A tactile skin that needs just one wire: super-resolution factor above 10,000 from palm-sized to back-sized patches
Synopsis
Large-area tactile sensing has meant tiling many sensing units and routing a wire to each, so scaling area or resolution drives up wiring and cost. A serpentine electromagnetic surface-wave transmission line on flexible polyimide is fed swept signals from both ends, and the dual-port reflection spectra encode contact position, which a multi-branch multilayer perceptron decodes into coordinates. Two prototypes an order of magnitude apart in area, 5×5 and 16×18 cm, both exceed a super-resolution factor of 10,000, with two-dimensional localization RMSE of 0.368 mm for the small skin and 1.209 mm for the large one.
Interpretation
The tactile skin no longer buys resolution by packing in sensing units; a single electromagnetic surface-wave transmission line serves the whole sensing plane. Raising resolution previously meant adding smaller sensing units and routing dedicated interconnects to each, and enlarging the area likewise required more units, escalating wiring complexity, fabrication difficulty and cost. The two flexible prototypes differ by an order of magnitude in area, and the 5×5 cm and 16×18 cm skins reach super-resolution factors of 12477.4 and 16886.4, both above 10,000.
Contact position is written into the dual-port reflection spectrum of one routed line and read back as coordinates by a neural network. The line follows a meandering topology that maps its one-dimensional propagation coordinate onto known positions in the plane, and an object near the line perturbs nearby segments through evanescent coupling, so the reflection spectrum is a geometry-weighted projection of the two-dimensional perturbation onto the path. The small prototype was sampled at 13005 points over 7 to 13 GHz with 121 frequency points, the large one at 3772 points over 4 to 10 GHz, and localization is performed by a multi-branch multilayer perceptron.
The same skin also distinguishes objects and follows continuously drawn trajectories. Objects of different material, shape and size leave different spectral signatures, and trajectories are assembled from point-by-point localization. Eight everyday object classes are recognized at 91.3% accuracy and the metallic versus non-metallic split at 100%; digit trajectories 0 to 9 written by a robotic arm reach about 96% under a CNN trained on MNIST, and a seven-key mini piano is recognized at 100% even with a support vector machine.
Perspective
The demonstration covers two flat flexible prototypes, 5×5 cm and 16×18 cm, measured with a vector network analyzer over 7 to 13 GHz and 4 to 10 GHz respectively, with localization and recognition performed on a benchtop setup.
Acquisition rate trades against accuracy: on the small prototype, a single dual-port measurement over 6 GHz bandwidth with 121 frequency points takes 150 ms on average and 250 ms at most, a mean sampling rate of about 6 Hz, and fewer frequency points speed acquisition but reduce precision. Wider bandwidth is not uniformly better either: pushing the small prototype's upper limit from 11 GHz to 13 GHz increases localization error, while on the large prototype extending from 8 GHz to 10 GHz yields little further improvement.