Slashdot Mirror
Google's Driverless Car and the Logic of Safety
mikejuk writes "Google's driverless car could save more than 1 million deaths per year and tens of millions of injuries. It is an impressive achievement, but will we allow it to take over the wheel? Sebastian Thrun puts the case for it in a persuasive TED Talk video. However it may be OK for human drivers to kill millions of people each year but one human fatality might be enough to finish the driverless car project — in fact it might not even take a death as an injury might cause the same backlash. Robot drivers might kill far fewer people than a human driver but it remains to be seen if we can be logical enough to accept the occasional failure of algorithm or hardware. Put simply we might have all seen too many 'evil robot' movies."
I don't think people realize just how automated their vehicles already are. Sure, it's nice to be able to point to something and go "It parks itself! Ohmigawd!" but if you dig deeper you'll realize that the beginning of the "cars driving themselves" era has already passed us by. Thirty years ago when you mashed the brakes in your car, it pushed on a hydraulic, vacuum-assisted cylinder, and forced a fluid down to the brakes. That's it.
Now when you nail the brakes, a computer is deciding that the "rapid engagement of the brakes" is really a request for 100% braking power and fully actuates the master cylinder by itself regardless of your exact input. Some cars will even adjust your steering inputs for you. Meanwhile another computer is looking at the rotating speed of each wheel, comparing them, and reducing and/or modulating the pressure to keep them from locking up. Another computer (or maybe the same one) is checking the speed of all four wheels versus the angle of the steering wheel versus roll/pitch/yaw sensors, and further adjusting the brakes and engine torque split to ensure that the vehicle isn't spinning or attempting to roll. Yet another computer is seeing that a massive load is being placed on the front suspension and actuates a set of valves or magnets to firm up the front shocks to reduce braking dive. Meanwhile a front facing sensor is comparing your rate of deceleration with the speed at which you're approaching an object, and when the check fails it weighs each occupant and primes a series of airbags for them, fires the seatbelt pretensioners, unlocks the doors, brings the seats upright, rolls up the windows, closes the sunroof, disables non-essential electrical systems, and basically does it's best to prepare the cabin for a crash. Some cars even have microphones tuned to listen for the sound of impact as a queue for firing the airbags! And how many cars these days phone home (OnStar, etc) when you're in an accident? You smash into a tree and before the fog clears from your eyes there's a friendly sounding lady on the phone going "We've detected a crash. Sir, are you alright?"
Cars already drive themselves. We just point them in the direction we want to go. One day we won't even have to do that, we'll just say "take me home" and it will figure out the rest. Why that is so much more terrifying than our present state is largely a matter of perception.