Ubers use of fewer safety sensors prompts questions after Arizona crash
Source: Reuters
TEMPE, Ariz./PITTSBURGH (Reuters) - When Uber decided in 2016 to retire its fleet of self-driving Ford Fusion cars in favor of Volvo sport utility vehicles, it also chose to scale back on one notable piece of technology: the safety sensors used to detect objects in the road.
That decision resulted in a self-driving vehicle with more blind spots than its own earlier generation of autonomous cars, as well as those of its rivals, according to interviews with five former employees and four industry experts who spoke for the first time about Ubers technology switch.
Driverless cars are supposed to avoid accidents with lidar which uses laser light pulses to detect hazards on the road - and other sensors such as radar and cameras. The new Uber driverless vehicle is armed with only one roof-mounted lidar sensor compared with seven lidar units on the older Ford Fusion models Uber employed, according to diagrams prepared by Uber.
In scaling back to a single lidar on the Volvo, Uber introduced a blind zone around the perimeter of the SUV that cannot fully detect pedestrians, according to interviews with former employees and Raj Rajkumar, the head of Carnegie Mellon Universitys transportation center who has been working on self-driving technology for over a decade.
Read more: https://www.reuters.com/article/us-uber-selfdriving-sensors-insight/ubers-use-of-fewer-safety-sensors-prompts-questions-after-arizona-crash-idUSKBN1H337Q
That Arizona woman's family should be able to win a lawsuit: "Velodyne acknowledged that with the rooftop lidar there is a roughly three meter blind spot around a vehicle, saying that more sensors are necessary."
FrodosNewPet
(495 posts)Apparently even THEY don't have an infinite source of venture capital, and they are burning through cash by the pallet.
Marie Marie
(9,999 posts)decides, this early in the game, to cut corners. The more I hear about Uber, the less I like them.
Egnever
(21,506 posts)and also not surprising actually.
I don't think this is the actual cause.
Even with the three foot blind spot it should have seen her.
Gore1FL
(21,130 posts)That's a little over a 9.8 ft blind zone around the perimeter. I don't know what size of blind zone the 7-lidar vehicles had but I must admit 3 meters seems jaw-dropping to me!
Read it as feet. Still even at 9 feet it should have seen her.
I don't think breaking in 9 feet at 40 MPH is feasible regardless.
From the video she looks to be half way across the road when hit. That would seem to imply she would have been visible far further than 9 feet as she started to cross.
Maybe not to the human eye but in the Google cars they claim to see somewhere in the neighborhood of 300 yards.
Perhaps the lidar is not nearly as good in the Uber cars. More than likely is not. Still even at half the range it should have seen her.
Moving at 40 MPH is almost 60 feet a second it had to take more than a second to cross the lane she crossed before being hit.
From my understanding the car did not break meaning it did not see her.
It should have even with a 9 foot blind spot
pnwmom
(108,977 posts)and that it apparently completely missed seeing her. And yet . . .
ToxMarz
(2,166 posts)The Uber self driving car struck and killed a pedestrian crossing the road.
These are as yet not connected and my be either the cause (which you are assuming), contributory, or unrelated to the actual cause.
The take away is just that they have serious problems.
pnwmom
(108,977 posts)They also said that the position of the lidar on the top of a taller vehicle made a difference too.
I read feet instead of meters.
Absolutely a ridiculously large blind spot. Still she should have been outside of that blind spot when she entered the road for the driver side and crossed a lane before being hit at least for a bit.
I think something else went wrong.
gyroscope
(1,443 posts)destroys the aesthetics and aerodynamics. the added weight and drag in turn brings down fuel efficiency. the thing is I don't think the eyesore can be minimized much, without sacrificing safety, because the sensor array and lidar have to be mounted high on the roof in order for the car to adequately see everything around it.
Which is probably why Waymo used a minivan for its SDC platform since minivans have a higher roofline. Since minivans have a boxy, utilitarian appearance anyway, having all that extra stuff on the roof doesn't hurt the appearance as much. Car companies spend billions of dollars to make their cars look good, as well as aerodynamic, etc. But with SDCs all that may go out the window.
pnwmom
(108,977 posts)And one of the problems, according to the article, was that the sensor on the car was too TALL, because of the height of the vehicle.
gyroscope
(1,443 posts)it wasn't tall enough to see over the roofline of the SUV.
they may have attempted to improve/streamline the appearance of the sensor array by lowering it which of course sacrifices the effectiveness and safety of the system.
pnwmom
(108,977 posts)And, judging by the photos, the lidar is positioned at the edge of the front part of the roof -- so it shouldn't have had trouble seeing "over" the roof. It's too soon to know what problems they will discover, but the change of vehicle might have added to the problems.
A photo is here:
https://www.theverge.com/2016/8/18/12541672/uber-volvo-partnership-autonomous-self-driving-car
DeminPennswoods
(15,286 posts)Even if the radar did not detect her, the person in the car should have been doing what all good drivers do, paying attention to the road and continuously being on guard for something to unexpectedly pop up out of nowhere. I'd speculate that the person in the car was simply trusting the technology and not paying enough attention to the surroundings or the road.
pnwmom
(108,977 posts)didn't add to safety. That's why the move has been to driverless vehicles. There is too much lag time between even a good back-up driver's response to a vehicle's error.
These vehicles aren't being marketed with the idea that people in large numbers will want to buy cars that run themselves while humans have to sit there, just as engaged as ever.
bucolic_frolic
(43,149 posts)Exactly. It's like a correction to a correction. How does the driver pay close attention knowing that the vehicle is driving itself?
An actual driver might well have seen the woman before she entered the roadway. An actual driver could anticipate, and slow down, or lightly brake, enabling him to stop before the fatality, or lessen the impact.
Computers are running several ETF stock portfolios. They trade better than humans. But thus far they can't make judgments about what's happening in three months. So like real human drivers, portfolio managers aren't dead yet.
DeminPennswoods
(15,286 posts)The person sitting behind the wheel is the backup system. No matter how good the technology may or may not be, it's still experimental. The human driver was irresponsible and should have taken control of the car as soon as he spotted the pedestrian instead of, presumably, trusting the car to detect her and stop itself. It's not that different from a driver's training instructor sitting in the passenger seat and having their own brake to stop the car if the student driver errs.
pnwmom
(108,977 posts)don't increase safety and can't react fast enough in most situations to correct machine error. They don't know the error is going to happen till it's happened -- and then it's usually too late.
For example, this study at Duke:
https://slate.com/technology/2018/03/safety-drivers-attention-spans-might-slow-self-driving-car-progress.html
Research conducted by Missy Cummings, a Duke University professor who studies interaction between humans and unmanned vehicles, found that humans have trouble staying focused when expected to monitor an autonomous system. Cummings studied 27 subjects in a four-hour-long driving simulation and found that, on average, their vigilance decreased after just under 21 minutesa physiological phenomenon known as the vigilance decrement. (In other studies, its been documented as occurring between 10 and 30 minutes.) After this period, driver performance worsened: That is, drivers were more likely to deviate from their lane, use their mobile phone, and cross road edge lines. Self-driving car operators typically work in eight-hour shifts, well beyond that 21-minute good attention span. This observation could foretell problems with semi-autonomous cars and trucks.
The bottom line is that humans are terrible babysitters of automation, which is why we are in such a dangerous period of time with good, but not great, autonomous cars, Cummings told Slate in an email.
LiberalLovinLug
(14,173 posts)"it also chose to scale back on one notable piece of technology: the safety sensors used to detect objects in the road"
Probably THE most crucial aspect of the technology, and the most concerning aspect from the public's point of view. Not to mention avoiding the kind of business backlash if anything like what happened, happens. Who would think this was a good idea?