As noted, I’m working on the intersections between physics and A.I., and though I don’t spend much time thinking about robotics, it comes up now and then, since I think about complexity of motion or gesture, for example. And what I realized is, what you can probably get away with, in most practical circumstances, is a robot that can move, grab objects, and move objects –

This is not going to work for everything, but it will work for many practical real world problems, which usually involve only these things.

A robot comprised of a series of flexible, retractible joints, arranged as a panel, with retractable wheels below, to allow for motion, and grabbing.

Taking it to its extreme, just create a panel that is entirely comprised of powered, flexible joints, with wheels on the bottom. This will allow the panel to change its shape, into pretty much anything, by twisting, contracting, expanding, etc. And if you make the wheels extendible, then a group of wheels can together squeeze an object, operating as a grip, leaving the balance of the robot to make contact with the ground below, thereby allowing the robot to grab and move objects. My software can definitely power something like this, leaving the engineering questions of materials, motors, and electrical power. Note that adding vision to something like this is trivial.

You could even imagine making these capable of being combined, since they would be fungible panels. It shouldn’t take much to make them autonomous, which means they could even combine and separate on their own, based upon the task at hand, allowing for even greater flexibility in problem solving. For example, a group of small panels could break off from the whole, to tackle a problem that requires small scale manipulation. This could allow a large number of small panels to adjust the position of a complex object, that would, for example, then allow for the object to be grabbed by the entire unit, post reassembly. Or panels could partially separate, allowing for the emergence of appendages.