MicCheck: Repurposing Off-the-Shelf Pin Microphones for Easy and Low-Cost Contact Sensing

Steven Oh, Tai Inui, Magdeline Kuan, Jia-Yeu Lin · 2025

Read MicCheck: Repurposing Off-the-Shelf Pin Microphones for Easy and Low-Cost Contact Sensing by Steven Oh, Tai Inui, Magdeline Kuan, Jia-Yeu Lin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-18299.txt.

Introduction

Steven Oh, Tai Inui, Magdeline Kuan, and Jia-Yeu Lin turn an ordinary Bluetooth pin microphone into a cheap touch sensor for robot hands. Manipulation is rich in contact, yet imitation learning leans on vision, which misses stiffness, roughness, and slip. Dedicated touch sensors tend to be costly, fragile, or hard to fit. MicCheck instead clips the microphone into a printed gripper insert and streams its audio through a standard receiver, with no custom electronics.

Even so, the sound of contact carries real information. On a ten-class material test across tapping, knocking, pressing, and dragging, the sensor sorts materials with high accuracy. Plugged into an imitation learning pipeline, it also improves manipulation. The appeal is practical: a sensor anyone can buy and clip on.

Catalog record

CREATORSteven Oh, Tai Inui, Magdeline Kuan, Jia-Yeu Lin
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-18299

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MicCheck: Repurposing Off-the-Shelf Pin Microphones for Easy and Low-Cost Contact Sensing Steven Oh, Tai Inui, Magdeline Kuan, Jia-Yeu Lin Abstract Robotic manipulation tasks are contact-rich, yet most imitation learning (IL) approaches rely primarily on vision, which struggles to capture stiffness, roughness, slip, and other fine interaction cues. Tactile signals can address this gap, but existing sensors often require expensive, delicate, or integration-heavy hardware. In this work, we introduce MicCheck, a plug-and-play acoustic sensing approach that repurposes an off-the-shelf Bluetooth pin microphone as a low-cost contact sensor. The microphone clips into a 3D-printed gripper insert and streams audio via a standard USB receiver, requiring no custom electronics or drivers. Despite its simplicity, the microphone provides signals informative enough for both perception and control. In material classification, it achieves 92.9% accuracy on a 10-class benchmark across four interaction types (tap, knock, slow press, drag). For manipulation, integrating pin microphone into an IL pipeline with open source hardware improves the success rate on picking and pouring task from 0.40 to 0.80 and enables reliable execution of contact-rich skills such as unplugging and sound-based sorting. Compared with high-resolution tactile sensors, pin microphones trade spatial detail for cost and ease of integration, offering a practical pathway for deploying acoustic contact sensing in low-cost robot setups. I Introduction Imitation learning (IL) has advanced robot manipulation substantially, yet many everyday skills remain contact–rich: task success often hinges on cues that are difficult to perceive with vision alone (e.g., stiffness, roughness, damping, incipient slip, and micro-impacts at the contact interface). Tactile and acoustic feedback can complement vision by sensing contact events directly and by improving robustness under occlusion or challenging illumination. However, practical adoption of tactile sensing faces a deployment gap: higher-performance solutions (e.g., vision-based tactile sensors, custom contact microphones, piezoelectric arrays) tend to be costly, fragile, or integration-intensive (amplifiers, drivers, bespoke software), limiting their use outside well-equipped laboratories. Notably, many manipulation tasks do not require ultra-fine spatial resolution; rather, they benefit from reliable, timely signals that separate no-contact from meaningful contact, discriminate broad material classes, and flag events such as slip or impact. In such regimes, a simpler, lower-cost, and easier-to-integrate sensor can be a reasonable trade-off. Fig. 1: Overview of MicCheck. We repurpose low-cost pin microphones for contact sensing. We demonstrate this through two experiments: robotic manipulation and object classification. We propose MicCheck , a plug-and-play approach that repurposes an off-the-shelf Bluetooth pin microphone as a low-cost contact sensor. The microphone clips into a 3D-printed gripper insert and operates out of the box—no custom electronics or drivers—while the stock foam provides compliance and robust acoustic coupling. Despite its simplicity, MicCheck yields signals that are informative for both perception and control: in material classification, it attains 92.9% window-level accuracy on a 10-class benchmark, and in manipulation, incorporating audio into an IL policy improves bottle-cap pouring success from 0.40 (vision only) to 0.80 (vision+audio) and supports two additional contact-rich tasks (texture sorting and high-friction unplugging). These results indicate that low-cost acoustic sensing can provide useful contact awareness, though it does not replace higher-resolution tactile sensors for fine spatial discrimination. Our contributions are as follows (Fig. 1 ): (i) A minimal, low-cost acoustic contact sensor using an unmodified consumer microphone and a simple mechanical mount; (ii) integration into classification and an IL pipeline showing consumer microphone complements imitation learning and is beneficial to contact-rich tasks. In this paper, Sec. II reviews tactile and acoustic sensing and positions our work. Sec. III details hardware, signal processing, and the learning setup. Sec. IV presents material classification and real-world manipulation results with ablations and discussion. Sec. VI concludes, and Sec. V outlines limitations and future directions. II Related Work II-A Passive Tactile Sensing Passive tactile sensing methods measure interaction signals without injecting external stimuli. Traditional approaches rely on resistive, capacitive, or force–torque sensors, while more recent systems leverage vision-based tactile designs. Notable examples include the TacTip family of optical biomimetic fingertips [ 17 ] , GelSight for high-resolution geometry and force estimation [ 19 ] , and derivatives such as GelSlim [ 4 ] , DIGIT [ 5 ] , and OmniTact [ 14 ] . Magnetic-based sensors such as ReSkin [ 2 ] offer scalable, low-cost tactile skins. Passive acoustic sensing has also been explored, e.g., SonicSense [ 9 ] , which embeds contact microphones into a multi-fingered hand to capture vibrations for object recognition and material classification. These methods demonstrate that passive signals can encode rich contact information, though often at the cost of custom fabrication or integration. II-B Active Sensing In contrast, active tactile sensing injects energy—through motion or actuation—into the environment and interprets the response. Active haptic perception surveys [ 15 , 1 ] highlight how active strategies enable disambiguation of materials and shapes. Lepora et al. demonstrated exploratory tactile servoing using TacTip [ 6 ] , while more recent work formalizes servoing with pose and shear features [ 7 ] . Martinez-Hernandez et al. [ 12 ] applied active sensorimotor control for autonomous tactile exploration, and Shahidzadeh et al. [ 16 ] combined reinforcement learning with active tactile exploration for shape inference. Within acoustics, Lu and Culbertson [ 11 ] demonstrated active acoustic sensing for grasp state estimation, while VibeCheck [ 20 ] achieved robust peg-in-hole insertion using only an emitter–receiver acoustic pair. Active methods thus provide controllable, discriminative signals but increase hardware complexity. II-C Acoustic Sensing in Robot Learning Acoustic signals have recently been explored as a complementary modality for robot learning in contact-rich tasks. Some studies use audio–visual pretraining to improve generalization in low-data regimes [ 13 ] , while others show that incorporating audio with vision enables better adaptation to texture changes, slip events, and hidden object states [ 10 ] . Multimodal systems that fuse vision, touch, and audio also improve performance on tasks such as dense packing and pouring by combining global, temporal, and local cues [ 8 ] . Beyond passive use, active acoustic sensing has been applied to infer material properties and grasp states through wave transmission [ 11 ] , and recent work shows that audio-only feedback can support robust imitation-learned peg-in-hole insertion [ 20 ] . Together, these results highlight how acoustic cues capture contact events that are hard to perceive visually or tactually. Our work follows this trend but emphasizes accessibility, using an off-the-shelf microphone as a plug-and-play solution for perception and imitation learning. Fig. 2: Architecture of the contact-based object classification model. Single-channel Mel spectrograms from 4 types of mic–object interactions (tap, knock, slow, drag) on 9 objects plus a “blank” no-contact class are fed into a compact 2D CNN comprising three Conv–BN–ReLU blocks, followed by global (adaptive) average pooling and a linear classifier. Models were trained with an 8:2 train/validation split (stratified by class) using cross-entropy loss and the Adam optimizer (learning rate 3\times 10^{-4} , batch size 32) for 2000 epochs, with the best checkpoint selected by highest validation accuracy. The blank class in the softmax serves as a rejection threshold for low-evidence windows. Fig. 3: Teleoperation setup. We employ the Lerobot SO-101 teleoperation setup with a modified gripper. A commonly found commercial bluetooth microphone is embedded onto the gripper. The microphone is connected to a PC via a wireless USB retriever. Fig. 4:

Read the complete text