Self-driving cars? No so fast, thanks

I’ve long been intrigued by autonomous vehicles, or more specifically “self-driving cars.” Being no expert in AI or robotics, I can’t estimate when or if they would be practical without a fully attentive human driver.

That said, I’ve just enough common sense to question. Don’t like to be skeptical of technological progress, but in this case maybe we’re jumping on an exciting bandwagon, promising too much too soon. Twenty years? No way. Fifty, maybe.

I quickly jotted a list of more than 200 situations that would logically have to be recognized and accommodated. Examples: defaced/obscured/missimg signage; ruts and potholes; objects in road; moving around garbage or delivery or mail trucks; cops’ and workers’ hand signals overriding lights; driving into sun; funeral processions; high water; cars rolling through four-way stops; lane closures; checkpoints; accident scenes; emergency vehicle horns and sirens and flashing lights…

And these are just one-dimensional situations. There are hundreds of complex combinations. One of the reasons the Air Force wants to keep pilots, and that man is destined for space despite the hostile environment, is that “man is receptive to the unanticipated.”

Even proponents caution that there’s lots of work to be done to accommodate situations and develop sensors. But they stress the sensors; the rest of the systems and the environment in which they must operate is seldom mentioned.

Yes, we need the sensor technology: reliability, operating range, vibration and temperature resistance, etc. Do they work when snow- or mud-covered? At night? In fog? Rain? At night in rainy fog? And sensing is only part of the system. There’s the transducer, the computer, the actuators, and the feedback to be reliably designed as well, and the communication among many such systems.

But we also need to better know the real (not fantasy or sterile cartoon) world we have to sense, the complex environmental part.

Most tests have been done on test tracks or well-maintained highways in good weather. Then proponents merely conquer one simple one-dimensional hurdle at a time, over-optimistically extrapolating to all situations.

Then there’s the demanding behavioral and even moral decision-making. Can they be polite or recognize courtesy? Do they let someone merge? Do they recognize a smiling wave to go ahead? Do they respond to road rage? And many situations require decisions between the “lesser of two evils.” One source’s example: If an animal (how big is that beast?) jumps in front it might be better to hit it than run into a ditch at speed (how deep is that ditch?), but if it’s a child head for the ditch. Does the car even sacrifice itself? Does it stop after a small bump?

And where does it stop, and what happens when it does?

Realistically, I think the best we can do for decades is allow some vehicles to drive autonomously sometimes in some places. Maybe some trucks on interstate highways between cities without orange barrels in good weather. But even then the required Teamster-member driver would have to be crazy to lie down and take a nap.

About the Author