In a rather lengthy twitter conversation on the viability of the driverless car, there was some skepticism expressed about the idea that driverless cars would interact with pedestrians and cyclists.
There’s already been a good deal of fairly impassioned discussion on the topic of how driverless cars will interact with pedestrians.
I agree with the conclusion in David Alpert’s post:<
Driverless cars are sure to lead to big fights. Will they shift the balance farther toward pedestrians, as Kevin Drum believes, or away? I hope the former, but the technology won’t magically solve this problem. Instead, we’ll have to fight it out through the democratic process, as we do most other issues affecting the public sphere.
However, I don’t think there’s much informed doubt expressed that public policy decisions about the interaction of pedestrians with vehicles, including the status quo, could be implemented in robocars. Even at the status quo, it is highly likely that pedestrians will be much safer than with conventional vehicles, as robocars won’t speed, will have faster reaction times, are unaffected by glare or other vision problems.
As best I can tell, nobody, at least in the public domain, has explicitly considered this question with respect to cyclists.
Google’s driverless car has undoubtedly passed a few by now, so they must have thought about it, but they’re not saying anything yet.
In short, I do not see any reason why cyclists and driverless cars would not integrate at least as well, and likely much better, than cyclists and conventional vehicles.
Identifying the presence, speed and direction of a cyclist on the road is not a fundamentally difficult problem, given the sophistication of the sensors available to a driverless car. It’s worth noting, again, that a driverless car can have 360 degree vision, unaffected by the blind spots that afflict human drivers. Similarly, identifying and respecting bike lanes – and leaving an overtaking gap on roads without them – is a trivial extension of the myriad other problems that a viable driverless car will have to solve.
One of the more difficult challenges that a driverless car faces when dealing with pedestrians and cyclists (and unleashed pets for that matter) is determining not only where they are and their current trajectory, but what they might do next. In principle, it shouldn’t be hard to build a system to identify cyclist’s hand signals, but in any case cyclists do not universally follow such signals. There are other cues that the system could potentially pick up on, including positioning on the road, sudden decelerations, even the direction in which a cyclist is looking. This kind of thing may sound complex – and it is – but it doesn’t have to work perfectly. If there’s uncertainty, the car can simply slow sufficiently that if the cyclist does end up on a collision course it can stop in time.
A related question is how a cyclist divines the intention of a driverless car and determines whether the car has “seen” them; making eye contact with a driver is something we all do (whether driving or riding). In my view, that is unlikely to be much of a problem – the driverless car will see you. Furthermore, the most dangerous situations for cyclists are situations where either the cyclist or the driver has no opportunity to make eye contact anyway – getting hit from behind, or hitting a car, for instance, in a “dooring” situation or when a car pulls out from the curb without seeing the cyclist. I’d finally note that driverless cars are more likely to behave predictably, with fewer sudden bursts of hard acceleration, changes of mind in the middle of intersections, and so on.
There are a couple of ways in which driverless cars could improve cyclist safety further. One dangerous situation for cyclists is when “undertaking” in a cycle lane in heavy traffic, heading into an intersection where cars cross the cycle lane to turn left. While a driverless car with rear-facing sensors is better placed to detect the cyclist than a human driver, they still may be obscured by other traffic. However, if driverless cars were equipped with vehicle to vehicle data links (V2V is the favoured acronym, apparently) cars detecting the cyclists could pass that information on to other cars in the vicinity.
A second piece of infrastructure which may help is equipping bicycles with a V2V transponder. At a minimum, it would avoid the problem of non-detection of bikes. However, if the transponder could produce visible or audible warnings, it could warn the rider of cars on a collision course. Such transponders would be very cheap by the time driverless cars became commonplace (less than $50 is my guess) and could easily be integrated into cycle computers (or Google Goggles, which may be quite common by the time this becomes a live issue) which would reduce the cost even further. For those worried about mandating such systems, it’s worth pointing out that we mandate that cyclists have to have front and rear lights if they wish to ride at night; a half-decent set of lights easily runs to that, or more.
None of this goes to whether cyclists will be given higher or lower priority on the roads than they currently are; that, as noted earlier, is a policy question. But at a minimum, I don’t see any technical impediment whatsoever to implementing the status quo, with more safety and much less harassment for cyclists than is presently the case.