  * If systems do require human "safety drivers" as a backup then various monitoring systems and protocols must be in place to ensure that the humans are actually doing the work. You can't just hire random people at $12/hour, give them 3 hours of training, and hope. That is guaranteed failure.

  * Companies doing this type of testing need to be 100% responsible for anything that goes wrong. The fact that some employee wasn't doing something 100% right is no excuse. The companies need to have enough of a safety culture, safety system, and safety protocol in place that they know whether or not any individual tester is doing what they should or not.

  * Most of all, these safety-critical systems must be engineered in an environment of safety-critical engineering. Not the "move fast and break things" bullshit software development culture that is currently so pervasive.

"Move fast and break things" might be a great philosophy for developing a cell phone app, but operating a motor vehicle is a safety critical system operating in an environment with very high risk of serious injury and death. The systems and the testing must be designed to take this seriously from top to bottom.

FWIW Uber's corporate culture is like the polar opposite of this from top to bottom.

Congress is trying to pass a bill to allow nationwide testing of self-driving vehicle that is laughably lacking in any type of oversight. More here:

http://saferoads.org/2018/06/1...

https://www.cnbc.com/2018/06/1...

Re:I remember a lot of people defending Uber

[kiviQr]

note that "dimly list street" - what you have seen is footage from camera that does not perform well in low light condition. Human eye works way better - as long as you focus it on the road...

Re:I remember a lot of people defending Uber

[Anonymous Coward]

Correct - testing auto pilot includes the part where we see the effects of a distracted driver if something goes wrong with the automation... automation who's very existence practically begs the human driver to ignore the road. Similar to texting while driving - drivers SHOULD pay attention to the road rather than text, but we know many will. Deciding on if auto pilot is safe enough to use in mass production must account for the fact the human drivers won't pay attention to the road as this accident revealed in testing.

Re:I remember a lot of people defending Uber

[sexconker]

You have no clue how good the human eye is and how poor a digital replica is, do you?

Re: I remember a lot of people defending Uber

[mlyle]

The drivers testing the vehicles are employees. :P

Re:They still should have had 2 people

[magarity]

in the car. If nothing else it decreases the odds. They'd both have to be watching Hulu to mess up.

It is a sad comment on society of epic proportions if companies need to hire two people to police each other from cell phone addiction.