Mixed-Stiffness Fingertips Widen the Operating Range for Coin Pickup
Picking a coin flush on a hard table is still awkward for robots. Many stacks cheat with under-edge nails, scoops, or prior flip. Humans often just tip a pad into the coin’s side and rotate it free — no underside access.
Researchers at the University of Electro-Communications (Tokyo) ask whether where you put pad stiffness matters as much as how soft the pad is. Their paper, arXiv:2608.07887 (submitted Aug 8, 2026), fabricates anthropomimetic fingertips with Shore E10 silicone in the center and Shore A60 on the sides, then runs automated oblique rotational tip pinches against uniformly soft E10 tips and a nail-free control.
Setup
All nails (when present) are SLA-printed; pads share geometry and a 1.5 mm skin. An Arduino-driven apparatus varies horizontal approaches, vertical displacements, index rotation, and coin position on a Japanese 500-yen coin (26.5 mm diameter, 1.8 mm thick). Ten trials per condition. CAD, STL, code, and trial tables are on GitHub (TogoLab).
Results that stuck
- Nail-free pair: 0/36 conditions formed a geometric constraint in Experiment 1-1.
- Mixed pair integrated success score on the d1–d2 grid: 23.5 (mean 0.653) vs uniform 18.5 (0.514); bootstrap 95% CI on the difference [3.8, 6.2].
- Mixed also won on vertical displacement and rotation-angle sweeps, and on transition to stable grasp (mean 0.693 vs 0.350 on 14 shared conditions).
- Trade-off: when coin position along the finger varied, the uniform pad scored higher.
- After per-denomination tuning, both mixed and uniform grasped all six Japanese coin types at 100%.
A Human’s Take
This is the kind of small, measurable hardware paper I want more of. Not a new foundation model — a stiffness map that widens the set of poses that still form a constraint. The honest bit is the coin-position trade-off: mixed is not free lunch. If they open the molds and someone ports this to a commercial two-finger tip for card and coin trays, that is a real product conversation.