The proposed Google self-driving car, once on the road, presents problems beyond the mere problem of connecting a car's controls to an internet map, and avoiding collisions with stuff in its way, or stuff approaching it at a closing velocity.
For it to work, the maps must be in place, and the GPS connection must be available, and there must be a connection to a satellite, so that the vehicle can be operated remotely. The operation of the vehicler is, in this case, a computer making decisions, and sending the decisions remotely to the vehicle controls.
Presumably, the operator supplies the destination address to initiate the travel, and there is probably some sort of log-on process.
Is it at least possible, or conceivable that the destination signal could be input into the vehicle remotely, as well?
Sure. One of the imagined benefits of a self-driving car could be to call your car, from someplace else, to come get you. You could have the car parked a short three blocks away, and have it come to your door so as to avoid walking in the rain.
What if somebody else, not the owner/operator could run the vehicle somewhere, using owner's identification and password, or even through some backdoor conveniently left available for government law enforcement.
I think what you have, then, is an unmanned drone.
Based on the recent revelations about government surveillance . . . can we trust that it would never happen?