In plain words
OpenTac is an ecosystem for making tactile skins that fit different robot shapes and for using their readings in robot tasks. A soft bump spreads pressure over nearby sensing spots: similar signals indicate a push into the skin, while differences indicate a sideways push.
On an SO-101 robot arm without vision, a model using touch succeeded in 82% of peg-in-hole trials, compared with 66% for a version using joint positions alone. This result covers that task and its 50 evaluation trials.
The research
Robot skins can be shaped to a machine, but that leaves a practical research question: can custom tactile hardware and its software provide useful contact information for manipulation, rather than merely fit the surface? The paper tests both the fit of generated layouts and whether touch readings help robot actions.
OpenTac combines customizable tactile skins with design, simulation, transfer, and readout tools, then tests their layouts and touch signals in robot tasks.
Across five robot-derived regions, OpenTac-CAD covered more area than an equal-pitch rectangular grid on two irregular shapes, matching it on three near-rectangular ones. On an SO-101 arm without vision, touch-input models trained on 100 demonstrations succeeded in 82% of peg-in-hole trials versus 66% for pose-only models, and 70% versus 54% on plug insertion, each over 50 trials. The results support usefulness in these tasks, while the design gain depended on shape.
These results come from selected hardware and tasks; they do not establish performance on other robots or manipulation settings. Adaptive sampling trades spatial completeness for update rate, and piezoresistive readings can drift with time and temperature.
Analysis
Deep
- Original paper authors
- Ariel Slepyan, Geeta Chandra Raju Bethala, Jiafeng Gu, Anagha Rajendra Dangle, Yinzhe Zhou
- Submitted to arXiv
- License
- arXiv non-exclusive