An Electromagnetism-Inspired Method for Estimating In-Grasp Torque from Visuotactile Sensors

TL;DR

The paper tackles the challenge of estimating in-grasp tilt torques from visuotactile sensors without relying on deep learning or sensor-specific mechanical models. It introduces the Tactile Dipole Moment, computing a dipole-like quantity $\mathbf{p}_{tilt} = \frac{1}{N} \sum_{i=1}^N \mathbf{r}_i ( \nabla\cdot\mathbf{v}_i )$ from the divergence of the marker velocity field and mapping it to tilt torque via $\boldsymbol{\tau}_{tilt} = [c_x (p_{tilt})_y, -c_y (p_{tilt})_x]^T$, with the origin chosen from the centroidal midpoints of positive and negative divergence regions. The method is evaluated across two visuotactile sensors and three object geometries, and applied to a USB-stick insertion task with a compliant wrist, showing a linear relationship to ground-truth torques and improved accuracy over a baseline analytic approach. It generalizes to the DIGIT sensor and to peg-shaped objects, albeit with per-object calibration, highlighting its practicality for real-time, extrinsic-contact sensing and manipulation. The work offers a simple, interpretable, sensor-agnostic tool for extracting torque information from visuotactile data and provides open-source resources for replication and extension.

Abstract

Tactile sensing has become a popular sensing modality for robot manipulators, due to the promise of providing robots with the ability to measure the rich contact information that gets transmitted through its sense of touch. Among the diverse range of information accessible from tactile sensors, torques transmitted from the grasped object to the fingers through extrinsic environmental contact may be particularly important for tasks such as object insertion. However, tactile torque estimation has received relatively little attention when compared to other sensing modalities, such as force, texture, or slip identification. In this work, we introduce the notion of the Tactile Dipole Moment, which we use to estimate tilt torques from gel-based visuotactile sensors. This method does not rely on deep learning, sensor-specific mechanical, or optical modeling, and instead takes inspiration from electromechanics to analyze the vector field produced from 2D marker displacements. Despite the simplicity of our technique, we demonstrate its ability to provide accurate torque readings over two different tactile sensors and three object geometries, and highlight its practicality for the task of USB stick insertion with a compliant robot arm. These results suggest that simple analytical calculations based on dipole moments can sufficiently extract physical quantities from visuotactile sensors.

Figures (8)

• Figure 1: The core contributions of this work. Left, top: Similarities between the electric fields produced from a dipole charge distribution, and visuotactile marker motion fields produced from tilt torques. Left, bottom: The corresponding equation defining the tactile dipole moment used in this work to estimate tilt torques. Right: The contact-rich USB stick insertion alignment problem that we use to demonstrate our method.
• Figure 2: The diverging components of the tactile field before and after grasping an object respectively (a,b), and the field under the application of tilt torque with and without zeroing the tactile field after grasping, respectively (c,d). Note that the dipole field pattern only appears after zeroing the field after grasping. The diverging vector field component is obtained through nhhd.
• Figure 3: The experimental setup for our main calibration and evaluation experiments, showing the 3D printed jig to enable direct measurement of the torques applied to a grasped object (top), and the way in which a human manually applies the torque (bottom).
• Figure 4: The main result of this work, showing the linear relationship between our method to estimate tilt torques, and the ground truth provided by the FT sensor (Left). This is compared against the method proposed in fingervision, which also estimates tilt torques without the use of deep learning (right). The colors correspond to different experiments, each involving different maximum applied torques. Units for RMSE and Slope are in Nmm.
• Figure 5: A typical time series plot of the torque applied during an experiment, both as obtained from the FT sensor as well as the estimation obtained from our method. This shows the quasi-static nature of the loads that we applied for evaluation. The scaling factor for the tactile estimation curve is obtained through the experiment illustrated in Fig. \ref{['figure:main-result']}.