Uber's autonomous vehicles require frequent human intervention


#1

Originally published at: https://boingboing.net/2018/03/23/nyt-ubers-autonomous-vehicl.html


#2

The thing to remember about this Uber crash is that this was a perfect scenario for machine assistance.

A woman was crossing a street near a city park, walking a bicycle across the street. A single human-shaped object moves perpendicular to vehicle motion at a slow, steady pace. Constant bearing, decreasing range. If the Uber lidars were working, this is a very standard test scenario. It’s a scenario so favorable to the cars, they put it in car advertisements for emergency braking features (usually with deer).

And the car failed to respond. It’s a reasonable question to ask if a $25k Subaru with the Eyesight package would have done fine here.

We accept that autonomous vehicles are going to screw up. The world is chaotic and software doesn’t handle that well. But this wasn’t that class of problem. This was a lab test scenario for protecting pedestrians, and Uber failed.

People should maybe go to jail for this one.


#3

autonomous vehicles require frequent human intervention

To be fair, so do I.


#4

Dang. When was the last time two major corporations had that big of a technological gap for the same class of product? I mean people gave Microsoft a lot of flak for being behind the curve but the iPod still couldn’t hold 430 times as many songs as the Zune.

(Also the Zune never killed anyone.)


#5

Luckily the local chief of police is on the case with a harrowing account of how the reckless pedestrian sprang from the shadows in a move nobody could have predicted.

Progress cannot fail; it can only be failed, apparently.


#6

In this country you have to be driving drunk on an expired license to go to jail for hitting someone with your car. (as long as you don’t leave the scene)

I wonder though what happened to all that CMU talent they poached way back when. Those guys were some of the best-- Surely they’re at similar spending levels as Google on this one, but maybe Google is able to self-poach/self-pollinate across their organization in an effective way?


#7

Does Yahoo (a division of Oath Inc., keeping AOL company, A Verizon Company) count? They were still banging rocks together and hand-populating a ‘portal’ until surprisingly late in the game.


#8

My spitball (prompted by Cory’s comment re PageRank in yesterday’s 3D printing article) was that maybe something bled over from Google Search being constantly challenged by SEOs. Goddess knows that their algorithms have been under attack for longer than Uber has existed.


#9

Human intervention every 13 miles.
See, now this is real proof Uber shouldn’t be testing these vehicles on real roads yet.


#10

Subaru owner here (including one with EyeSight). While it’s very cool technology, I doubt it would have helped here. It uses front-facing cameras and line of sight. It’s optimized for avoiding collisions with other vehicles, not other kinds of obstacles. Maybe it could have avoided the collision here but it’s not designed for it.


#11

My mistake, I was thinking of Volvo’s system. Uses radars and cameras to reduces impact speeds with people.

http://wardsauto.com/news-analysis/volvo-demos-pedestrian-protection-system


#12

Yeah, whatever it is that went horribly wrong, it seems like it’s fully a result of what Uber has been doing, no one else. This should have been a situation where the car worked really well.

I guess I always assumed, based on what I’d been reading, that Uber’s efforts were a ridiculously long way behind Google’s, but I also assumed, that since they were on the roads, that they at least functioned, on some basic levels. Apparently not… The gap is far worse than I assumed.

Or, at the very least, not without two backup drivers with their eyes firmly fixed on the road and hands on steering wheels. I can’t help but think that Uber only got approval to do these road tests because legislators lumped them in with Google’s actually functional technology.


#13

See, this is something I’ve been wondering. Do the Volvos that Uber uses have built in features that would have prevented this death?

I get that Uber would shut off all the built-in functionality to test their own, but it really does not bode well if existing ‘not-so-smart’ warning systems would have worked.


#14

And it was a Volvo XC60, which, optionally has:

Corporations are people. :slight_smile:


#15

Honestly, this means that Uber’s self driving car project was so ineffective that it was basically a fraud. I see a few different self driving car companies driving test vehicles around SF, I wonder how similarly crappy their tech is. I like to pull up next to them whenever I’m on my bicycle and they’re stopped at a red light, and try to get a look at the laptop that the passenger is invariably coding on. They’ll always close it when they see me, and I always figured they were worried about trade secrets, but now I’m thinking that their trade secret might simply be that their system doesn’t actually work.


#16

No, it means they’re about 2-3 years behind Google…


#17

Right, but I see the self-driving heavily branded Uber Volvos driving all around San Francisco, in heavy traffic, at least 2 or 3 times a day. That, to me at least, is a pretty heavy statement of competency, meaning their system is safe enough to be used on populated roads, and almost ready to be applied commercially (and hence a factor in their stock valuation). That’s not to say their tech doesn’t work at all, but that they vastly over stated how well it works, and how far they have to go before it’s even close to commercially viable. A similar exaggeration in, for example, the pharmaceutical industry, would be considered outright fraud.


#18

Google has been doing this for much longer, and they started when their intervention mileage was similar.

Well, there is an ongoing crisis in repeatability and cherry-picking in the pharmaceutical industry, so I’d disagree on that specific point…

But as you pointed out and apparently edited out, Uber is sketchy and should not be given much benefit of the doubt.

To me, the epic joke in all this is that these cars are going to have a fit with Texas (and specifically Dallas) as the roads here and the typical construction markings make them barely navigable by a human standards.


#19

Uber’s “God view” & settlement fine, Uber absolving themselves of employer responsibility by having drivers be independent contractors, the Susan Fowler sexual harassment scandal, Uber’s “Hell” program to spy on Lyft drivers … why would we think Uber would change its tune and think of pedestrian safety first, over putting their cars into widespread use …


#20

Even that is an average of two accidents a year if you aren’t continually monitoring it. Not ready for prime time IMHO.