> The car that they put forward first needs to be non-threatening, safe, and easy to adopt.
Moreover, I think Google wants their car to be recognizable as a driverless car. They don't want it to blend in, because people would (at least for now) think that was "spooky"—that any car anywhere around them could be being secretly driven by a robot.
Instead, the visual distinctness means that it's instantly clear that these are their own thing, in much the same way that golf carts, scooters, and backhoes are—despite all being wheeled vehicles that can be driven in car lanes—their own thing. Because they're recognizable, it makes it possible for you to build a separate autonomous response into your subliminal "driving event loop" for the class of "driverless car", the same way that you will react separately to seeing any of the above in your lane. Which is exactly what people want, in order to feel in control: the ability to notice and compensate for the driverless cars around them if they all encounter some pathological edge-case.
(Just as an example of what I'm talking about—and this is a completely untested hypothesis on my part—I bet that driverless cars come to a complete stop a lot more frequently, in surprising places, because their sensors tell them something is standing in front of the car—e.g. a pigeon. A regular driver knows a pigeon will just sit there minding its own business as the car approaches it, but fly away right before the impact would happen; a driverless car might not know that. Because of this, your driving algorithm for being behind a driverless car will likely have a component similar to driving behind a school bus: "this vehicle makes frequent sudden stops, and I need to be ready to stop too. I should be more car-lengths away from it than are strictly necessary.")