[siliconwrath]

It certainly does look bad for self driving vehicles, and I definitely feel terrible that a life was lost from this.

However, how does one develop a self driving vehicle that’s 100% automated without the ability to test in real driving conditions? Despite this accident, self-driving vehicles have a fairly safe driving record for the number of miles and time they have been active.

Details on California accident rates for self driving vehicles, for example, show mostly minor fender benders despite more frequent accidents: http://journals.plos.org/plosone/article?id=10.1371/journal....

[danso]

It's a difficult question. Because we cannot just dive into it assuming that the toll (in life/safety) of beta testing AVs on public roads will result in a net benefit within a reasonable timespan. The human driving fatality rate is ~1 death per 100 million miles. Uber has 2 million miles driven and 1 fatality. It's obviously unfair to extrapolate and say that Uber has 50x the fatality rate of normal driving. But that means we have to keep testing Uber AVs on public roads.

What if an Uber AV accidentally kills someone at the 2.5M mark? That's still not enough data to statistically compare apples to apples. Maybe the next 100M miles of Uber testing is fatality free...that still wouldn't be completely enough (right? I'm not great at stats but I would think we need at least a billion?). Of course, it could go the other way, with Uber AVs killing someone every 1M miles.

As a general tech optimist, I'm inclined to think tech will get better, overall. But let's face it, that's not a given. And in the meantime, it's likely the tech upper-class won't be the ones who suffer the most while tech improves. The case at hand being the prime example: a homeless recently-imprisoned woman was killed.

Earlier today someone submitted an interesting RAND study that argued that the testing time for autonomous vehicles to meet statistical reliability for safety testing would be on the order of decades, or even centuries, and there would still be no guarantee that AVs would be safer. I'm hoping RAND is just being really pessimistic here...

https://www.rand.org/pubs/research_reports/RR1478.html

[blt]

Build quite large test facility with diverse artificial scenes. Pay people to walk/bike/drive around. Use external motion capture, GPS, stationary radars, etc. with additional offboard computation to act as a watchdog and step in when the onboard systems fail. Would be expensive, but the companies pursuing autonomous driving can afford it.

[InclinedPlane]

There are several parts here: the hardware a self-driving car uses to "see", the software a self-driving car uses to process sensor input into a representation of the vehicle's surroundings, and then the software that makes decisions and issues commands to the car to actually "drive" it.

You have a car with all of this running but a human driving while you drive around for hundreds of millions of vehicle miles. You then review the data for these trips and use it to assess both the ability of the hardware/software to maintain a meaningful degree of "situational awareness" as well as the reasonableness of the software to control the car if it had actually been doing so. From this you can determine how good of a job the system is doing and build a fairly good idea of how much you can actually trust the car to drive on its own. Then you can let the automation drive the car with a human behind the wheel and actually paying attention. From that you can further improve your assessment of how well the car operates under realistic driving conditions.

However, if the human isn't paying attention then you potentially significantly increase risk. Especially if you short-changed the previous step of monitoring the automation's performance while "side seat driving".

In the case of, say, Waymo, they've done a fairly good job here because they've been careful and thorough in each step. In the case of Uber, as is reflective of their corporate culture, they have rushed ahead and taken on a lot more risk than they should have, in this case putting bystanders in harm's way.