At the 1990 Robot Olympics in Glasgow, a headless, armless machine called Shadow Walker stood on two legs and refused to fall. When pushed, it shifted its weight and recovered. Its 28 pneumatic air muscles, hand-built from salvaged parts and medical supply tubing, responded to perturbation with something that looked like biological reflex. It had twelve degrees of freedom across eight joints. It could absorb impact, redistribute force, modulate pressure in real time. By every metric that matters for standing, it was competent.
It could not take a single step.
Richard Greenhill, a photographer with no robotics training, spent three years building the machine in his attic in London. Every Saturday, his wife Sally made spaghetti for a rotating crew of hobbyists who called themselves the Shadow Group. They worked from medical textbooks on human anatomy. The skeleton had deliberate simplifications — one bone per lower leg, single wide toes — but the actuation was faithful to biology’s principle: pneumatic muscles that contracted and expanded with compressed air, producing smooth analog force across a continuous range. The McKibben muscle design they adapted was thirty years old in 1987, invented in the 1950s to mimic how mammalian muscles work. Greenhill’s contribution was proving it could hold a body upright.
Walking is not standing with forward progress. Standing is static equilibrium: forces balanced, position maintained. Walking is controlled falling. Each step begins with the body losing balance, tilting forward past the point of recovery, and catching itself with the next footfall. The twelve degrees of freedom that made Shadow Walker an elegant stander made it an impossible walker. Its joints yielded on contact, and that flexibility meant the space of possible movements was enormous. Programming a sequence that navigated that space without collapsing required coordinating 28 muscles across millisecond timescales. Honda’s ASIMO research team, with industrial resources Greenhill’s attic lacked, would need another decade to solve a simplified version of the same problem using rigid electric motors.
The trombone has the same structural feature and the opposite outcome.
Among brass instruments, the trombone is the only one in a classical orchestra where pitch is controlled by moving a slide rather than pressing valves. A trumpet player pushes three buttons in various combinations, each of which reroutes air through a fixed additional length of tubing. The positions are discrete. A trombonist extends a slide along a continuous track. There are seven named positions, but the slide can stop anywhere between them, and the distance between positions varies because frequency is inversely proportional to tube length. First to second position is shorter than sixth to seventh.
This gives the trombone infinite positions. The slide has no frets, no stops, no mechanical lock to tell the player where a note lives. And the slide is only half the problem. To play different notes in the same position, the player changes embouchure — the configuration of lips, tongue, jaw, and airflow — to select a different harmonic partial. Playing a trombone requires managing slide position, embouchure, tongue articulation, air speed, and dynamic balance across all of them simultaneously, with no tactile feedback confirming that any single variable is correct.
A beginner trombonist faces Shadow Walker’s problem: too many degrees of freedom, no fixed steps to navigate by. Every note is an act of coordination across five or six analog variables. The temptation is to overstress the lips, to force pitch by pushing harder, the way a programmer might try to solve walking by adding more constraints to the pneumatic system.
What experienced trombonists develop instead is what Bryan Hu, writing about his own learning, describes as using “ears to determine if it sounds good.” They calibrate by listening to ensemble context, the surrounding harmonic environment, and adjusting moment to moment. The body learns to navigate infinite positions not by mapping them but by tuning itself to the sound field it inhabits. Decades of practice build a kinesthetic vocabulary that relates tongue position to air column resonance to what the player hears reflected back. The analog freedom that paralyzes a mechanical system becomes, in a trained body, the instrument’s advantage: real glissandos, just intonation, pitch that bends into the exact harmonic slot the ensemble needs.
Shadow Walker had biological-style actuation without biological-style control. It had the muscles but not the ears. The system was rich enough to stand because standing has a clear target: balance. Walking has no single target. It is a continuously managed sequence of imbalances. The trombone has no single target either. Every note is a continuously managed convergence of variables, no two performances landing in exactly the same place. The trombonist manages this because the body doesn’t solve the problem once. It solves it differently each time, by listening.
After the 1990 failure, the Shadow Group didn’t try harder at walking. They pivoted. The organization evolved into Shadow Robot Company, now Britain’s oldest robotics firm, and their primary product is a specialized robotic hand. They moved from whole-body locomotion to fingertip dexterity — from one analog-control problem to another, but a smaller one, where the space of possible movements is constrained enough that computational control can manage it without a body in the loop.
The trombonist went the other direction, starting with the smallest analog controls (lip pressure, tongue placement) and building upward to whole-body coordination, from fingertip-scale adjustments to room-scale listening. The same kind of freedom, traveled in opposite directions.
Twelve degrees of freedom held Shadow Walker upright and kept it from moving. Infinite slide positions give the trombone a range that valved instruments cannot reach. The mechanism is the same: analog control across a continuous space with no discrete steps to anchor the system, and whether the freedom paralyzes or liberates depends on nothing the mechanism itself provides.
The difference is not in the degrees. It is in what, or who, is navigating them.