WaveTouch: Active Tactile Sensing Using Vibro-Feedback for Classification of Variable Stiffness and Infill Density Objects

TL;DR

Problem: classify object stiffness and infill density during gripping to guide safe, adaptive grip force. Approach: active vibro-feedback using an emitter finger and a receiver accelerometer, transmitting chirps through objects and extracting differential FFT features. Findings: soft materials absorb more at low-frequency bands ($100-400~\mathrm{Hz}$) and rigid materials amplify at high bands ($400-1000~\mathrm{Hz}$); hollow versus solid infill shows distinct absorption patterns; 50 trials confirm repeatability. Significance: demonstrates a lightweight tactile sensing approach for real-time object characterization during grip and motivates extension to broader object properties and safer human-robot interaction.

Abstract

The perception and recognition of the surroundings is one of the essential tasks for a robot. With preliminary knowledge about a target object, it can perform various manipulation tasks such as rolling motion, palpation, and force control. Minimizing possible damage to the sensing system and testing objects during manipulation are significant concerns that persist in existing research solutions. To address this need, we designed a new type of tactile sensor based on the active vibro-feedback for object stiffness classification. With this approach, the classification can be performed during the gripping process, enabling the robot to quickly estimate the appropriate level of gripping force required to avoid damaging or dropping the object. This contrasts with passive vibration sensing, which requires to be triggered by object movement and is often inefficient for establishing a secure grip. The main idea is to observe the received changes in artificially injected vibrations that propagate through objects with different physical properties and molecular structures. The experiments with soft subjects demonstrated higher absorption of the received vibrations, while the opposite is true for the rigid subjects that not only demonstrated low absorption but also enhancement of the vibration signal.

Figures (4)

• Figure 1: Representation of the Experimental Setup. (1) Schunk gripper with plastic fingers and silicone tips is squeezing a testing object. (2) Photo of the real setup. (3) Vibrations injected by a haptuator propagate through the object.
• Figure 2: Signals used during the experiments: (a) Raw signal from the MCU; (b) Original FFT spectra of emitted and received signals and their difference.
• Figure 3: Classification results for variable stiffness: (a) FFT signal difference between emitter and receiver with low range peaks, and high range trends used for 2-feature classification; (b) Classification map
• Figure 4: Classification results for variable infill: (a) FFT signal difference between emitter and receiver with low range peaks, and high range trends for 2-feature classification; (b) Classification map