Self-driving Google cars: 300,000 miles, 0 crashes — if only your PC was as stable

Share This article

In a little over a decade, self-driving cars have advanced from “unlikely” in the ’90s, to “maybe in our lifetime” in the 00s, to today, where autonomous driving seems very likely in the next 10 years. A reliability milestone was reached this summer when the fleet of self-driving Google cars passed 300,000 miles with no accidents. Conveniently for the rest of the world, it’s a round number in metric distances, approximately 500,000 kilometers. Google has a dozen test cars on the road most of the time. There’s always a driver behind the wheel ready to take over and a bevy of instruments monitoring data feeds and the on-board software. Soon, Google says the self-driving car team will be a solo show with the driver-in-waiting only. It’s also adding different vehicles to the fleet, most recently a Lexus RX450h hybrid SUV (main photo).
Advances keep coming and not just from Google. Ford announced it envisions a self-driving car by 2017 (that is, in five years) building on technologies that are currently available. In-car machine vision systems in Europe can now read road signs to warn of, say, “Detour 1 Km”; it’s a bit harder in the US because states may use different road sign formats where Europe uses the same symbols. Google recently announced its Google Translate can read text from images which can only help with road sign recognition.

How close to autonomous driving?

How far are we from self-driving cars, also called autonomous driving or driverless cars? The car as chauffeur that drives you anywhere on any kind of roadway is beyond the horizon. The more limited but useful vision announced by Ford could well happen this decade. The car pilots itself on a limited-access road with no driveways, no pedestrians, no bicyclists, and no animals. The road might be the HOV lane of an existing highway. Driven and self-driving cars can share the same road. The road would need well-defined lane markings. Then it’s a comparatively simple matter, as self-driving goes, to adapt the available lane departure warning (LDW) and lane keep assist (LKA) cameras, and adaptive cruise control (ACC) radars. LKA currently nudges a car back into lane when it veers onto a lane boundary; a self-driving car would use LKA to recenter the car as soon as it’s no longer exactly in the middle of the lane. ACC keeps pace with the car in front.
Blind spot detection that alerts drivers to cars behind or alongside would be employed to tell the car if it could swerve into an adjacent lane in emergency. In this first stage, a driver would still be behind the wheel and would have to be paying attention, more or less, but couldn’t be napping. The cameras would have to see far enough ahead to know there’s a problem and give the driver a couple seconds to take over. Some automakers are using rear-facing cameras for LDW/LKA to save costs; that would have to change. Even if it works, it’s going to sound indefensible in court in the very first lawsuit: “Did you tell the court, Mr. Chief Engineer, that for cars moving forward under their own control, you have the cameras facing backwards?”
There are still many problems to solve. Google in a recent blog said it needs to build its ability to self-drive when snow covers road markings. The same would hold for fog and heavy rain. Not to mention highways with worn-out pavement markers. Negative terrain (a cliff next to a mountain road) remains difficult to sense. It’s much harder to drive on city streets and deal with cars pulling out of driveways, bicyclists, pedestrians jaywalking, traffic lights, and road restriction signs (no left turn 4-6pm).
Self-driving cars would not use some once-promising technologies. The 1990s idea that cars would drive themselves by following magnetic markers in the roadway is deader than Toad the Wet Sprocket. Magnetic guidepaths are working for warehouse automated forklifts and delivery carts.

Self-driving vehicles for the military and industry

It’s not clear if a vehicle can drive on highways solely by ultra-precise GPS fixes. For military vehicles in the desert, GPS holds promise. The same is true for mining vehicles. The military also wants self-driving vehicles with no humans aboard for battlefield situations — think drones on wheels or tracks. They don’t fall asleep and don’t require funerals.
Follow-the-leader self-driving hold promises on public roads and for the military. The first vehicle in a convoy could be driven by a human and the following vehicles would only need to follow the leader.

The green angle

The environmental angle may be the one that moves states to approve autonomous driving roads. In announcing Traffic Jam System, its initial self-driving effort, Ford says rush hour travel times could be reduced 38%, more cars can travel a roadway per hour, and fuel economy goes up. That’s even on a roadway with a mix of self-driving and human-driven cars. The self-driving cars are better able to modulate their speed, notice slowdowns up ahead, not not make sudden stops because of misjudged distances. They act as shock absorbers between driven cars. Automakers might be able to get higher EPA mileage ratings numbers on cars with something like Traffic Jam Assist. They already get EPA fuel economy ratings credits for things a vehicle might do, such as being able to run on E85 (85% ethanol, 15% gasoline) even if it never uses E85.

Source : www.extremetech.com