While thousands of independent force-pixels allow the powerful combinatorics of a high-dimensional space, for all kinds of deep reasons proprioceptive vibrations must be digested and their 3-D calculations performed in as low-dimensional a medium as possible, that is in their original 3-D spacetime form. Of the three different 3-D coordinate systems available, a cylindrical one best matches the symmetry of spinal control. In it force-geodesic stripes correspond to splines, existing either on the surface of a cylinder of muscle fibers, or threaded through a thicker 3-D mat of them.
Describing a curve of muscle activation on a cylinder takes two variables (z, phi(z)), and describing the curve through a thick mat takes three (z, phi(z), r(z)), regardless whether the reference frame is a true cylinder, a twisty one, or even a multi-scale one. But to describe the curve of the spinal bones in both cases takes a different three variables (z, phi(z), theta'(z)).
These principles (force-geodesics and optimally-symmetric reference frames) apply even as skeletal structures become more elaborate. One can imagine a long slow series of evolutionary hardware upgrades as various new skeletal features appear atop the spinal blueprint: first add ribs and lungs, then a head and tail, then fins, then legs, then hands, then walking. Each of these contains new structures and new symmetries, and hence requires new and more detailed reference frames (or eigenspaces) operating alongside the old ones. But because in evolution new features appear incrementally over time, each new frame must be built out of the previous versions and subservient to them. Hence even in bipeds, spinal and respiratory computations are more elemental than dexterity.