The authors introduce the Tactile Dipole Moment, a method for estimating in-grasp tilt torques from visuotactile sensors. The key idea is to draw an analogy between the marker displacement vector field patterns caused by tilt torques and the electric fields produced by electric dipole moments.
The authors first demonstrate the improved accuracy of their Tactile Dipole Moment method compared to an existing analytical technique for estimating tilt torques from 2D marker displacement data. Experiments are conducted using a Gelsight Mini sensor and a force/torque sensor as ground truth.
The authors then show how the tilt torque estimates from their method can provide useful feedback signals for a contact-rich robot manipulation task, such as USB stick insertion. They also test the generalization of their approach to a different visuotactile sensor (DIGIT) and various grasped object shapes, finding that the method can be applied across different hardware and geometries.
The results suggest that simple vector calculus techniques inspired by electromagnetism can effectively extract physical quantities like tilt torques from visuotactile sensor data, without requiring deep learning or detailed sensor modeling.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Yuni Fuchiok... ב- arxiv.org 04-25-2024
https://arxiv.org/pdf/2404.15626.pdfשאלות מעמיקות