Uberized auto cars is still a long ways off, if you can believe this report.
Internal report says its autonomous cars can't even go a single mile before a person has to take control
by David Booth | 5 hours ago
Here’s the reality behind all the hype you’re reading about self-driving cars: Uber’s autonomous test fleet drove 32,000 kilometres last year and, in almost every single one of those kilometres, the minder in the driver’s seat had to take over the wheel.
Officially, the statistics are that Uber’s active robotic cars drove 20,354 miles and, according to Recode.net, which obtained some internal performance documents, “during the week ending March 8, the 43 active cars on the road only drove an average of close to 0.8 miles before the safety driver had to take over for one reason or another.” In other words, Uber’s vaunted self-driving cars couldn’t get you safely to the closest grocery store without one of those pesky carbon-based humans having to take back control from the computer.
The reason for taking over command and control, says Recode, “can include navigating unclear lane markings, the system overshooting a turn or driving in inclement weather.” What it doesn’t include are “critical interventions,” those situations where the safety driver had to take over to prevent harmful situations, which the company defines as the car hitting a person or causing more than $5,000 in property damage. According to Uber’s own numbers, that was once in every 320 clicks. Essentially, Uber’s cars would have hit two people or had two accidents (were it not for we lowly “safety drivers,” of course) driving from Toronto to Montreal or Vancouver to Nelson.
Even the robot’s ability to drive smoothly is called into question by Uber’s own statistics. According to Recode, Uber “auto-detected ‘bad experiences'” — hard deceleration or abrupt car jerks — every three kilometres or so. This was especially bad in Phoenix along the city core’s trendy Scottsdale Road, where the (not) taxis could barely get two-thirds of a mile before a lowly human had to intervene.
What makes this so very interesting is that the reason Uber moved to Phoenix from its original test facility in San Francisco was California’s insistence that it report all such “disengagements” as a condition of being issued a permit to test its autonomous automobiles on public roads.
RELATED Self-driving cars approved for 10 U.S. testing sites
According to the author of the Recode report, Johana Bhuiyan, other suppliers and manufacturers have a little better record than Uber at keeping their cars handsfree. Cruise, the self-driving startup now owned by General Motors, reported that drivers had to take over an average of once every 85 km, Nissan once every 240 clicks and Ford only once every 300. Reasons for the interventions ranged from the potentially fatal — the software module froze or the vehicle was about to hit another car — to the innocuous — the driver simply wasn’t comfortable with the visibility of the lane demarcation lines.
This last, relatively easy point in sunny Arizona, one would think, points to the challenges that self-driving cars will face here in the Great White Frozen North. At the very same time that Uber was suffering its travails in sunny Arizona — where the weather is so uniformly fair that it’s an inside joke amongst air force pilots to be grounded due to “bad weather” — I was in a luxury sedan with semi-autonomous features driving to Ottawa to take care of my folks.
On the way to the nation’s capital, conditions were Phoenix-like sunny and bright (though not quite as temperate, of course). The car’s robot had little trouble navigating between the 401’s lines. The straights were handled with unwavering aplomb and corners with steady, gentle arcs. Oh, I had to intervene every 15 seconds or so with a gentle tug of the wheel (automakers disable their current systems automatically to make sure you’re paying attention), but in general, the system was pretty reliable. The only emergency “disengagement” was when an 18-wheeler abruptly moved over with feet to spare, and I didn’t want to wait around to see if the computerized pilot would get with the program before I became one with a Peterbilt. Nonetheless, the system was fairly impressive.
On the way back, however, Ontario was hit with the snowstorm that wasn’t — we were threatened with 30 centimetres; only about two arrived. But even that light dusting sent the autonomous systems into a tizzy. Turns now became a nightmare of delay and over-correction, while navigating the straight sections felt like a drunken sailor weaving down a sidewalk after three nights of shore leave. Eventually, the system just shut down, a few flakes of the powdery white stuff enough to overwhelm the camera systems trying to guide its path. I should mention again that this was no whiteout blizzard, just a light dusting of snow that did not cover the lane markings — I could see them clearly despite my 59-year-old, I’m-probably-due-for-a-new-prescription eyes. How such systems will deal with a real snowfall is hard to fathom.
So why then all the hype surrounding self-driving cars? Perhaps more importantly, why are the monies being spent on autonomous technology so outrageous? Intel, you may have read, recently paid US$15 billion for Israeli self-driving tech supplier Mobileye (the company, you may remember, that abandoned Tesla after its over-hyping of the Model S’s Autopilot system). You might think that’s because CEO Brian Krzanich sees the self-driving industry growing to US$70 billion by 2030. In fact, there may be an even bigger reward.
RELATED Intel buys Mobileye, instantly becomes giant in autonomous car arena
“Data is the new oil,” says no less than the Financial Times; “who’s going to own it” determining the corporate haves and have-nots of the future. Between cameras (collecting between 20 and 40 megabytes per second), sonar (10 to 100 KB/sec) and Lidar (10 to 70 MB/sec), Intel estimates that autonomous vehicles will collect 4,000 gigabytes of data every day. Multiply that by the roughly 300 million cars on North American roads today and you have – well, my calculator won’t actually go that high, but you have a whole whack-load of data to sell.
And finally, what makes these revelations all the more interesting is the continued ambivalence that consumers profess towards the prospect of the future’s autonomous automobiles. Study after study (the most recent by Deloitte indicating that fully three-quarters of Americans “believe fully autonomous vehicles will not be safe”) reveal a populace still strikingly uncomfortable with the prospect of a computer assuming complete control of its car.
Makes you kind of wonder what the motivation behind all this automation really is.
driving.ca
|
