Opinion: autonomous cars will make traffic *worse* before it gets better

Many things bug me about the prospect of autonomous cars. I bring those concerns to researchers and developers, who calmly reassure me they’ll eventually solve most of them. But there’s one issue that no engineer will address. Or at least none that I’ve met. They just laugh it off.

Here’s the theoretical example I give them. On my pushbike in London, if I need to cross the traffic, I know it’d be insane to swerve in front of moving vehicles. But what if in the future I saw an autonomous car? Clearly I could pedal across in front of it, trusting it to slam on its brakes to avoid me. Now extend the problem’s scope. I could do the same if I were driving, knowing I could happily barge in front of a robo-car, because it’d be programmed to proceed with saintly defensiveness. Pedestrians too would treat autonomous cars in the same way: no more waiting patiently for the green man. Prankster kids would dawdle in the middle of the road. Which means all the cars behind the autonomous vehicle would grind to a halt too. Traffic chaos ahoy.

Now here comes an autonomous driving researcher addressing the issue. And he’s indicating that yes, simulations predict that traffic flow will likely get worse when traffic is a mixture of autonomous and normal cars. This is Nils Berkemeyer, a director of self-driving AI company Autobrains, into which BMW and various system suppliers have invested. Berkemeyer is bound to be a robo-car cheerleader, yet he admits the problem.

However, he reckons it’s not as simple as the modellers – and grumpy Top Gear columnists – predict. For a start Berkemeyer says that if autonomous cars prove safer (and if they don’t prove safer they’d surely be banned pretty darned sharpish) then we’ll see fewer accidents and fewer of the snarl-ups that result.

He also reckons that traffic will flow better because autonomous vehicles drive more smoothly. This means they have a dampening effect on what’s behind them. One well known property of a traffic stream is that one driver accelerating and braking sets up a standing wave that brings things to a halt some way behind them. Robo-cars would reduce that.

He also says that the robo-cars won’t be mixing it entirely with “dumb” cars (aka ones driven by intelligent humans) but with others that have at least Level 2 assistance systems so there will be sensors on every car, and they can cooperate.

I get his point about the lowered accident rate, but others don’t cut much ice, at least except on motorways, where I don’t live and I hope you don’t either. Where humans live the local traffic is jammy, and it’s mixing with people on bikes and on their feet. Cyclists don’t have Level 2 assistance. Pedestrians dodging across roads mean the vehicles have to stop, so flow becomes lumpy.

This autonomous vehicle maven concludes that “mixed traffic will not be quite as big a challenge as today’s simulations would have us believe”. But still a challenge then. For which read more gridlock, getting worse before it gets better.