Monday, July 22, 2013

The drones are coming. Slowly.

Long-time readers of my blog will know the disdain I have for auto-driving cars, but this week I wanted to try to expand a bit further on the issues with them right now.
Look at the limitations of drone cars right now. We'll use Google as an example as they're one of the most advanced in this area. The cars cannot drive on a route that has not already been driven, catalogued and stored as GPS data by Google's engineers. (ie. it cannot navigate on its own). And when I say "catalogued" I do mean "catalogued" - the engineers have to enter every street sign, lamp post, intersection and road marking and 'teach' the computer what they all are. For every road. Ultimately that's why Google has Streetview - it's not so we have a fun way to look at the world, it's so they can collect this data for everywhere their car could ever drive.
It can't merge on or off freeways - that has to be done by hand. It has trouble in traffic that isn't single lane. It can't deal with unexpected events (unexpected debris in the road, for example) because it's not sentient - it's a programmed series of instructions that cannot adapt unless the code has been supplied for every eventuality. That in itself is an impossible task.
It can't drive if one or more of the cameras is covered (think wet leaves sticking to the windshield on an autumn day).
It can't drive in the snow because the world doesn't look like the pristeen clear, sunny day the car expects. It can't drive in heavy rain because the rain and road spray introduce enough data distortion into the cameras and radars that they don't work. More to the point, in heavy spray, adaptive cruise doesn't work for the same reason - radar doesn't work well in precipitation (think of the TV weather radar - it's reflecting off water, basically).
Can these problems be overcome? For the most part, probably. Adapting to changing and unexpected road conditions is the thing that engineers cannot code around. That's either going to require a very fast-learning AI, or it's going to become the achilles heel of drone cars. People do seem to have a hard time understanding why this type of project is so difficult. After all, we've had autopilot in aircraft for decades, right? Yes, but aircraft operate in a fundamentally different environment. TCAS systems talk aircraft-to-aircraft to avoid collisions. Aircraft don't find themselves nose-to-tail in the sky, they don't have to deal with pedestrians walking off the kerb, or cars, trucks, motorbikes, segways, street furniture, signals, illogical road layouts, faded paint or any of the millions of other variables that car and driver have to.
I'm not just picking on Google here. All the manufacturers who are trying this tech have the same problems to a greater or lesser degree. Volvo and Mercedes can't even get their auto-braking systems to work reliably, and unless it's bombproof, 100% reliable, nobody is going to want it in their cars. Hunt around YouTube for videos of the various public and very embarrassing tests where manufacturers have demonstrated their auto-braking systems mowing down cardboard pedestrians, driving through foam brick walls and in one very prominent case, burying the car under the back of a real truck with some considerable force.
If you still think I'm being pessimistic, think of this : drive-by-wire cars - in particular some Toyotas. That's a single system - a throttle. Not a complex series of computers, sensors, cameras and radars. It's a pair of potentiometers a length of wire and a servo motor. And yet this very system has been accused of causing 89 deaths in Toyota vehicles - a figure Toyota deny and have now been accused of covering up evidence about. If (and it's a big IF), it turns out that their unintended acceleration woes were caused by an electrical fault in something as simple as throttle-by-wire, and if they're spending millions covering it up, can you seriously imagine any manufacturer taking responsibility for deaths caused by fully autonomous drone cars?
I for one am not looking forward to a future where drone cars are the norm. In fact, as a daily driver and motorcyclist, I can't imagine a more bleak or dangerous place to live. However if you really are pining for the day when you'll be able to eliminate all personal responsibility and get in a vehicle that just drives you to where you want to be, I have good news. That day is already here. It's called a taxi. Or a bus. Or tram. Or train.
Footnote : Forbes did a short writeup on similar issues back in March - a one-page worthwhile read if you have another three minutes : Forbes and the Google car

No comments: