Uber self-driving crash: Footage shows moment before impact [video] • Hacker News Board

Re: Uber self-driving crash: Footage shows moment before impact [video]

#11

Post by notahacker » Thu, Mar 22, 2018, 5:55 PM UTC

Beamer92 wrote: post #4

I can't be alone in thinking that the woman walking the bicycle didn't seem to be in a crosswalk... Like, ok so autonomous driving isn't perfect, and yeah the backup driver maybe should've been paying attention to the road instead of whatever she was doing. Fine. But I can't be alone in thinking it looks at least partly the woman's own fault. Right? Walking your bike, in the dark, across a road without looking when y…

Uber is a well-funded corporation with a long history of disregard for the law which in order to beat competitors has chosen to conduct real world road tests of many instances of an "autonomous driving" system apparently inferior to heavily intoxicated humans at reacting to unexpected pedestrians in certain conditions.

The woman was a homeless person who is now dead.

The above may influence people's emphasis on whose actions warrant the most scrutiny.

Re: Uber self-driving crash: Footage shows moment before impact [video]

#12

Post by jey » Thu, Mar 22, 2018, 6:00 PM UTC

LinuxBender wrote: post #3

I would argue these are not self driving cars. I would expect a self driving car to have multiple compute nodes in the trunk that use something more advanced than a gaming engine, a voting system and numerous sensors. LIDAR, infrared, thermal, sonar, other. An autonomous car that expected to operate in locations and conditions that humans often fail in should be advanced enough to know everything around it and then s…

That's a software quality issue, not one of having the right hardware. We went to the moon using pretty weak hardware and clever software.

Re: Uber self-driving crash: Footage shows moment before impact [video]

#13

Post by LinuxBender » Thu, Mar 22, 2018, 6:56 PM UTC

jey wrote: post #12

LinuxBender wrote: post #3

I would argue these are not self driving cars. I would expect a self driving car to have multiple compute nodes in the trunk that use something more advanced than a gaming engine, a voting system and numerous sensors. LIDAR, infrared, thermal, sonar, other. An autonomous car that expected to operate in locations and conditions that humans often fail in should be advanced enough to know everything around it and then s…

That's a software quality issue, not one of having the right hardware. We went to the moon using pretty weak hardware and clever software.

I think it is both. Human lives are at risk, so I expect solid battle hardened software and hardware.

Re: Uber self-driving crash: Footage shows moment before impact [video]

#14

Post by TYPE_FASTER » Thu, Mar 22, 2018, 6:57 PM UTC

mslev wrote: post #8

I'm not sure how closely we should analyze this footage. Its relatively low-quality dashcam footage, and I highly doubt its representative of what the car "sees" and what a human would see in similar conditions. That being said, Uber's LIDAR should've seen the woman in the road, right? Does the range not go beyond the lane ahead?

Published vs. real-world LIDAR ranges are radically different. Even if the LIDAR sensor is looking 360deg around the car, the Uber software may only be looking at a small cone in front of the car.

The challenge is: what do you do when something is moving toward the vehicle's path? You could track every object in front of the car, but since people walk toward the road, then stop, you can't just apply full brakes every time something is moving toward the road.

It's really, really hard to get this right: we humans can look at people standing next to a road, and, based on movement and body language, predict whether pedestrians will walk into a road.

And since there will be cases where pedestrians don't look and walk in front of a vehicle going 40mph, autonomous technology will be limited by the physics of stopping a vehicle with full brake applied.

It's impossible to figure out what the Uber vehicle should have, or could have, done autonomously without looking at the code, and all the data.

Re: Uber self-driving crash: Footage shows moment before impact [video]

#15

Post by TYPE_FASTER » Thu, Mar 22, 2018, 7:15 PM UTC

TYPE_FASTER wrote: post #14

mslev wrote: post #8

I'm not sure how closely we should analyze this footage. Its relatively low-quality dashcam footage, and I highly doubt its representative of what the car "sees" and what a human would see in similar conditions. That being said, Uber's LIDAR should've seen the woman in the road, right? Does the range not go beyond the lane ahead?

Published vs. real-world LIDAR ranges are radically different. Even if the LIDAR sensor is looking 360deg around the car, the Uber software may only be looking at a small cone in front of the car. The challenge is: what do you do when something is moving toward the vehicle's path? You could track every object in front of the car, but since people walk toward the road, then stop, you can't just apply full brakes every…

Also challenging: looking ahead around the corner. You have to be able to predict somewhat where the vehicle is going, so you have to have an accurate map of the road, and know accurately where you are.