TactileAR: Active Tactile Pattern Reconstruction

TL;DR

This paper builds a Gaussian triaxial tactile sensor degradation model and proposes a tactile pattern reconstruction framework based on the Kalman filter that enables the reconstruction of 2-D HR contact surface shapes using collected LR tactile sequences.

Abstract

High-resolution (HR) contact surface information is essential for robotic grasping and precise manipulation tasks. However, it remains a challenge for current taxel-based sensors to obtain HR tactile information. In this paper, we focus on utilizing low-resolution (LR) tactile sensors to reconstruct the localized, dense, and HR representation of contact surfaces. In particular, we build a Gaussian triaxial tactile sensor degradation model and propose a tactile pattern reconstruction framework based on the Kalman filter. This framework enables the reconstruction of 2-D HR contact surface shapes using collected LR tactile sequences. In addition, we present an active exploration strategy to enhance the reconstruction efficiency. We evaluate the proposed method in real-world scenarios with comparison to existing prior-information-based approaches. Experimental results confirm the efficiency of the proposed approach and demonstrate satisfactory reconstructions of complex contact surface shapes. Code: https://github.com/wmtlab/tactileAR

Figures (8)

• Figure 1: An illustration of proposed active tactile pattern reconstruction process. The tactile sensor taps the contact surface vertically downward and reaches a given height to collect a LR tactile pattern $\mathbf{I}^{\mathrm{LR}}_t$. The collected LR data is used to update the $\mathbf{S}_t$ , which is defined as the 2D shape of contact surface. In the first several taps, the sensor tends to prioritize unexplored areas (e.g. $t=1, 2$). As the number of taps $t$ increases, the sensor gradually focuses on the contour information of the reconstructed surface (e.g. $t=5$).
• Figure 2: An example of 3-axis LR tactile signals collected by the Xela tactile sensor. The sensor has 4 $\times$ 4 taxels with $\vcenter{\hbox{\texttildelow}}$ 20$\times$20 mm$^2$ sensing area (adapted from wubing_tactile_sr_iros_2022).
• Figure 3: Problem description. Suppose the LR tactile sensor has $N\times N$ taxels with $l^{\mathrm{sensor}}\times l^{\mathrm{sensor}}$ sensing area. Assume that a HR sensor has the same size as the LR sensor, but with $M\times M$ taxels. The data collect by the LR sensor can be considered as the degradation of the HR sensor. The reconstructed area is $\alpha$ times larger than the sensor's sensing area, but has the same resolution as the HR sensor. The sensor center of the first tapping is taken as the origin of the reconstruction system.
• Figure 4: An illustration of the degradation process from the HR tactile data $\mathbf{I}^{\mathrm{HR}}_t$ to LR data $\mathbf{I}^{\mathrm{LR}}_t$. The HR sensor model is an idealized representation with smaller taxel size and smaller distances between adjacent taxels than the actual LR sensor (adapted from wubing_tactile_sr_toh_2023).
• Figure 5: An illustration of the Sequential filtering process of the Kalman filter. The triaxial data gathered by the sensor can be considered as three independent observations. The posterior of the previous tapping and the X-axis LR data are used to update the posterior of the X-axis. Then, this posterior serves as the prior for the Y-axis, resulting in a sequential state update.