The Spec Is Already Written

Human glabrous skin runs four principal classes of mechanoreceptor, and each owns a separate mechanical channel. Merkel cells, the SA-I class, are the static channel. They fire continuously under sustained pressure, from DC to about 5 Hz. Meissner corpuscles, RA-I, report change rather than load, across 5 to 50 Hz. Pacinian corpuscles, RA-II, cover 50 to 1000 Hz with peak sensitivity at 200 to 300 Hz. Ruffini endings, SA-II, read skin stretch and sustained shear from DC to about 10 Hz. Four classes. Four jobs. Four bandwidths. One hand. The published physiology is not a research target. It is a channel allocation table with the arguments already settled. No single artificial transducer class covers all four channels. Piezoresistive and capacitive materials give you the static ones. Piezoelectric and triboelectric materials give you the dynamic ones. A skin built on one material has already dropped two of the four.

The densities are published, and they are not uniform. On the fingertip, Merkel cells sit at roughly 100 per square centimetre. Meissner corpuscles sit near 50, Pacinian near 5, Ruffini near 10. On the palm the same four channels fall to about 10, 15, 3 and 5. On the forearm, about 5, 8, 2 and 3. The whole hand carries roughly 17,000 mechanoreceptors. Dense where the work happens, sparse where it does not. That gradient is the answer to the resolution, coverage and wiring problem, and it was reached without a bill of materials. A hand that runs the same taxel pitch everywhere has spent its wiring budget on the forearm. Read the four columns and you have four taxel densities, not one.

Receptive field size is the second published number. Merkel cells carry fields of about 2 to 5 millimetres with sharp borders. Meissner corpuscles carry 3 to 5 millimetres. Pacinian and Ruffini fields are large and diffuse, covering an entire finger. A receptive field is a spatial filter. Its width decides what the channel can and cannot report, and no amount of downstream processing recovers detail the filter never passed. The fingertip resolves two points 1 to 2 millimetres apart. That limit is set by Merkel spacing, not by nerve count. The engineering translation is a taxel pitch. Piezoresistive or capacitive arrays at 0.5 to 2 millimetres reproduce the SA-I channel. Anything coarser is a different channel carrying a different claim.

The band edges are the part most specifications leave vague. The biology is specific. DC to 5 Hz is the static channel. 5 to 50 Hz is the flutter band, and it is where slip begins. 50 to 1000 Hz is tool-mediated vibration. DC to 10 Hz covers stretch and shear. Meissner corpuscles earn their place on one function: incipient slip detection. An object starts to slide long before it is dropped, and the vibration that begins at that moment sits inside 5 to 50 Hz. The grip correction that follows in a human arrives in roughly 60 to 80 milliseconds, against a conscious reaction time nearer 200 milliseconds. That window is why the channel exists. Nothing downstream of the sensor can make it longer. Put a slip detector behind a 20-millisecond software path and the hand is still late. Five hertz sounds like a rounding error. It is the difference between holding an egg and cleaning up after one.

The biology did not leave a range to argue about. It left a table.

Pacinian corpuscles are the strangest of the four, and the one worth copying deliberately. Their layered capsule behaves as a mechanical high-pass filter. Static load never reaches the nerve ending. Only frequencies above roughly 50 Hz get through. That filter produces distal attribution. A hand holding a pen feels texture at the pen tip, not in the fingers holding it. Surgeon, forceps, tissue. The channel transmits the touch of a tool as though the tool tip were skin. Any system that manipulates objects indirectly needs that bandwidth, and it has to sit in the sensor rather than in software, because the information is gone before software sees it. Broadband piezoelectric sensing from 10 Hz to 1 kHz is the published equivalent. Bands above 1 kHz are the ceiling most transducer materials care about. A surgical robot without the vibration channel is a hand wearing a glove it cannot take off.

Add the four columns up and the specification is finished. Four channels, four bandwidths, four receptive field sizes, four densities, and one graded distribution across the hand. The numbers come from physiology going back decades, not from a competitor's datasheet. The probability that a tactile specification survives procurement is set almost entirely by whether its numbers trace back to something the buyer already accepts. A specification citing 100 per square centimetre and a 1-to-2-millimetre discrimination limit argues from a benchmark published before the product existed. A specification citing internal targets argues about targets. I have built companies across twelve countries and deployed 210 energy systems. The pattern repeats. The teams that move found a number that already existed and inherited it. The teams that stalled spent two years drafting a specification the biology had already written. Copy the receptors, and the argument shifts from what the specification should be to who can manufacture it. That is a manufacturing question, and manufacturing questions get answered.