Acoustic Sensing for Universal Jamming Grippers
Lion Weber, Theodor Wienert, Martin Splettstößer, Alexander Koenig, Oliver Brock · 2026
Read Acoustic Sensing for Universal Jamming Grippers by Lion Weber, Theodor Wienert, Martin Splettstößer, Alexander Koenig, Oliver Brock (2026) 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-2603-00351.txt.
Introduction
Weber and colleagues present acoustic sensing for jamming grippers, whose soft bodies grasp objects but lose compliance with tactile sensors. A speaker and microphone inside the cavity, away from the membrane, turn the soft body into the sensor. Sound through gripper and object encodes traits for machine learning, with compliance intact. Size comes within under three millimetres and orientation within under one degree. It resists 80 dBA noise, tells materials apart with up to 100 percent accuracy, and names 16 objects with 85.6 percent accuracy. Sorting runs 53 minutes of uninterrupted grasping and sensing.
Catalog record
| CREATOR | Lion Weber, Theodor Wienert, Martin Splettstößer, Alexander Koenig, Oliver Brock |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39465 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.00351
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Acoustic Sensing for Universal Jamming Grippers Thanks: Equal contributions Thanks: Robotics and Biology Laboratory, Technische Universität Berlin Thanks: Science of Intelligence, Research Cluster of Excellence, Berlin Thanks: Robotics Institute Germany Thanks: This paper extends our workshop contribution [ 20 ] by evaluating sensor resolution across object orientations and providing an extensive discussion of the approach. Portions of the earlier version, including text and figures, are reused or adapted in this manuscript. Thanks: This work has been partially supported by the German Federal Ministry of Research, Technology and Space (BMFTR) under the Robotics Institute Germany (RIG) and funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy – EXC 2002/1 – project no. 390523135; and project no. 405033880. Thanks: Project page with videos and code: https://rbo.gitlab-pages.tu-berlin.de/papers/acoustic-jamming-icra26 Lion Weber, Theodor Wienert, Martin Splettstößer, Alexander Koenig, Oliver Brock Abstract Universal jamming grippers excel at grasping unknown objects due to their compliant bodies. Traditional tactile sensors can compromise this compliance, reducing grasping performance. We present acoustic sensing as a form of morphological sensing , where the gripper’s soft body itself becomes the sensor. A speaker and microphone are placed inside the gripper cavity, away from the deformable membrane, fully preserving compliance. Sound propagates through the gripper and object, encoding object properties, which are then reconstructed via machine learning. Our sensor achieves high spatial resolution in sensing object size ( 2.6 mm 2.6\text{\,}\mathrm{mm} error) and orientation ( 0.6 ° 0.6\text{\,}\mathrm{\SIUnitSymbolDegree} error), remains robust to external noise levels of 80 dBA 80\text{\,}\mathrm{dBA} , and discriminates object materials (up to 100 % 100\text{\,}\mathrm{\%} accuracy) and 16 16 everyday objects ( 85.6 % 85.6\text{\,}\mathrm{\%} accuracy). We validate the sensor in a realistic tactile object sorting task, achieving 53 53 minutes of uninterrupted grasping and sensing, confirming the preserved grasping performance. Finally, we demonstrate that disentangled acoustic representations can be learned, improving robustness to irrelevant acoustic variations. I Introduction Universal jamming grippers [ 1 ] grasp by enveloping an object with a flexible membrane filled with granular media, and then stiffening through jamming. This simple, compliant mechanism lets them conform to a wide variety of unknown objects, making them particularly attractive for grasping in unstructured environments. Despite their versatility, universal jamming grippers remain difficult to sensorize: traditional rigid sensors compromise compliance, reducing grasping performance. Yet, tactile sensing is essential in realistic environments. Acoustic sensing offers a solution that preserves compliance while providing rich tactile feedback. Acoustic sensing is a form of morphological sensing [ 19 ] , where the gripper’s soft body itself functions as the sensor. The key idea is that interactions between the gripper and environment induce measurable changes in the gripper’s morphology. Sound propagating through the gripper captures these changes as modulated signals, from which machine learning reconstructs desired sensor measurements. In this paper, we explain how to build an acoustic sensor that fully preserves the gripper’s compliance (Fig. 1 ). We then show that a single acoustic sensor can perceive many physical quantities that may otherwise require specialized sensors. The acoustic sensor estimates object size with millimeter-scale resolution and remains robust to substantial external acoustic noise. It reliably senses object materials , which is impossible for cameras when objects appear visually identical. Our approach also senses the object orientation with sub-degree resolution. In a realistic application, we discriminate the class of 16 household objects with a high accuracy of 85.6\text{\,}\mathrm{\%} and show the gripper’s unaffected grasping performance in a tactile object-sorting task. Finally, we demonstrate that the high-dimensional acoustic data can be transformed into disentangled latent spaces, isolating task-relevant information and improving robustness against irrelevant variations such as object pose. Fig. 1: Left: Universal jamming grippers conform to unknown objects, enabling versatile grasping, but traditional sensors restrict this deformability. Right: We address this with acoustic sensing. A speaker emits sound into the gripper cavity, which propagates through the gripper and object, encoding object properties (size, material, orientation, class, etc.) in the modulated signal shown in green. A microphone then records the signal, and machine learning reconstructs the object state. II Related Work We first analyze existing sensing methods for universal jamming grippers, with a focus on their impact on gripper compliance, and then present prior art on acoustic sensing. II-A Sensorized Universal Jamming Grippers Some prior works on sensorizing universal jamming grippers rely on mounting rigid cameras on or inside the gripper. For example, a small camera can be placed at the center of the gripper membrane [ 12 ] , but the rigid mount inside the gripper likely worsens compliance. Other efforts relying on cameras [ 8 , 15 , 14 ] fill the gripper cavity with liquid and glass beads of the same refractive index. Cameras mounted inside the gripper can then see the membrane deformation through the glass beads to estimate the object’s shape and pose. A challenge of these approaches is protecting the electronics inside the gripper and other robot parts from liquid leakage. Most of these works [ 15 , 14 ] require custom membranes with physical markers, which were shown to worsen grasping capabilities [ 15 ] and are complex to manufacture compared to using latex balloons as membranes [ 1 ] . Other works directly sensorize the gripper membrane with flexible sensor arrays [ 10 , 13 ] . These sensors can predict proximity, membrane deformation, normal forces [ 10 ] , and slippage [ 13 ] . However, the attached sensor arrays again limit the membrane’s compliance, require a complex manufacturing process, and have coarse spatial resolution of 6 \times 6 [ 10 ] or 3 \times 3 [ 13 ] sensor grids. Our sensor is mounted far away from the membrane and therefore preserves its compliance, is easy to manufacture, and can accurately reconstruct various physical quantities from a single sensor signal. II-B Acoustic Sensing Industry and robotics research increasingly leverage acoustic signals. In industrial settings, acoustic sensing detects tool wear [ 9 ] , rail infrastructure failure [ 7 ] , and motor defects [ 4 ] . In robotic manipulation with rigid grippers, acoustic sensing classifies household objects [ 16 , 21 ] , estimates object materials and contact positions [ 11 , 21 ] , and enables peg-in-hole insertion [ 23 ] . Acoustic sensing is also applied in soft robotic grasping. Wall et al. [ 18 ] present an acoustic sensor integrated into soft robotic fingers, which accurately predicts contact locations, forces, materials, and temperature. They demonstrated that active acoustic sensing, where sound is injected rather than passively observed, yields more robust predictions [ 18 ] . Other works reconstruct the deformed shape of soft robotic actuators using acoustics [ 22 , 17 ] . While prior work demonstrated the versatility of acoustic sensing across many robotics domains, it has not yet been applied to universal jamming grippers—despite these grippers being challenging to sensorize due to their deformable nature. To our knowledge, we are the first to bridge this gap. III Acoustic Sensing for Universal Jamming Grippers This section explains how to build and operate the acoustic jamming gripper, showing how the gripper’s deformable body, which is normally a challenge for sensorization, turns into an asset for morphological sensing. III-A The Gripper Body as a Sensor Our design goal