Introduction

A human-machine interface (HMI) establishes a connection between individuals and machines, systems, and devices. [1][2][3][4][5][6] The interface is poised to revolutionize the future of robotic prosthetic hands, reshaping the way people engage with artificial limbs and enhancing their overall functionality and seamless integration into daily activities. [1,5,7] The utilization of flexible DOI: 10.1002/adsr.202400086 electronics, especially flexible sensors, facilitates the advancement of wearable device platforms within HMIs. [6,[8][9][10] Recent advances in wearable electronics have enhanced HMIs through the incorporation of a set of functional electronics based on micro-electromechanical systems. [11][12][13][14] This has led to the creation of compact and skin-compatible designs. However, existing approaches do not effectively address the fundamental challenges associated with the utilization of traditional rigid electronic components, such as batteries and sensors. In addition, many sensors used in HMI applications such as electromyography (EMG) sensors, which can detect the voltage potential difference between muscle groups, need to overcome the bulkiness of the electrodes in order to achieve highlevel biocompatibility. [15][16][17] Therefore, it is crucial that HMIs embrace innovative technologies and materials, moving beyond the limitations of traditional electronic components.

Moreover, the implementation of selfpowered sensing operations would significantly enhance the practical utility of HMIs. However, strain sensors, which are broadly used in the applications of HMIs, are externally powered. These sensors generate electric signals due to varying resistance or capacitance during the deformation of the human body. [18][19][20][21] Despite the use of various strain sensors to detect body deformation, their drawbacks, such as slow response time, restricted sensitivity within the material's elastic limitations, and poor sensing linearity, significantly limit their applicability in HMIs. In comparison, triboelectric sensors utilized in HMIs can generate electric signals through the triboelectric effect, wherein electrons can transfer between two materials solely through physical contact, thus demonstrating self-powered ability. [22] Researchers have recently enhanced triboelectric sensors in HMIs by developing larger (centimeter-scaled) mechanical structures that improve motion detection, [23] sensitivity, [24] and sensing precision. [25] Despite advancements, triboelectric sensors in HMIs still face issues like bulkiness, low durability, electric stability, and the requirement for physical contact. [26][27] In contrast, piezoelectric materials can generate electric charges in response to applied mechanical stresses, making them intrinsically self-powered for sensing applications. Specifically, when piezoelectric materials are subjected to mechanical stresses, such as those from vibrations or movements, their internal crystal structures become distorted. This process, called the piezoelectric effect, occurs as the distortion of the crystals causes a rearrangement of charges, leading to polarization from the lattice's asymmetry and generating an electric potential on the material surface. [28][29] Based on this mechanism, a variety of sensors, made of piezoelectric materials, have been employed in HMIs. [30][31][32][33][34] For example, piezoceramic sensors like Lead Zirconate Titanate (PZT), known for their high piezoelectric coefficient, have been adopted for various body monitoring applications within HMIs. Examples include hand gesture monitoring, [35,36] and robotic prosthetic hand control. [37][38] However, the primary use of those rigid piezoelectric sensors, which are attached to the human body parts, can reduce the comfortability and potentially restrict certain subtle muscle movements. Concerns regarding toxicity often arise with such materials, especially when they are used in long-term wearable devices. Additionally, piezoelectric ultrasonic sensors, made from piezoceramics and utilizing acoustic transduction for sensing, tend to be rigid. [39] They require ultrasonic coupling agents to maintain their sensitivity when applied in HMIs applications, resulting in increased inconvenience.

Some other researchers explored piezoelectric polymers for HMI applications, such as polyvinylidene fluoride (PVDF) and its copolymer poly(vinylidene fluoride-trifluoroethylene) (P(VDF-TrFE)), which offer advantages such as flexibility, biocompatibility, lightweight construction, chemical resistance, and thermal stability for long-term use. [30][31][40][41][42] However, the piezoelectric constants of those piezoelectric polymers are lower than compared to those of ceramics and single crystals. [43][44] To boost piezoelectric polymer's efficiency, researchers have explored different fabrication methods such as incorporating doped piezoceramic particles, [40,[45][46][47] or inducing microstructures into PVDF matrix to improve its piezoelectric performance. [40][41][42][48][49][50][51] Additionally, post-processing methods like annealing and poling have been utilized to further increase the piezoelectric coefficient. [52,53] Recently, piezoelectric sensors based on PVDF have been attached to human body parts to facilitate interaction with robotics. Examples include human-robot synchronization, [54] mobile robotic car control, [55] acoustics-based robotic interaction, [33] and robotic hand control. [30,33] Nevertheless, the work involving these piezoelectric polymeric sensors has mainly focused on robotic control driven by hand-manipulation, primarily through significant bending motions of the fingers or finger motion-based muscle movement. There has been limited research on piezoelectric polymers in HMIs for controlling robotic hands solely through subtle muscle movements without any hand manipulations. That requires higher sensitivity and advanced algorithms capable of accurately interpreting these minimal signals for precise control.

In HMI applications, the requirements for user comfort often entail limiting the use of sensory channels to minimize the impact on the human body. For the implementation of the HMIs with functionalities recognizing those of body motions, such as joint rotation and muscle movements, effective control algorithms are essential. [23,[32][33][34]56,57] Among different control methods, comparing the acquired sensing signals with preset thresholds has been commonly used. This involves acquiring each sensor's signal from a particular data acquisition (DAQ) channel. [29][30][31]58] Advancements have been achieved in analyzing statistical features from the complex signals, such as EMG, to monitor the specific status of muscles. [56,[59][60][61][62] However, those applications usually require multiple sensors attached to the human bodies, in which each sensor requires one acquisition channel, resulting in increasing computational complexity, higher hardware requirements, and increased bulkiness of the HMI systems.

To tackle the aforementioned challenges, in this work, we introduce a self-powered, soft, and breathable HMI based on piezoelectric sensors. These sensors not only have the capability to detect subtle muscle movements but also enable effective control of robotic prosthetic hands (Figure 1). Here, we design and fabricate polymeric materials with engineered microstructures to enhance piezoelectric performance and improve compressibility and permeability for sensing applications. We utilize the piezoelectric polymer P(VDF-TrFE), which enables a selfpowered mechanism by generating electrical signals in response to skin stimuli through its piezoelectric effect. By creating an HMI based on piezoelectric sensors, this interface precisely detects subtle changes in body and muscle movements, enabling effective communication and control of robotic prosthetic hands across various applications. Furthermore, the developed control algorithm empowers a single sensor to comprehensively control robotic hands, allowing for the execution of complex tasks, including Morse Code communication and manipulating of objects such as a cup of water (Figure 1). Therefore, the advancements in sensing materials, fabrication methods, microstructure designs, robotic control algorithms, and comprehensive application strategies lead to innovative self-powered, flexible, and breathable HMIs, specifically for robotic prosthetic hand control applications. Such designs are expected to contribute toward the realization of widespread, sophisticated, comfortable, and intuitive HMI control systems.

Results and Discussion

Material Design and Sensing Performance Evaluation

In this work, we have engineered polymeric materials with porous structures to improve their piezoelectric performance, as well as their compressibility and permeability, for use in sensing technologies. This involves the use of the piezoelectric polymer P(VDF-TrFE), which, due to its piezoelectric effect, can generate electrical signals in response to stimuli from the skin, enabling a self-powered sensing capability. The design of porous material not only improves their breathability but also increases the material compressibility, which in turn enhances the piezoelectric coefficient (See Supporting Information). The porous piezoelectric material of P(VDF-TrFE) was fabricated through a combination of spin-coating and water vapor phase separation techniques. Essentially, the water droplets penetrated the matrix of the polymer-rich P(VDF-TrFE), resulting in the formation of pores within the material during the solvent evaporation process (See Experimental Section). Such porous P(VDF-TrFE) material was characterized based on its chemical, mechanical, and electrical properties. As shown in Figure 2a, the cross-sectional of the porous P(VDF-TrFE) was captured by using scanning electron microscopy (SEM), revealing an average pore size of 7 μm. The crystalline phase composition was analyzed using X-ray diffraction (XRD). Figure 2b displays the diffraction spectra of both porous and non-porous P(VDF-TrFE) materials, which were both annealed and poled (See Experimental Section). In contrast to non-porous P(VDF-TrFE), the porous material exhibited an enhanced 𝛽 phase formation, featuring distinct characteristic peaks of 2𝜃 = 19.8°, 35.2°, 40.7°with a crystallinity of 82.5%, while the non-porous P(VDF-TrFE) had a crystallinity of 75.6%. Fouriertransform spectroscopy (FTIR) was further conducted to characterize the crystalline 𝛽 phase formation for both porous and nonporous P(VDF-TrFE) films. Based on the FTIR spectrum analysis in Figure 2c, the porous P(VDF-TrFE) exhibited an enhanced relative percentage of 𝛽 phase, up to 83.6%, while the non-porous P(VDF-TrFE) material had a 𝛽 phase relative fraction of 73%. Next, a tensile test was performed to evaluate the mechanical properties of the materials. Figure 2d,e shows the tensile test results and the corresponding Young's Modulus of both porous and non-porous P(VDF-TrFE) materials. The porous P(VDF-TrFE) exhibited a Young's Modulus of 400 MPa, while the non-porous material's Young's Modulus was determined to be 1.15 GPa. This demonstrates that the porous P(VDF-TrFE) showed better flexibility compared to the non-porous P(VDF-TrFE) material.

Additionally, it is crucial to take into account breathability as an essential factor for extended HMI applications, making it more comfortable for users. Breathability in wearable applications refers to a material's ability to allow the passage of air and water vapor. This can be achieved through porosity in the material, which forms specified channels and spaces that facilitate the movement of air and water vapor, thus facilitating breathability. [63,64] To assess the breathability of porous P(VDF-TrFE), the permeability test was performed by measuring the evaporation water weight loss. Two control groups were employed: one consisting of non-porous materials and another using a fully sealed cup. Figure 2f shows that the permeability of the porous P(VDF-TrFE) measured as 6.3 kg m -2 d -1 , which was higher than that of the non-porous material, 1.7 kg m -2 d -1 (See Experimental Section). This indicates a significantly enhanced breathability attributed to the internal porous structures within the piezoelectric material.

The piezoelectric sensor was then fabricated by encapsulating the porous P(VDF-TrFE) layer between two polydimethylsiloxane (PDMS) membranes coated with single-wall carbon nanotubes (CNTs), connected to a pair of wires functioning as electrodes (See Experimental Section). The sensing performance of the piezoelectric sensor was evaluated through conventional tests by using a shaker-based testing platform. (See Experimental Section; Figure S1, Supporting Information). Briefly, the piezoelectric sensors were mounted by a fixture using optics components, and subjected to mechanical excitation at different frequencies and forces via a shaker driven by a function generator. The resulting piezoelectric voltage outputs were captured and recorded by an oscilloscope. In Figure 3a, the porous P(VDF-TrFE)-based sensor produced an open circuit peak-to-peak voltage of 2.1 V, which was 2.6 times the output generated from the non-porous piezoelectric sensor. This is due to the internal porous structures within the P(VDF-TrFE) material. Essentially, those engineered microstructures allow for high compressibility and stress concentration and thus generate higher piezoelectric outputs (See Supporting Information). During the conventional tests, as the normal forces increased from 5 to 25 N at a constant frequency of 1 Hz, the piezoelectric sensor demonstrated a significant increase in peak-to-peak voltage outputs from 1.3 to 3.4 V (Figure 3b). The result suggests that greater input energy leads to higher piezoelectric outputs.

By defining the sensitivity as the voltage output per pressure, such as S = ΔV/ΔP, the resulting sensitivity of the developed sensing device was 38.4 mV kPa -1 , which was significantly higher than that of the non-porous P(VDF-TrFE) device, whose sensitivity was determined by 26.4 mV kPa -1 (Figure 3c). Furthermore, we evaluated the sensing performance under various bending conditions, which replicated the body motions in practical settings, as later we would wear it on the finger joints for recognizing different gestures. Four bending angles of 15°, 30°, 45°, and 60°were applied via bending a notebook where the device was attached at the middle of the hinge, resulting in the peak-topeak piezoelectric voltage outputs of 0.11, 0.18, 0.26, and 0.36 V, respectively (Figure 3d). This result indicates that the developed porous P(VDF-TrFE) sensors could be further utilized to control the angle of joints for applications. (See Figures S2 and S3, Supporting Information) To test the cycling stability of the sensor, we further evaluated the anti-fatigue performance of the device by applying a load of 10 N at 1 Hz over 10000 cycles. The result in Figure 3e shows a consistent sensing performance, demonstrating no observed degradation in electrical outputs.

Evaluation of Finger Flexion Sensing for Controlling Robotic Hands

Utilizing the developed porous P(VDF-TrFE) sensors, we demonstrated the applications in robotic hand control, enabling the robotic hand to mimic the user's gestures effectively. Here, the sensors were fixed to the surface of the human body to facilitate the HMI application. As the skin beneath the P(VDF-TrFE) sensor moves, the sensor deforms in response to this motion. Consequently, the piezoelectric material converts the mechanical motion into electrical signals, such as the piezoelectric voltage outputs. These acquired piezoelectric signals were then fed into the HMI system and interpreted as commands to control the robotic hands. Such an application of the self-powered piezoelectric sensor enables the developed HMI to work without reliance on batteries for the porous P(VDF-TrFE) sensor. The experiment setup and hardware connection are shown in Figure 4a. The sensors were placed on the finger joints and connected to the DAQ system. The acquired output signals were driven to the LabVIEW program to further control a Microcontroller unit (MCU) for executing the robotic hand. The motions of the robotic hands were triggered via piezoelectric voltage output signals generated from the developed porous P(VDF-TrFE) sensor.

Figure 4b shows the piezoelectric signals to manipulate a single finger of the robotic hand. When the sensor detected flexion of the attached finger, it generated a positively concave piezoelectric pulse to activate the flexion of the robotic finger. Conversely, when the previously bent finger extended, a negatively convex piezoelectric pulse was generated to facilitate the extension of the robotic finger. The index finger flexed and extended with angles from 15°to 60°, producing corresponding piezoelectric voltage outputs of 0.3, 0.8, 1.4, and 2.0 V. In addition, the connection of two piezoelectric sensors in parallel allowed the control of two robotic fingers using only one DAQ channel. The control algorithm, which was designed to detect any maxima and minima of all existing pulse waveforms within a piezoelectric signal, was employed for the piezoelectric signals generated from multiple sensors connected in parallel. (See Experimental Section; Figure S4, Supporting Information). This enables the identification of specific finger movements, given that each sensor produces voltage outputs within its distinct range. In Figure 4c, the index finger and the thumb, which were equipped with the developed sensors located at finger joints and connected in parallel with only one DAQ channel, moved in a sequence of motions, namely, small extension of the index finger, big extension of the index finger, and full extension with both the index finger and the thumb. This sequence resulted in generating piezoelectric voltage outputs of 0.5, 1.5, and 2.7 V, respectively. Meanwhile, the algorithm detected the peaks of each pulse waveform; then, based on those peak values, commands were assigned to activate different robotic fingers accurately with the preset thresholds.

We further attached the piezoelectric sensors to all five fingers on the user's left hand to comprehensively control the robotic hand for making the gestures of "High-five", "I", "Okay", and "Victory", by using only two DAQ channels (Figure 4d). Sensors on the ring finger and pinky finger were connected to channel 1. Sensors on the index finger, the middle finger, and the thumb were connected to channel 2. During the gesture, the hand initially started off as a fist, and specific finger extension led to specific gestures. For example, all five fingers were extended to make a gesture of "High-five", while the index finger and middle finger were extended to make the "Victory" gesture. After making each gesture, the hand went back to the initial fist position. Specifically, Figure 4d shows the piezoelectric voltage outputs for each gesturing task, in which finger extension provides convex signals and finger flexion provides concave signals. For the "High-five" gesture, two fingers from channel 1 and three fingers from channel 2 generated convex signals during the initial finger extension, followed by concave signals during the subsequent flexion. Employing the same control algorithm, specific finger movements from both channel 1 and channel 2 generate corresponding convex and concave signals during extension and flexion for the "I", "Okay" and "Victory" gestures.

In this work, to control multiple robotic fingers with only two DAQ channels, the developed algorithm identifies finger motions by detecting the peak values of the convex pulse signals. Each finger motion produces distinct ranges of piezoelectric pulse convex signal peaks. For example, to make the gesture "Okay," two fingers from channel 1 and the middle finger from channel 2 extended, resulting in a smaller convex peak compared to that in the "High-five" gesture. With a similar control mechanism, the counting task was also performed (See Movie S1 and Figure S5, Supporting Information). These results validate that attaching the developed piezoelectric sensors to human joints (such as fingers), coupled with our unique control algorithm, can effectively achieve the control of multiple actuators using minimal DAQ channels in robotic hand control applications.

Evaluation of Muscle Movement-Enabled Sensing for Robotic Hand Control

The same piezoelectric sensor demonstrated the capability to control robotic hands not only through finger joints but also by detecting muscle movements. This highlights the potential for prosthetic applications, especially for upper limb amputees, using the developed porous P(VDF-TrFE) sensor. Using a small piece of Tegaderm, the porous P(VDF-TrFE) sensing device was affixed to the surface of the bicep muscle between the antecubital and brachium sides to control a robotic hand. This design was to accurately capture the movements of the bicep muscle, facilitating precise control of the robotic hand based on the detected signals generated from the piezoelectric sensors. The schematic illustration of the bicep muscle movement-enabled sensing is shown in Figure 5a. On one aspect, based on the clenching and releasing movements of the bicep muscle with different amplitudes, the generated piezoelectric signals were encoded into Morse Code, in which the letter of the English alphabet corresponded to unique piezoelectric output peaks. Morse

Code, formed by a sequence of dots and dashes, is highly secure, making it a chosen communication method for humanrobot interaction. Additionally, by incorporating extra servo motors and arm components into the robotic hand, we introduced two additional degrees of freedom for both rotation and translation. This modification enabled the robotic hand, when integrated with a robotic arm, to react to the action of clenching and releasing the bicep muscle. As shown in Figure 5a, these bicep muscle movements, occurring within specific intervals, translate into distinct commands, allowing the robotic hand to perform a variety of motions for interacting with objects in the environment.

We demonstrated various patterns of piezoelectric outputs corresponding to different letters of the English alphabet. For example, in Figures 5b-e, the generated piezoelectric outputs encoded through Morse Code, represent the letters "D", "L", "A", and "B", respectively (See Movie S2, Supporting Information). In detail, a gentle clenching of the bicep muscle generated a piezoelectric voltage output at around 30 mV, representing a dot in Morse Code, while an acute clenching produced a large pulse of around 160 mV, symbolizing a dash. Multiple pulses were produced to convey desired dots and dashes in Morse Code. A specific command, such as two large pulses in a short period, triggered the robotic hand to perform the corresponding hand sign of the alphabet.

In addition, we demonstrated the process by using the example of performing the daily task of "drinking a cup of water". This involves controlling the robotic hand with the integrated arm to execute a sequence of actions: moving forward to reach the cup, gripping the cup, tilting the cup for drinking, releasing the cup, and finally, returning to its initial position by rotating backward (See Movie S3, Supporting Information). Figure 5f shows the piezoelectric outputs due to the bicep clenching at different times within the assigned period. The precise control of the robotic arm was achieved by decoding the pulse numbers of the pulses of around 50 mV generated during gentle clenching within the 5 s timeframe. Specifically, two pulses switched the hand's status between open or closed, three pulses controlled the angle of the wrist to be straightened or tilted, and four pulses or five pulses activated the robotic elbow to rotate 15°forward or 15°backward, respectively.

To further conduct an assessment of the developed sensor's capability for controlling automated prostheses with different muscle groups, we also attached the sensor to the surface of the forearm muscle near the extensor carpi ulnaris, close to the elbow side. This allowed us to control the robotic hand for making different gestures (See Figure S6 and Movie S4, Supporting Information). Performing repetitive clenching and releasing of the forearm muscle resulted in the piezoelectric sensor producing pulses due to the change of muscle status between clenching and relaxing. The piezoelectric signals produced a peak-to-peak voltage output of 115 mV when the forearm muscle was in a clenched state (See Figure S6 and Movie S4, Supporting Information). Using the pulse numbers from the generated piezoelectric signals, various robotic hand gestures were triggered. Such piezoelectric sensors are simple to use, and those results show their potential for controlling prosthetic hands by clenching forearm muscles to create hand gestures for daily communication. In addition, the response time of the piezoelectric output voltage signals, generated from various body parts (finger joints, bicep muscles, and forearm muscles), was measured. The signals exhibited an average response time of 0.065s, demonstrating rapid responsiveness of the piezoelectric sensors. (See Table S1, Supporting Information) Overall, those results demonstrate the capability of porous P(VDF-TrFE) sensors as a new HMI for robotic hand control through muscle movement detection.

Conclusion

We have developed a self-powered, soft, and breathable HMI by using porous P(VDF-TrFE) materials for robotic prosthetic hand applications. By leveraging the porous structures within the polymeric material, the P(VDF-TrFE) sensor exhibits a significantly enhanced sensitivity, flexibility, and permeability, highlighting outstanding HMI capabilities in effectively controlling the bending motion of the finger joints and the clenching motion of the forearm and bicep muscles. The developed control algorithm empowers comprehensive control of robotics using fewer sensors and minimized DAQ channels, facilitating applications such as gesturing, Morse Code communication, and manipulating a cup of water. Such functionality of the control algorithm is based on the high sensitivity of the porous P(VDF-TrFE) sensor, enabling it to identify both the pulse numbers and the amplitudes of the output signals in response to subtle changes in body and muscle movements. The sensor is self-powered for operation, allowing it to be worn on the body without needing an external power source, thereby increasing convenience. The excellent breathability of the sensing material due to its porous structures makes it well-suited for long-term use. Moreover, our developed sensors exhibit scalability, allowing them to be positioned on various human body parts, such as the surfaces of body joints or muscle groups. This versatility is particularly beneficial for amputees who need precise control over prosthetics. Such HMIs pave the way toward self-powered, soft, and breathable biomimetic systems, contributing to the evolution of prosthetics in the future.

Experimental Section

Fabrication of Porous P(VDF-TrFE) Film: The fabrication process began with the preparation of the P(VDF-TrFE) solution by dissolving P(VDF-TrFE) powder (Arkema 70/30 mol) in the N, N-Dimethylformamide (DMF) solvent procured from Sigma-Aldrich with the concentration of 15 wt.% over 8 h of continuous magnetic stirring. The choice of 15 wt.% concentration of the P(VDF-TrFE) solution was based on our previous studies. [42,51,[65][66] A non-porous P(VDF-TrFE) film was created by spincoating 1 mL of the P(VDF-TrFE) solution onto a glass slide at 1000 rpm, and the film was then heated in an oven at 50 °C for 10 min. Subsequently, the spin-coating procedure was repeated at the same speed by directly applying a 2 mL solution of P(VDF-TrFE) onto the solidified non-porous P(VDF-TrFE) layer. The porous layer was then formed by immersing the material in a water bath with a relative humidity of 90%. Next, the porous P(VDF-TrFE) film was annealed at 135 °C for 2 h. Then, a 60 min electrical poling procedure was conducted with an applied electric field intensity of 80 V μm -1 .

Assembly of Porous P(VDF-TrFE) Sensing Device: The PDMS layers were first prepared by spin-coating a mixture of silicone elastomer base and agent (10:1; Dow Sylgard 184) at 500 rpm and curing at 60 °C for 20 min. Next, the CNT solution was then prepared by 30 min-long ultrasonication of the mixture of CNT ink (Sigma-Aldrich) with 1 wt.% sodium dodecyl sulfate solution (Sigma-Aldrich), isopropanol (Sigma-Aldrich), and deionized (DI) water. Then, the vacuum filtration method was used to create the CNT percolating through the Polytetrafluoroethylene (PTFE) filters (47 nm pore diameter). Afterward, the patterned CNT electrodes were obtained by pressing the CNT mats on the surface of PDMS using a mask (Mylar film). The fabricated porous P(VDF-TrFE) film was sandwiched between two CNT electrodes coated PDMS layers. Finally, the CNT electrodes were connected to the cable wires.

Material Characterization Methods: Both samples of the porous and non-porous P(VDF-TrFE) films were prepared using the freeze-fracture method. Then, the morphologies of the cross-section for the porous and non-porous samples were observed by using scanning electron microscopy (SEM, JEOL JSM-9700F instrument). The X-ray diffraction (XRD) analysis was conducted by utilizing a PANalytical Empyrean X-ray diffractometer equipped with Cu-Mo radiation, covering the 2𝜃 range from 10°to 70°. The Fourier transform infrared (FTIR) measurement was conducted using the Agilent Cary 620 FTIR Imaging Microscope with Cary 670 Bench equipment to form the infrared absorption spectrum. Finally, the Young's Modulus of the porous and non-porous P(VDF-TrFE) films were evaluated through the tensile tests by using a motorized force tester (MARK-10). The gas permeability experiment was performed by utilizing cups filled with DI water and covered with test materials at openings. The experiment consisted of two control groups: one with an opening sealed using a nonporous P(VDF-TrFE) film and another with an opening totally sealed with plastic wrap. The test group involved containers filled with DI water and their openings sealed using a porous P(VDF-TrFE) material. The containers were positioned within a chamber that maintained a consistent relative humidity of 20-30% relative humidity. The permeability was determined by monitoring the weight loss of DI water every 24 hours over a period of 7 days.

Sensing Performance Evaluation Methods: A conventional testing setup was developed to evaluate the sensing performance of developed sensors. Here, a customized fixture using optics fixing parts was developed to affix the sensing device made of the porous P(VDF-TrFE) film. A shaker (Modal Shop) was connected to the wooden shaker stinger with a cross-sectional area of 2 cm × 2.5 cm. The shaker was actuated by a sinusoidal signal from a function generator (Keysight) with a power amplifier's (Modal Shop) assistance to apply various loads to the testing devices. The oscilloscope (Keysight) captured the real-time piezoelectric output produced by the porous and non-porous P(VDF-TrFE) sensing devices.

Hardware Connection and Control Algorithm for HMI: For the hardware connection, the piezoelectric sensors were connected to a preamplifier, and the outputs were collected with a DAQ card (National Instrument) and analyzed by LabVIEW programs. To control the robotic hand via finger flexion, multiple sensors were connected in parallel to a single DAQ channel. A program was developed to compare the positive and negative peak values of the superimposed piezoelectric voltage outputs when they reach their peak values (See Figure S2, Supporting Information). Such an algorithm could prevent the system from mistakenly identifying the incorrect finger, especially when the peak value of the signal undergoes changes in the initial phase of the user's finger movement with the piezoelectric signal still rising or dropping without having yet reached the peak. Specifically, four threshold voltages were defined for each finger (two positive and two negative values), namely, V p_upper , V p_lower , V n_upper , V n_lower . V t_upper and V t_lower were trigger voltages to detect if the piezoelectric signal would start rising or dropping due to finger movements. While the peak values from a particular finger were detected, two positive threshold voltages V p_upper and V p_lower were employed to identify which finger was flexing from the concave waveform, while the negative thresholds V n_upper and V n_lower were utilized to identify which finger was extending from the convex waveform. When a concave signal waveform was detected within the positive threshold value, such as V p_lower < V maximum < V p_upper , then the corresponding robotic finger would flex, and vice versa.

In order to control the robotic hand through muscle clenching movements with a single piezoelectric sensor, two variables were defined to track the pulse numbers of the output signals associated with mild and acute clenching motions, namely, Count mild and Count acute . Two thresholds were set accordingly, namely, V mild and V acute to identify if the clench was mild or acute by comparing them with the positive peak values of piezoelectric outputs. Additionally, a timer that triggered a recording of a few seconds of the signal was activated whenever a clench signal was generated. When the timer was active, the count value would start accumulating whenever the peak value of a clench was detected and within the corresponding threshold range, such as if V maximum > V acute , then Count acute + = 1. The accumulation ended whenever the timer stopped, and then the pulse numbers were decoded into commands to control the robotic hand. Specifically, the number of mild clenches would be decoded into commands for controlling the robotics to perform different motions. Meanwhile, the presence of the acute clenches would turn the control pattern into Morse Code, such that a small piezoelectric pulse due to a small clench would be translated into a dot, and a large piezoelectric pulse due to an acute clench would be translated into a dash. Acute clenching and releasing twice would be recognized as the "action" command to trigger the translation of the maintained dots and dashes into letters. The experiments involving human subjects were performed with the full, informed consent of the volunteers, who are also authors of the manuscript.