Reproduce “DexSkin: High-Coverage Conformable Robotic Skin for Learning Contact-Rich Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Suzannah Wistreich, Baiyu Shi, Stephen Tian, Samuel Clarke, Michael Nath, Chengyi Xu, Zhenan Bao, Jiajun Wu
DexSkin is a soft, conformable capacitive electronic skin that provides high spatial coverage and calibrated contact sensing for robotic manipulation. It enabled robots to learn complex in-hand object reorientation with robust perturbation recovery and delicate grasping, reducing contact pressure on artificial berries by 90% and improving real blueberry integrity from 20% to 60%.
Key insights
- Expanded tactile coverage and conformability, as provided by DexSkin, are essential for robots to learn and robustly perform contact-rich manipulation tasks involving dynamic interactions and external perturbations.
- Practical calibration methods are crucial for transferring learned policies across different physical sensor instances, addressing a major bottleneck in deploying data-driven tactile robotics due to manufacturing variability and sensor wear.
- The ability to derive directly interpretable physical quantities like contact force from tactile sensor outputs significantly simplifies the design of effective reward functions for real-world online reinforcement learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.