SoylentNews is people
SoylentNews
Waymo has announced a driverless taxi service called Waymo One, but it will only be usable for around 400 preapproved "early riders" in the Phoenix metro area, rather than the general public. While self-driving Chrysler Pacifica hybrid minivans will be used, they will continue to retain a safety driver behind the wheel.
Waymo's "new" service could be described as a launch in name only:
The banner Waymo is unfurling, though, is tattered by caveats. Waymo One will only be available to the 400 or so people already enrolled in Waymo's early rider program, which has been running in the calm, sunny Phoenix suburb of Chandler for about 18 months. (They can bring guests with them and have been freed from non-disclosure agreements that kept them from publicly discussing their experiences.) More glaringly, the cars will have a human behind the wheel, there to take control in case the car does something it shouldn't.
So no, this is not the anyone-can-ride, let-the-robot-drive experience Waymo and its competitors have been promising for years. Building a reliably safe system has proven far harder than just about everyone anticipated and its cars aren't ready to drive without human oversight. But Waymo promised to launch a commercial service sometime in 2018, it didn't want to miss its deadline and risk its reputation as the leader of the industry it essentially created, and not even the might of Waymo parent company Alphabet can delay the end of the calendar year.
So Waymo is pushing out a software update, tweaking its branding, and calling it a launch.
Also at Reuters, Gizmodo, The Atlantic, and Ars Technica.
See also: Waymo's driverless cars on the road: Cautious, clunky, impressive
Previously: Google/Waymo Self-Driving Minivan Tested with the Public in Phoenix AZ
Waymo Orders Thousands More Chrysler Pacifica Minivans for Driverless Fleet
Walmart and Waymo to Trial Driverless Shuttle Service in Phoenix for Grocery Pickups
Google's Waymo Plans to Launch a Self-Driving Car Service in December (the service falls short of what is described in this November article)
(Score: 1, Informative) by Anonymous Coward on Thursday December 06 2018, @09:56AM
The specific reasons for crashes don't matter in the slightest if they kill less people per mile. This only matters if you think a) we won't be able to beat human safety levels without robust recognition (plausible) and b) we won't be able to achieve robust recognition (doubtful). Let's consider b further since I don't take issue with a.
Neural networks are universal function approximators, either they can do robust-as-humans recognition or the physics ruling human brains is uncomputable. The question isn't whether it's possible, but rather whether we can practically train them to be robust. Showing that networks not trained to be robust aren't robust doesn't mean that they can't be practically trained to be, it just means that robustness isn't automatic with current training techniques and we need to design for it. I don't see any reason to believe it's not practical to train it to recognize the class ``unclear signage'' once that's a design goal and sufficient training data is accrued.
If these cars are being rolled out then I expect they're robust enough for the test data (though they probably didn't test in neighborhoods covered in graffiti/snow/mud/&c). This particular implementation may not be sufficiently robust, but sooner or later someone will collect enough data and find the right techniques. This is a difficulty, but not a fundamental limitation.