Reproduce “Learning to Detect Slip through Tactile Estimation of the Contact Force Field and its Entropy” 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
Xiaohai Hu, Aparajit Venkatesh, Yusen Wan, Guiliang Zheng, Neel Jawale, Navneet Kaur, Xu Chen, Paul Birkmeyer
Researchers from the University of Washington and Amazon Robotics developed a real-time slip detection system for robotic manipulation, employing GelSight Mini tactile sensors to estimate contact force field entropy. The system achieved 95.61% accuracy using entropy-based features and demonstrated effective generalization to unseen objects, enabling robots to maintain stable grasps by adapting grip force.
Key insights
- The inhomogeneity of tactile sensor readings, quantified by the entropy of the contact force field, serves as a highly discriminative indicator for incipient slip events.
- The rate of change of entropy (dE/dt) offers improved generalizability for slip detection on unfamiliar objects compared to features based solely on mean velocities.
- Combining entropy and its rate of change with mean velocity features results in the highest overall accuracy for slip classification, enhancing the robustness of detection across various objects and slip conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.