Design and Characterization of a Multimodal Flexible Electronic Skin with CPLD-Based Data Acquisition for Tactile Intensity and Gesture Recognition

Summary

This protocol describes the fabrication and operation of a flexible, multimodal electronic skin system. It utilizes a CPLD-based backend to achieve synchronized, hybrid-frequency data acquisition from 36 sensor channels, enabling the quantitative analysis of complex tactile gestures for advanced robotic perception and human-robot interaction.

Abstract

This work presents the design and characterization of a flexible multimodal electronic skin (e-skin) platform that supports high-resolution tactile sensing and gesture recognition via a CPLD-based data acquisition system. The system integrates a 36-channel, 14-bit hybrid-frequency sampling architecture (2 kHz, 1 kHz, 100 Hz), with hardware-level support for pressure, acceleration, light, temperature, and environmental signals. A flexible FPC substrate enables conformal integration on curved robotic surfaces while maintaining mechanical stability. To demonstrate tactile perception capability, this study focuses on the pressure sensing channel, which serves as the primary modality for contact force estimation and gesture dynamics. Four representative human interactions, gentle touch, tap, light pinch, and strong pinch, were analyzed. Two dimensionless metrics, equivalent waveform skewness and equivalent load, were introduced to distinguish force intensity and gesture categories. Experimental results show that the system achieves clear gesture separation and skin-like viscoelastic response, with an average unloading time of 0.4 s ± 0.2 s. Real-time wireless data transmission at 1.5 Mbps is supported, and the modular design provides a scalable foundation for future integration of additional modalities. This work establishes a hardware framework for force- and gesture-based human-robot tactile interaction, with extendable capacity for multimodal perceptual systems.

Introduction

Tactile sensing has been extensively investigated for robotic perception^1,2, however, most electronic skin (e-skin) systems still rely on single-modality, low-density configurations that can limit the interpretation of complex human touch. Despite significant advances in flexible tactile materials and device architectures³, system-level integration of multimodal and multichannel tactile sensing with high-precision, real-time acquisition presents ongoing technical challenges⁴. Consequently, key issues such as signal synchronization, timing coordination, and data throughput optimization require systematic addressing in large-scale or high-channel-count tactile systems⁵. To address these requirements, a high-bandwidth front-end architecture capable of managing synchronized multichannel data streams, preserving signal integrity, and supporting quantitative gesture and force characterization is necessary^6,7,8.

Multimodal e-skin systems have emerged as a significant development for enabling robotic platforms to emulate the complex tactile sensing capabilities of human skin. Traditional single-modality sensors, which detect only normal pressure or temperature, may not fully capture the complex and dynamic nature of real-world tactile interactions. In contrast, human-inspired multimodal systems are designed to detect multiple types of tactile stimuli in real time, including shear force, vibration, and thermal cues. For example, Lee et al. developed a bioinspired tactile perception system capable of real-time detection across four distinct sensory channels, which improved the recognition of texture, contact mode, and object interaction states⁹.

Recent advancements in flexible electronics have further expanded the horizons of these multimodal systems, particularly through the development of sophisticated electronic skin (e-skin) that mimics the mechanical compliance and high sensitivity of biological tissue. For instance, the exploration of novel active materials and hierarchical structural designs has led to next-generation e-skin platforms with unprecedented tactile sensitivity and multi-functionality¹⁰. Beyond material innovation, the transition toward large-scale application necessitates robust and scalable fabrication techniques. Printed electronics have emerged as a pivotal solution, enabling the precise and cost-effective integration of high-density sensor arrays onto diverse elastomeric substrates while maintaining device performance¹¹.

To ensure the reliability of these systems in practical scenarios, researchers are increasingly focused on optimizing the electromechanical properties and sensing mechanisms of flexible sensors to mitigate signal crosstalk and environmental interference¹². Ultimately, the current paradigm in the field is shifting from isolated sensing components toward fully integrated, intelligent, flexible platforms that synergize advanced materials with high-efficiency signal processing frameworks¹³.

However, integrating multiple sensing modalities into a unified array presents substantial challenges in signal coupling, synchronization, and data interpretation. Kong et al. highlighted the necessity for robust integration and decoupling strategies in high-density multimodal sensor arrays to ensure accurate signal isolation and minimize cross-talk¹⁴. Moreover, the mechanical behavior of the soft sensing substrate significantly influences signal fidelity. Mu et al. demonstrated that the viscoelastic nature of conductive polymer composites could induce signal drift and hysteresis, emphasizing the role of material modeling and system-level compensation strategies in multimodal sensing platforms ¹⁵. Collectively, these findings suggest that developing high-bandwidth, viscoelasticity-aware, and decoupled multimodal e-skin systems is essential for advancing intelligent tactile perception in robotics and human-machine interfaces¹⁶ .

To ensure real-time acquisition of multichannel tactile signals with minimal latency and high signal integrity, a complex programmable logic device (CPLD)-based acquisition framework was adopted. Prior studies have demonstrated that CPLDs are suitable for managing high-speed data pipelines and timing-critical signal streams in sensor systems^17,18.

Compared to traditional data acquisition architectures, this CPLD-based synchronized hybrid-frequency approach offers several distinct advantages. While microcontroller unit (MCU)-based systems often rely on software-level synchronization or sequential polling, they are prone to non-deterministic timing jitter and interrupt latency, which can compromise signal phase alignment across high-channel counts¹⁹. In contrast, the hardware-logic-based architecture of a CPLD ensures strictly parallel execution and nanosecond-level deterministic timing for all 36 channels. Furthermore, while field-programmable gate arrays (FPGAs) provide higher computational power, CPLDs offer a more cost-effective and power-efficient solution for the logic depth required in hybrid-frequency state machines, facilitating long-term stability in wearable e-skin applications. Additionally, unlike lower-channel multiplexing methods that sacrifice effective sampling bandwidth per channel, the proposed synchronized architecture maintains a consistent 2 kHz rate for tactile modalities, overcoming the bottlenecks of heterogeneous data fusion identified in recent literature¹⁹.

This study presents a multimodal flexible electronic skin integrated with a CPLD-based acquisition system. The system supports 36 channels of 14-bit hybrid-frequency sampling, operating at 2 kHz for dynamic tactile channels and 100 Hz for auxiliary modalities, including acceleration, temperature, humidity, illumination, and sound sensors (Figure 1). The system integrates these heterogeneous sensors on a flexible printed circuit (FPC) substrate, ensuring both mechanical adaptability and electrical stability. A compliant viscoelastic encapsulation layer¹⁵, segmented into small rectangular units, was adopted to improve local deformation adaptability and contact stability during tactile interactions.

Regarding practical suitability, the high-speed parallel sampling architecture is optimized for capturing rapid, impulsive tactile events such as tapping or texture sliding within a controlled laboratory environment. However, operating constraints include potential signal fidelity degradation due to the viscoelastic drift and hysteresis of the soft substrate, which necessitates periodic calibration or software-based compensation. Furthermore, while the system is scalable to higher channel counts, the current implementation is bounded by the wireless transmission bandwidth for real-time visualization.

The CPLD manages synchronized multichannel acquisition and wireless data transmission, ensuring stable real-time performance. This approach enables high-density multimodal signal collection and provides a scalable hardware foundation for robotic tactile perception. To evaluate tactile discrimination, four representative gestures, gentle touch, tap, light pinch, and strong pinch, were experimentally analyzed.

This work focuses on demonstrating the feasibility and signal integrity of the high-bandwidth multimodal e-skin architecture. Pressure sensing is implemented using a piezoresistive mechanism based on viscoelastic sensing material, a conductive polymer composite characterized by a volume resistivity of <500 Ωcm. In this configuration, mechanical pressure alters the contact resistance within the material, which is subsequently captured by the acquisition system. Pressure channels were selected as the representative modality to validate real-time acquisition and data quality. While the system supports additional modalities such as acceleration and temperature, their implementation and analysis are reserved for future application-specific studies.

To address the challenges of high-density tactile data acquisition, this work employs a CPLD-based hardware architecture integrated with a Wi-Fi 6 module. As shown in Figure 2, the system is engineered to manage 36 channels with 14-bit resolution at a high-speed sampling rate of 2 kHz, generating a raw data throughput of approximately 1.01 Mbps. By leveraging the 802.11ax protocol, the platform achieves a stable, effective transmission rate of 1.5 Mbps, maintaining a packet loss rate of less than 0.2% in typical laboratory environments. This hardware-logic-based approach eliminates the non-deterministic timing jitter and interrupt latency common in traditional MCU-based architectures, while offering a more power-efficient and cost-effective alternative to FPGA-based solutions. Such a robust data pipeline ensures the precise real-time synchronization and signal integrity necessary for complex gesture recognition and multimodal tactile sensing.

Two dimensionless indicators were defined: equivalent waveform skewness (EWS) and equivalent load (EL). The EWS quantifies the temporal asymmetry of the tactile response waveform, while the EL represents the overall force intensity integrated over the contact duration. Analysis of these indicators revealed distinct mechanical signatures for each gesture, with an average unloading time of 0.4 s ± 0.2 s, suggesting that the substrate's viscoelastic response contributes to the temporal characteristics captured by the system.

These results demonstrate the feasibility of accurate tactile intensity and gesture recognition using synchronized, multimodal acquisition. The proposed system thus provides a functional front-end platform for quantitative tactile analysis and establishes a foundation for future developments in affective or emotion-related human-robot interaction.

Protocol

All experimental procedures described herein were non-invasive and did not involve animal or human clinical subjects; therefore, no ethical approval was required for this study. The reagents and the equipment used are listed in the Table of Materials.

1. Fabrication of sensors

1. Preparation and integration of the viscoelastic sensor
   1. Mark the target dimensions [e.g., 5 mm x 5 mm] on the viscoelastic sensing material using a precision ruler and a fine-tip permanent marker to ensure dimensional accuracy. Cut the material along the marked lines using sharp, fine-tip scissors. Perform the cuts in a single, continuous motion where possible to ensure clean edges and minimize mechanical deformation or fraying at the boundaries.
   2. Clean the surface-mount pads of the Flexible Printed Circuit (FPC) using [e.g., isopropyl alcohol and a lint-free wipe] to remove contaminants and ensure optimal adhesion (Figure 1).
   3. Bond the viscoelastic material pieces onto the surface-mount pads of the FPC using a conductive adhesive.
   4. Apply a thin layer of UV-curable resin around the base of the sensing material to encapsulate the adhesive joint. Cure the resin using a UV light source at 365 nm wavelength with an intensity of 100 mW/cm² for 5 min.
   5. Verify the electrical continuity of the assembled sensor using a digital multimeter.
      NOTE: Resistance values should fall within the range of 5 kΩ to 20 kΩ under zero-load conditions.
   6. Apply external pressure and measure the corresponding change in resistance to characterize the pressure-sensitive properties of the assembled device.

2. Implementation of the data acquisition board

1. Configuration of the multi-rate data acquisition system
   1. Integrate the analog front-end and data-conversion circuits onto a single flexible FPC substrate (Figure 3).
   2. Utilize five 8-channel analog switches (Figure 4A) for sensor channel selection. Interface these with three parallel 14-bit Successive Approximation Register (SAR) Analog-to-Digital Converters (ADCs) (Figure 4B) for signal digitization.
   3. Connect the ADC and switch network to a Complex Programmable Logic Device (CPLD) to manage channel addressing, timing control, and synchronization.
   4. Securely attach the sensor FPC to the acquisition board using a 40-pin FPC connector. Ensure the locking mechanism is fully engaged to prevent signal instability.
2. CPLD-based system control and data handling
   1. Program the CPLD logic to synchronously drive all ADCs by generating precise timing signals according to the required sampling intervals.
   2. Configure the internal logic gates of the CPLD to coordinate high-speed channel switching and merge the parallel data streams from the three ADCs into a unified data packet.
   3. Transmit the merged data via a serial interface at a baud rate of 1.5 Mbps to a Wi-Fi module for real-time wireless communication with a host computer.
3. ADC interfacing for hybrid-frequency sampling
   1. Control each of the three independent 14-bit SAR ADCs through a dedicated 3-wire serial interface consisting of Chip Select (CS#), Serial Clock (SCLK), and Serial Data (SDATA) lines managed by the CPLD.
   2. Implement channel-specific clock dividers within the CPLD architecture to generate the distinct sampling rates required for the 2 kHz, 1 kHz, and 100 Hz modalities (Figure 5).
   3. Store each sampled frame in a temporary buffer relative to a common timebase reference. Assert a one-shot data-valid flag within the logic to signal that the data is ready for downstream transmission.
4. Implementation of time-slice-based sampling synchronization
   1. Define a 10 ms global sampling frame in the CPLD as the base scheduling period and divide it into ten uniform 1 ms time slices.
   2. Assign one or more fixed time slots within this 10 ms frame to each of the 36 sensor channels based on its designated sampling frequency.
   3. Monitor the deterministic scheduling scheme to maintain strict phase alignment across all ADC groups and frames, ensuring synchronous hybrid-frequency data acquisition.

Results

This protocol enables the quantitative characterization and differentiation of distinct tactile gestures applied to the electronic skin (e-skin). Fix the e-skin sample on the test platform. A single operator then performs four defined tactile actions, gentle touch, tapping, light pinch, and strong pinch, according to the parameters listed in Table 1. The "gentle touch" gesture involved repeatedly placing a finger lightly on a single e-skin unit for about 1 min to ensure stable and consistent repetitions....

Discussion

The sensing mechanism of the e-skin is fundamentally rooted in the viscoelastic properties of its constituent polymer^6,15. This material exhibits both elastic deformation and viscous flow when subjected to an external force. The viscous component introduces a characteristic latency in the material's recovery; that is, it does not instantaneously return to its original state upon the removal of the force but instead requires a finite period to recover. These visco...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Flexible Printed Circuit Board	JLCPCB	any one	Used to obtain pressure data
Analog switches	TI	TMUX1108PWR	Used for selecting different channels
Analog to Digital Converter	Analog Devices	AD7940BRMZ-REEL7	Convert analog signals to digital signals
CPLD	Intel	MAX 10M02	Data processing unit
Oscilloscope	Tektronix	TBS2000	This oscilloscope is used to measure signal waveforms.
quartus ii	Intel	any one	Programming IDE for CPLD
viscoelastic sensing material	any one	any one	This material is used to generate pressure signals.
Wi-Fi 6 module	AI-THINKER	ad-m62-cbs	Data transmission module