Google’s report on its self-driving cars:
Most than half of the collisions reported in this document are slow-speed rear-enders of the Google cars. That’s unusual. It might be that the behavior of the Google cars is more cautious than what human drivers expect, so the Google cars stop more often abruptly or at unusual places, and so are not tailgater-friendly. I’d suggest that the Google cars might be equipped with a rear-facing warning device.
It seems to me that self-driving cars will be able to avoid any collision where a human driver could avoid fault, and others. In other words, operators of non-automated vehicles (including bicycles) and pedestrians who follow the conventional rules of the road will be able to operate safely around automated vehicles. Vehicles with automated crash avoidance (not necessarily completely automated vehicles, even) will not rear-end bicycles, and so the premise of fear from the rear evaporates if automated crash avoidance becomes universal with motor vehicles. Self-driving cars will not be able to avoid collisions where avoidance would require violating the laws of physics. Vehicles with automated crash avoidance will be able to avoid some collisions in which the potential colliding vehicle or pedestrian is outside the field of view of a human driver, such as right-hook collisions, as long as there is a clear sight line to the automated vehicle’s sensor. Same for a large truck’s high hood which prevents the driver from seeing a pedestrian crossing in front.
Automated vehicles will not be able to avoid left-cross collisions where the bicyclist or motorist is passing on the right of other vehicles and concealed by them, or pedestrian dart-out collisions. The concept of fully networked vehicles is supposed to address this problem. All vehicles approaching the same place in the road network are envisioned as communicating with each other even when they are hidden from each other’s view. As someone with an electrical engineering degree, I consider this at best a very difficult proposition, and it might be described as a pipe dream. Bandwidth, interference and reliability issues lead me to ask “what could possibly go wrong?” Also, instrumenting every object on the road is only practical on a limited-access highway — no, not even there, because there will still be breakdowns, wild animals, debris. On other roads, is every pedestrian going to carry a transponder? I don’t think so.
Automated crash avoidance is easily hacked by rolling a trash can out into the roadway, and the like. The caution which automated crash avoidance inherently incorporates changes the dynamic from the one among humans, which can involve a game of bluff. To me, this means that automated vehicles will be extra-cautious in the presence of other drivers and pedestrians who do play the game of bluff, and so the progress of automated vehicles will be slow and erratic in, for example, Boston traffic.
All this leads to the question: does behavior change as these vehicles become more common? Does infrastructure change? Every new technology takes a while to find its feet. As Marshall McLuhan said, “We look at the present through a rear-view mirror. We march backwards into the future.” Do conditions become better or worse for bicyclists and pedestrians? And why? We have some control over this depending on the direction which is set for the technology, but also, time will tell.
Another serious issue I’ve heard mentioned is the car which is not only driverless but passengerless. There is potential for an increase in traffic if a car can be called to meet a person (like a passengerless taxi), or directed to drive around and around the block empty when a parking space can’t be found. I can’t say how serious this problem will be. To some extent, that depends on the extent of freedom afforded to people’s control over the driverless cars. It’s an interesting legal question involving private use of public space. We already face this question with congestion-pricing schemes. But on the other hand, fewer cars on the road might be needed, because the car-sharing model works better when a car can be called rather than only stationed. Again, time will tell.