A Sensor Position Localization Method for Flexible, Non-Uniform Capacitive Tactile Sensor Arrays

Carson Kohlbrenner, Caleb Escobedo, Nataliya Nechyporenko, Alessandro Roncone · 2024

Read A Sensor Position Localization Method for Flexible, Non-Uniform Capacitive Tactile Sensor Arrays by Carson Kohlbrenner, Caleb Escobedo, Nataliya Nechyporenko, Alessandro Roncone (2024) 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-2412-00672.txt.

Introduction

Carson Kohlbrenner and co-authors present a tactile skin with uneven sensor spacing in this 2024 paper. Robot touch sensors are usually laid out on a regular grid. Dense sensing helps fine manipulation, while sparse sensing suffices for collision alerts. One array that mixes both densities could serve both jobs.

Their array, called VARSkin, uses mutual capacitance sensing on a flexible sheet. The hard part is knowing where each sensor sits inside a hidden, uneven layout. Their localization method recovers those positions from the sensor signals themselves. Tests on two skin patches with concealed layouts placed sensors within two millimeters. An error analysis points to ways to sharpen the accuracy further.

Catalog record

CREATORCarson Kohlbrenner, Caleb Escobedo, Nataliya Nechyporenko, Alessandro Roncone
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00672

Browse the full library · About the Tower of Babel

Full text

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

A Sensor Position Localization Method for Flexible, Non-Uniform Capacitive Tactile Sensor Arrays Carson Kohlbrenner, Caleb Escobedo, Nataliya Nechyporenko, Alessandro Roncone Abstract Tactile sensing is used in robotics to obtain real-time feedback during physical interactions. Fine object manipulation is a robotic application that benefits from a high density of sensors to accurately estimate object pose, whereas a low sensing resolution is sufficient for collision detection. Introducing variable sensing resolution into a single tactile sensing array can increase the range of tactile use cases, but also invokes challenges in localizing internal sensor positions. In this work, we present a mutual capacitance sensor array with variable sensor density, VARSkin , along with a localization method that determines the position of each sensor in the non-uniform array. When tested on two distinct artificial skin patches with concealed sensor layouts, our method achieves a localization accuracy within ± 2 \pm 2 mm. We also provide a comprehensive error analysis, offering strategies for further precision improvement. I Introduction Artificial skin technologies, which generally aim at replicating the tactile sensing attributes of humans’, find applications in diverse domains such as healthcare, prosthetics, and robotics [ 1 ] . While numerous frameworks have been proposed, a conspicuous gap remains: the lack of a scalable, variable-resolution architecture suitable for both large scale (e.g., to cover all or the majority of a robot’s body) and small scale (e.g. to perform fine manipulation) implementations. Existing high-fidelity sensor technologies such as DenseTact and GelSight excel in emulating the tactile resolution of human fingertips but are unable to cover large surface areas [ 2 , 3 , 4 ] . Conversely, designs aimed at large surface coverage tend to compromise on tactile resolution [ 5 , 6 , 7 , 8 ] . This dichotomy overlooks the innate versatility of human skin, as seen in Fig. 2 , which exhibits variable sensor density tuned to the functional requirements of different anatomical regions [ 9 , 10 ] . It is precisely this adaptive feature that existing artificial skins, characterized predominantly by their uniformly-distributed sensor arrays, fail to capture [ 11 , 12 , 6 , 5 , 13 , 14 ] . Much like the nuanced tactile resolution across different areas of the human body, the integration of artificial skin into robotic systems calls for a tailored approach to sensor density, particularly when these systems are designed for multifaceted applications ranging from locomotion to fine manipulation. Engineering artificial skins with optimized sensor distribution offers tangible benefits for robotic applications, including enhanced efficiency in data acquisition, power consumption, and wiring architecture, primarily through the strategic elimination of extraneous sensors [ 15 ] . Despite the benefits, a significant challenge emerges: the absence of a straightforward method to determine the relative locations of individual sensors. This issue becomes accentuated when the skin either conceals sensor locations or employs non-standard manufacturing processes, such as randomized sensor arrangements. Without known sensor locations, robots struggle to perform tasks such as object orientation detection and manipulation [ 16 , 17 ] . Fig. 1 : In this work, we propose a localization method to predict the location of each sensor in a non-uniform sensor density artificial skin. The method starts by constructing a point log map that shows a single sensor’s recorded intensities at different probing locations. The point log map is then interpolated, filtered, and averaged, resulting in a single sensor location prediction. This method is then repeated for each sensor to compile the final sensor location predictions. In this paper, we introduce what we believe is the first example of an artificial skin with non-uniform sensor density. This is complemented by a computational method to determine the location of each tactile sensor inside the skin, which we validate on real hardware with two distinct artificial skin patches. Our contributions are as follows: a) we present a flexible, variable-density artificial skin equipped with mutual capacitance sensors. The skin’s variable sensor distribution is fully compatible with our novel localization method, ensuring robust performance; b) we propose a systematic, repeatable localization method designed explicitly for mutual capacitance sensors. This algorithm accurately localizes each embedded sensor, making the use of this skin technology possible for applications requiring detailed spatial information; c) we validate the efficacy of our algorithm by applying it to two distinct artificial skin samples with concealed sensor layouts. Further, we conduct an extensive error analysis to identify and address potential avenues for optimization. Collectively, the contributions presented in this work not only advance the field of tactile sensing but also pave the way for more versatile and adaptive robotic systems. By tailoring sensor density and offering precise localization, our artificial skin can be integrated into a large array of robotic applications requiring complex spatial awareness, from fine object manipulation and perception to dynamic locomotion. Fig. 2 : The prevalent large-scale artificial skin implementation in the iCub robot has equal average sensor density across the robot arm. In contrast, the average biological sensor densities of human skin gradually increase from the base of the arm to the tip of the fingers. In our work, we aim to mimic the biological approach of variable sensor density which we see in humans. [ 9 , 7 ] . II Related Work II-A Past and Emerging Trends in Artificial Skins for Robotics In recent years, many artificial skin technologies have emerged that can measure force, strain, slip, heat, and other sensations related to the sense of touch. High resolution artificial skins have been proposed with acoustics [ 18 , 19 ] , computer vision [ 20 , 3 , 2 , 17 ] , capacitive proximity sensing [ 5 , 21 , 22 , 23 , 24 , 25 , 26 , 7 ] , and dense sensor packages [ 27 ] . Of note, large-scale artificial skins have been developed and successfully deployed on platforms such as the iCub and REEM-C robots, albeit with unoptimized, uniform sensor densities [ 7 , 6 ] . The artificial skin architecture proposed in this work was inspired by the mutual capacitance sensing paradigm articulated by [ 5 ] , and adopts and builds upon the Muca development board to record sensor data. Beyond the advancements in artificial skin technology, there is a corresponding need for solutions that maximize the utility of large-scale robotic skins, particularly in the domain of machine learning-based robotics [ 15 ] . Varied sensor density in artificial skins emerges as a pivotal factor, potentially reducing computational overhead by selectively emphasizing data-rich or mission-critical regions on the robot. This optimization not only minimizes computational and energy costs but also enriches the quality of feedback data, a crucial component in the success of reinforcement learning algorithms [ 28 , 29 ] . Other avenues of efficiency beyond optimized density distribution include reducing the overall number of sensors, a factor which is highly dependent on both the geometry and the specific application of the tactile sensor arrays [ 30 ] . II-B Algorithms for Kinematic Calibration of Robotic Skin with Concealed Sensor Layouts [ 31 ] presents a kinematic approach to determining the location of sensors in artificial skin. This approach requirses an Inertial Measurement Unit (IMU) device for each sensor to calculate

Read the complete text