A laundry robot that hangs every garment from the dryer has to carry a few hanger shapes, and the shape decides whether a shirt comes out pressed or with shoulder bumps. This repository has designed those hangers in cloth simulations for three weeks, across 25 small projects. What none of them asked is how good the simulator has to be before the hanger it picks is the right one. The new hanger-codesign programme asks exactly that, aiming at a robotics journal, and its proposal sets out the plan: three solver tiers, fabrics with published measurements, and a check of whether cheap models can certify a design on their own.
The first step ran our two cheap models, a 2D chain along the shoulder line and a 3D sheet over an arm of finite depth, on four fabrics whose bending rigidity and weight are published Kawabata (KES-FB) measurements. Both pick the same hanger on every fabric: the round-ended wood-shaped profile, about 0.4 cm RMS from the worn shape against 3.1 to 3.4 cm for wire and plastic, with the arm 1 cm wider than the half shoulder.
Then the less comfortable part. On a deep arm the two models agree to 0.004 cm, which is not two models confirming each other: it is one model written twice. On a shallow 3 cm arm, where the 3D sheet adds real physics (the front and back panels fall away), doubling the solver's iterations moves the error by 0.05 to 0.11 cm, the same size as the 0.08 to 0.13 cm margin that separates the best arm width from the next. The decision held, but the solver was not converged tightly enough to say so with confidence.
So the next steps are a validated solver as the top tier, a drape test calibrated to published drape coefficients, and a stopping rule tied to the decision margin rather than an iteration count. All of it is simulation: the stretch stiffness is assumed, the garment is one size-M strip, and no real shirt has been hung. The numbers are in the results file. Written with AI assistance (Claude).