The original note, as written at the time
Think of a robot body which has a sensor attached to an omni-movable element on each and every point of its surface. These are what you can call robocells. Each robocell can be the size of a small coin. The combined data collection and element movement of each robocell creates the general movement of the entire robot.
Ongoing development
Think of a robot body with a sensor on an omni-movable element at every point on its surface. Each one the size of a small coin. The combined data collection and element movement of all of them produces the general movement of the whole.
This is the correct architecture and almost nobody builds it, because it is inefficient in the way that living things are inefficient and we have been trained to think that is a fault.
The usual robot has a skeleton, joints, actuators at a handful of points and a controller deciding what each should do. It is a machine. It moves the way a machine moves: precisely, in the directions it was designed for, and not at all in the directions it was not.
A surface of robocells has no joints. It has a skin that can ripple, grip, push and flow, and the movement of the whole emerges from millions of small local decisions rather than from a central plan. That is how a slug moves, how a tongue moves, how your gut moves. It is also why those things can squeeze through a gap that no articulated limb could negotiate.
The sensing and the acting are the same element, which is the deep advantage. Every point that can feel is a point that can push back. A body like that does not need a map of its own surface because the surface is the map, and it does not need to model the object it is holding, because the grip is distributed and local.
The engineering objection is real: a million coin-sized units with power and communication is a nightmare of wiring, heat and failure. But the objection is about our current manufacturing, not about the idea. And there is a version of this that is already here — a printed sheet with addressable cells — which is how you would prototype it tomorrow.
I want a machine that can hold an egg and a rock with the same hand and no reprogramming. That is a skin, not an arm.
Since this was written, the field has named the idea: electronic skins, distributed tactile arrays, soft actuation, materials that sense and push with the same element. It is one of the more active corners of robotics research now. What remains unsettled is exactly what I said would be hard — power and wiring, a million small units that must each be fed. The architecture is accepted. The manufacturing is not.