Last week I talked briefly about self-driving delivery vehicles. I've become interested in what I'll call the "roadmap" (pun intended) for the adoption of self-driving cars. Just how do we get there from here, taking the technology as a given? I've seen and thought of many proposals, and been ignoring the one that should stare us in the face -- delivery. I say that because this is the application the DARPA grand challenge is actually aimed at. They want to move cargo without risks to soldiers. We mostly think of that as a path to the tech that will move people, but it may be the pathway.

Robot delivery vehicles have one giant advantage. They don't have to be designed for passenger safety, and you don't have to worry about that when trying to convince people to let them on the road. They also don't care nearly as much about how fast they get there. Instead what we care about is whether they might hit people, cars or things, or get in the way of cars. If they hit things or hurt their cargo, that's usually just an insurance matter. In fact, in most cases even if they hit cars, or cars hit them, that will just be an insurance matter.

A non-military cargo robot can be light and simple. It doesn't need crumple zones or airbags. It might look more like a small electric trike, on bicycle wheels. (Indeed, the Blue Team has put a focus on making it work on 2 wheels, which could be even better.) It would be electric (able to drive itself to charging stations as needed) and mechanically, very cheap.

The first step will be to convince people they can't hit pedestrians. To do that, the creators will need to make an urban test track and fill it with swarms of the robots, and demonstrate that they can walk out into the swarm with no danger. Indeed, like a school of fish, it should be close to impossible to touch one even if you try. Likewise, skeptics should be able to get onto bicycles, motorcycles, cars and hummers and drive right through the schools of robots, unable to hit one if they try. After doing that for half an hour and getting tired, doubters will be ready to accept them on the roads. The robots (with no passengers) will be better at doing that than passenger self-driving cars because they can accept high gee-forces to avoid hitting something else, and can react immediately to the other robots doing so. They will plot probable trajectories for other vehicles and passengers and just never go where one of them could go if it suddenly alters course if they couldn't also get out of it. They will not rely on communications with their brethren but can take advantage of it to improve this ability.

I expect at some point it will even become a game for children to try to touch a robot. (Unfortunately they will also try to gum them up by throwing things at them, which the robots may not be fast enough to avoid or may damage fragile cargo.) And there will also be thieves to deal with who want to steal the robots or their cargo, though since the robots of course will have cameras that can transmit if they detect a theft attempt, this will not be trivial.

The delivery robots will be very patient. They will happily take the less busy, less popular streets so that they interfere less with other traffic (though this may bother people who live on streets that were formerly traffic free. However, the robots would never hit kids in a stickball game, and might well back up or go slowly and to the side if they encounter one. Patient as they are, they will also always get out of the way of a human-occupied vehicle, so that the humans will almost never find one slowing them down. You won't find yourself stuck behind a robot at a red light because, aware of the timing of the lights, they would almost never end up stopped at one, except because they slowed to get out of the way of a human. We probably can't attain this if traffic is very heavy -- but again they will mostly not be on roads with heavy traffic, sticking instead to side streets. Being narrow, they can share or split lanes.

An interesting question is, "what percentage of vehicles on the road are really just moving cargo?" I don't mean just UPS trucks, but things like people on trips to pick things up, such as parts from a store, or take-out food. How often do you travel when you don't need to "shop" (as in look at choices and select one) but just "get?" For these choices the robots will be the right path. And being cheap, many stores would own one, but since they can be dispatched from wherever they happen to be, they will also be cheap to rent.

Sadly, the fear of terrorism may sit in the way of this. They will be a fabulous bomb delivery tool, especially the larger ones. I'm not sure how to solve this and it would be a shame if this stands in the way of deployment.

Of course, after the delivery robots become common, the drive to let people ride them will increase. Indeed, somebody may also try to develop an ambulance robot which is able to get to the hospital far faster than a human ambulance can, for other reasons, using its ability to safely drive through intersections or on unusual places. If it's demonstrated the robot ambulance can get you to the doctor in half the time, how long until we accept it?

This is one of several possible pathways I will discuss in the future. Others include:

• Reserved lanes for self-driving cars on highways
• Special parking lots for self-parking cars
• Use of BRT right of way by self-driving cars
• Long haul robot trucking
• The same experiment described above with full sized cars, where skeptics try to hit or get hit by them, and can't, no matter what they do.
• Automatic throttle cars with a human driver but AI throttle (super fancy cruise control) which is synchronized to traffic lights