By on November 5, 2018

uber volvo

Are two safety drivers better than one when it comes to the testing of self-driving cars? Uber Technologies feels it is, declaring as much to Pennsylvania’s road regulator. The company has filed an application with the state’s department of transportation to resume testing of autonomous Volvos, eight months after a fatal collision with a pedestrian on a darkened Arizona highway.

Uber stopped all autonomous testing in the wake of the March 18th collision, with the Arizona program dismantled for good. In Pittsburgh, the company hopes to show it learned from the safety lapses revealed in the accident investigation. These Volvos now have two fail-safes on board. Is it enough to restore the public’s trust?

In its preliminary report on the events leading up to the death of 49-year-old Elaine Herzberg, the National Transportation Safety Board revealed that the self-driving XC90 “saw” the victim 6 seconds before the collision, but didn’t try to stop. It was also revealed that the safety driver was streaming a reality show on her phone, glancing away from the road in the seconds leading up to the crash. (The driver claims she was monitoring a seperate screen used to display vehicle functions). It was ultimately the safety driver who applied the brakes, but only after the collision.

Unlike the Arizona vehicles, Uber said the Pittsburgh fleet will not have their manufacturer-supplied automatic emergency braking system disabled to make things smoother for the autonomous drive system fitted to the Volvos. Uber was accused of dumbing down the safety of its test vehicles to avoid jerky, unnecessary braking. While the Tempe, Arizona Volvo recognized the looming victim, sources claim Uber’s drive system was programmed to ignore “false positives.” The elimination of the car’s factory AEB system made the situation all the more dangerous.

Then there’s the human element. With two safety drivers on board, Uber seems to feel that at least one pair of eyes will be on the road ahead at any given time. That’s a better setup than before, but in these Wild West early days of autonomous driving, nothing’s foolproof.

“Our goal is to really work to regain that trust and to work to help move the entire industry forward,” Noah Zych, Uber’s head of system safety for self-driving cars, told the Associated Press. “We think the right thing to do is to be open and transparent about the things that we are doing.”

Following the Tempe crash, Uber brought aboard former NTSB chair Christopher Hart to perform a review and change the company’s “safety culture.” In a lengthy report attached to Friday’s DOT application, Uber said it has introduced more rigorous training for safety drivers, among other initiatives. Pennsylvania officials have until Nov. 13 to approve or reject the application.

[Image: Uber Technologies]

Get the latest TTAC e-Newsletter!

Recommended

10 Comments on “Uber Eager to Restart Autonomous Testing, This Time With Two Safety Drivers...”


  • avatar
    SCE to AUX

    An important question is this: What SAE autonomous level is the Uber system supposed to be?

    If it’s Level 2, then that’s like all the other driver-required systems out there. In the Arizona accident (*if* it was a Level 2 system) and the myriad Tesla accidents (which actually are Level 2), you rightly blame the driver.

    If it’s supposed to be Level 4 or 5, then Uber becomes much more liable, and the driver isn’t even needed, let alone two of them.

    Adding another driver tells me that Uber’s system isn’t trustworthy today (regardless of its SAE level), and is years from being reliable. Moreover, the long-term business case for driverless systems has to be looking more bleak every day.

    • 0 avatar
      Snooder

      Of course it’s not safe today. That’s the whole point of it still being in “testing”. To test, find bugs and fix them. Possibly also to do some mapping and machine learning to “train” the car on how to respond in various situations.

      Think of it likenhaving a 15 year old in drivers ed for the first time. Would you trust them onnthe road by themselves? Hell no. But that’s why you have an adult sitting shotgun with a footbrake who can take over.

      It’s also obvious that the reason for 2 drivers isn’t to watch the car. You only need 1 guy for that. The second guy is there to watch the first and make sure he’s actually doing his job and paying attention instead of staring at his phone.

  • avatar
    conundrum

    The irony of having not one but two humans monitor the prototypical autonomous vehicle should not be lost on anyone. Why not fill all the seats with experts and be done with it?

    Or recognize that this AV was an idea whose practical implementation, with its depths of chaos and uncertainty, was massively underestimated by technical desk jockeys living in Silicon Valley.

    Since the usual argument brought up is that “but, but, but, airliners can land and take off autonomously”, I took a look, and it turns out, no they can’t. Plane to plane communication standards have not been agreed upon, and the human Air Traffic Controller is responsible for traffic which naturally gets congested around airports of any size. So planes cannot decide placement and hierarchy among themselves. On a test with no other aircraft around, sure the plane can land autonomously, but these new autonomous car prototypes can drive down a freeway or pre-plotted city course as well with not much problem. It’s when you add in all the complications, they falter, such as losing white lines at the top of a steep hill when the sensors are pointed at the sky. etc., leaves over the sensors in fall, snow, ice the list is endless.

    All these tech companies have invested billions trying to thread a needle with rope, and they’re gonna squawk and scream if they have to write off the investments made. So, we’ll be bombarded with justifications of dubious merit for years that we should embrace the future, save lives, blah, blah blah. What we really need is better and cheaper sensors and far more intuitive computer neural networks or whatever they call them. That’ll take decades.

    No sir, hubris made these techies think, “No prob!” a few years ago at the dawn of autonomous driving. But as they got into it, they found out it was a very big problem indeed. Their bosses with an eye on the $ won’t let them quit. Kind of like weed killers, insecticides and GMOs. Every box of breakfast cereal has a bit of glyphosate for your delectation, along with a dose of chlorine based insect killer. Excellent. Released on the market without due care, attention and trials looking back with hindsight. Now we’re about to get bizzaro killer vehicles, despite hand-waving like mine and others that there is NO rush.

    Let’s do it right.

    • 0 avatar
      SCE to AUX

      Very well said.

      And I’m sure it wasn’t the engineers who said “let’s add another driver”, but the lawyers. They’re starting to realize just how costly headlines could become.

    • 0 avatar
      stuki

      +1

      A major reason AI in complex environments has always seemed like it was just on the cusp of working, is that it obeys the ultimate 99-1 rule: The first 99% is easy. And can look quite impressive. Look, Ma, Asimo can walk! And smile! And kick a ball!

      And then….. The only known example of getting the last, necessary for any real usefulness, 1% to work, took about a billion years worth of evolution… Even if we could beat that current world record 100-1, that still leaves 10 million years. Of the whole darned world plugging away at it.

  • avatar
    tylanner

    It is beyond ridiculous to test this system on a massive SUV….there is no excuse for this…put it on a smart car and you’d really have to get lucky to kill a pedestrian…and you don’t really lose anything in testing…

    The Honorable Christopher A. Hart seems credible but its unclear what he found or resolved or what his charter was…but it looks like he is an Uber employee now…so we can guess where his incentives lie.

  • avatar
    TimK

    Again, the real question here has nothing to do with gadgets and technology, it has to do with accountability and legal recourse. What engineer, manager, corporate officer is going to sign-off on a system and send onto the roads vehicles that can autonomously kill people?

  • avatar
    brn

    The safety driver was 1% of Uber’s autonomous issues. Uber’s attitude toward risk is 99%.

  • avatar
    W126

    I guess next time Uber kills someone they can just doctor the video again to make it look it is pitch black and that no human driver could have avoided the collision. No one in the media or the police department will question a video even though it is put out by Uber, because video is regarded as infallible. Uber will then pay a “big” seven figure settlement to the family, which is a drop in the bucket for them, and they can repeat this process to their heart’s content. Just goes to show you can be the shittiest company around and not give a damn about your customers, their safety, or their privacy, but people will happily give you their money as long as you enable their laziness and have a nifty app. I guess calling a cab is just so much more arduous than using uber.

  • avatar
    Gedrven

    So let me get this straight: the way to improve the safety of a system supervised by one person staring down into their smartphone is to have two people staring into their smartphones? Or better yet, yakking with and otherwise distracting each other as their gizmobile plows into cyclists and concrete dividers, or panic stops for no apparent reason?


Back to TopLeave a Reply

You must be logged in to post a comment.

Recent Comments

  • legacygt: So for the same money you get a Mini Countryman All4. For about $3K more you get the Minis Cooper...
  • Corey Lewis: Thanks! Just had to find the right one for me. The black interior restriction was a serious issue.
  • 28-Cars-Later: “effectively means electrification kills jobs” This is an interesting point. “were...
  • Blackcloud_9: Good one, Dave. Corey, Congrats on your purchase. I’m sure you’ll be very happy with it. I...
  • JMII: I bought my 2014 C7 Z51 3LT Corvette in NJ and drove it home to FL last summer. Total was 1,207 miles in 2...

New Car Research

Get a Free Dealer Quote

Staff

  • Contributors

  • Timothy Cain, Canada
  • Matthew Guy, Canada
  • Ronnie Schreiber, United States
  • Bozi Tatarevic, United States
  • Chris Tonn, United States
  • Corey Lewis, United States
  • Mark Baruth, United States
  • Moderators

  • Adam Tonge, United States
  • Corey Lewis, United States