Bioinspired Rheological Sensing for Robotic Liquid Identification in Sealed Containers via Ultrafast Incipient Slip Detection

ABSTRACT

The rapid, non‐invasive identification of sealed liquids is crucial for ensuring counter‐terrorism security and improving operational efficiency in various scenarios such as airports and large‐scale events. Existing technologies are limited by invasive procedures or reliance on vision. In this research, we developed a bioinspired iontronic tactile sensor with an interlocked “protrusion‐groove” structure. Its unique biomimetic structure allows for the detection of static pressure and incipient slip within a single device, with a peak sensitivity of up to 1873.83 kPa^{− 1} in the low‐pressure regime and a slip response time as fast as 46 ms. A robotic fingertip integrated with this sensor can effectively capture the transient signals generated by shaking liquids. Furthermore, by extracting physical features from the signals and applying a Random Forest classifier, we developed a non‐visual liquid identification system that achieves an accuracy of 99.04% ± 0.47% for different liquids. This work provides a practical solution for non‐visual liquid identification and demonstrates a robust approach for enabling robots to achieve multimodal tactile perception.

A bioinspired sensor detects incipient slip and static pressure via a unique protrusion‐groove structure, enabling a robotic fingertip to capture liquid sloshing dynamics. The extracted physical features are processed by a Random Forest classifier, achieving 99.04% ± 0.47% accuracy for non‐visual identification of liquids within sealed containers.

1. Introduction

Quickly and non‐invasively identifying liquids in sealed containers is crucial for ensuring counter‐terrorism security and improving operational efficiency in various scenarios such as airports and large‐scale events [1, 2, 3]. However, existing methods have inherent limitations due to their sample damage and contamination risk, while vision or spectroscopy‐based non‐contact technologies often fail in the presence of opaque containers or in low‐light environments [4, 5]. Notably, humans provide a solution that does not rely on vision. By shaking a container and sensing the tactile feedback it transmits, one can determine the presence, state, and even approximate viscosity of the liquid inside [6, 7]. This reveals that transient tactile signals are crucial for unlocking non‐visual, non‐invasive perception, providing an effective alternative to visual‐based liquid identification (Figure 1a).

FIGURE 1.

Bionic tactile sensor design inspired by human skin. (a) A robot equipped with tactile e‐skin identifies liquids inside opaque containers through non‐invasive shaking, analyzing motion‐derived signals to infer physical properties such as viscosity. (b) Schematic diagram of the dermis‐epidermis structure of human skin and the ion channel activation mechanism under mechanical stimulation. (c) Schematic of a human hand and a dexterous hand using the tactile sensor to grasp and slide an object, along with a signal diagram of the incipient slip. (d) Operating principle of the bioinspired sensor based on PVA material, where normal and shear forces increase the contact area of the electrode ion layer to enhance the double‐layer capacitance. (e) Schematic of the robotic tactile perception of hardness and surface macro textures. (f) Non‐invasive liquid identification using coupled normal and shear force signals combined with a Random Forest classifier, suitable for sealed container inspection without visual input.

Robotic tactile sensing technology has made significant progress in recognizing object geometry [8, 9] and surface properties [10, 11]. Notably, specific innovations such as laser‐induced gradient micro‐pyramid structures have further endowed flexible sensors with exceptional sensitivity across ultra‐wide linear ranges [12]. However, perceptual capabilities are still primarily limited to static normal pressure [13, 14] or macroscopic sliding detection [15, 16]. There is still a lack of efficient methods for detecting incipient slip signals, which are the tiny relative displacements at the contact interface occurring before full sliding begins [17, 18, 19, 20]. Incipient slip is not only the key mechanism that allows humans to achieve precise anti‐slip gripping but also carries vital information about the material and internal state of an object [21].

Consequently, flexible tactile sensors are increasingly being explored for advanced recognition tasks such as liquid identification. Recent studies have demonstrated the classification of fluids using electromechanical coupling at solid and liquid interfaces [22], thermal sensing [23, 24], or by analyzing the frequency decay of vibrations induced by active shaking [25]. While these methodologies offer effective chemical, thermal, or macroscopic vibrational cues, they often rely on sensors that primarily capture bulk resonance and lack the ability to monitor static grip states. In contrast, when a fluid shifts during dynamic motion, its changing inertia continuously induces microscopic incipient slip at the robotic grasp interface. Resolving these highly localized and transient shear signals, alongside steady‐state normal pressure, provides a distinct and richer mechanical perspective to characterize intrinsic properties such as viscosity. The ability to resolve these transient slip signals with high temporal resolution significantly enhances the capacity of tactile systems to safely handle and identify diverse fluid containers without visual guidance.

Currently, research on slip perception has made significant progress in the macroscopic sliding phase. Typical methods include vibration response‐based recognition [26, 27, 28, 29], array‐based pixel displacement tracking [30, 31, 32], and signal processing techniques such as wavelet transform for extracting shear force characteristics [33]. However, these methods largely rely on noticeable displacement to generate detectable signals, leading to delayed tactile feedback and often requiring complex algorithms and post‐processing, which makes them difficult to meet real‐time control requirements. Some studies have attempted to detect incipient slip through micro‐vibration analysis [34, 35, 36], shear strain field reconstruction [37, 38, 39], or vision‐assisted deformation analysis [40, 41]. However, these methods are mostly indirect mechanisms and still fall short in terms of high sensitivity and real‐time detection of transient shear strain at the interface. In applications emphasizing fast response and non‐visual environments, such as sealed liquid state discrimination and dynamic grasp adjustment, existing technologies face significant challenges in terms of timeliness and adaptability.

From the perspective of biological perception mechanisms, human sensitivity to incipient slip originates from the synergistic coding mechanism between multiple types of mechanoreceptors in the skin, particularly the fast‐adapting type Ⅰ (FA‐I) mechanoreceptors and slowiy adapting type Ⅰ (SA‐I) mechanoreceptors, which are densely distributed in the fingertip region [42, 43]. When the interface friction is insufficient to balance the shear force, a micro‐slip occurs at the outer edge of the contact area, inducing transient strain in the local skin [44]. The FA‐I mechanoreceptors respond rapidly to strain stimuli via ion channels, triggering an explosive neural discharge within approximately 100 milliseconds [45, 46, 47]. This response provides a feedforward alert for the impending slip event [48]. Meanwhile, the SA‐I mechanoreceptors continuously encode steady‐state pressure, working together to maintain the stability of tactile state judgments (Figure 1b) [49]. In contrast, most existing electronic skin systems are only capable of detecting macroscopic slip, and in order to achieve multimodal sensing, they require the integration of multiple sensing units. This results in complex structures, increased power consumption, and greater difficulty in algorithm analysis, which limits their application in platforms such as soft robots and dynamic interaction systems. Therefore, the development of a tactile mechanism that integrates a single device capable of high‐sensitivity coupling perception of normal and shear forces, along with rapid response capability, is crucial for expanding the sensory functions of robots (Figure 1c).

In response to the challenges mentioned above, this paper, inspired by the synergistic coding mechanism of SA‐I/FA‐I mechanoreceptors in human skin, proposes an ion‐based tactile sensor with a bioinspired protruding electrode‐microgroove structure. This sensor enables high‐sensitivity detection of normal pressure and dynamic incipient slip in a single device, providing an innovative solution for liquid recognition in non‐visual environments. The sensor, through its bioinspired design of dome electrodes and microgroove ion‐conductive layers, simulates the epidermal response to static pressure and the selective conduction of transient shear strain in the dermis. During the incipient slip phase, the abrupt change in the contact area of the interface's micro‐protrusions triggers millisecond‐level capacitive transitions, simulating the fast response mechanism of the FA‐I mechanoreceptors; sustained normal pressure induces stable capacitive changes, corresponding to the steady‐state encoding characteristics of the SA‐I mechanoreceptors (Figure 1d). Experiments show that the sensor has a normal pressure sensitivity of 1873.83 kPa^{− 1} and an incipient slip response time of 46 ms, effectively capturing microscopic deformations and distinguishing object hardness and surface macro textures (Figure 1e). By extracting a 30‐dimensional physical feature space and utilizing a physics‐informed Random Forest ensemble classifier, a non‐visual liquid recognition system was developed, achieving a robust classification accuracy of 99.04% ± 0.47% for five different liquids (Figure 1f). This system demonstrates significant potential for application in sealed, opaque containers and the perception of complex viscosity‐related fluid behaviors without visual input. Ultimately, this study introduces a biomimetic tactile sensor that leverages transient incipient slip signals, enabling the non‐destructive identification of liquid types through dynamic fluid‐structure interaction. It powerfully demonstrates the potential of transient tactile responses in non‐visual object perception and provides a new technical path and solid foundation for the development of advanced tactile systems in soft robotics, human‐robot interaction, and complex environmental sensing.

2. Results and Discussion

2.1. Device Design Based on Skin's Mechanical Structure

Inspired by the biomechanical conduction mechanisms in the multi‐modal tactile perception of human skin, we designed and constructed a hierarchical bionic flexible tactile sensor that is sensitive to incipient slip detection. Based on this mechanism, we built a bionic structure with three functionally differentiated layers: the top layer is an electrode structure with metal protrusions, corresponding to the stress dispersion function of the epidermis (Figure 2a); the middle layer consists of a micro‐grooved PVA‐H₃PO₄ ionogel, serving as a strain‐guiding component analogous to the dermal collagen network, which exhibits high sensitivity to shear strain (Figure 2b); and the bottom layer is a flat metal electrode, imitating the deeper dermal tissue and providing a stable electrical reference.

FIGURE 2.

Device Microstructure Characterization and Mechanical Simulation. (a) Optical image of the top PVA electrode layer and micro‐protrusion structure. (b) Optical image of the PVA‐H₃PO₄ ion layer and micro‐groove structure. (c) Schematic diagram of the device's hierarchical structure and deformation mechanism under normal/shear load. (d) Finite element simulation model of normal and shear force loading. (e) Morphological changes at the electrode‐ion layer interface under normal‐shear synergistic loading (front view/side view).

In terms of device fabrication, we first utilized soft lithography to prepare a PDMS template with micro protrusions followed by thermal metal evaporation. This template was then cast with a PVA solution to form metal protruding electrodes on a top PVA substrate (Figure S1). The middle layer ionogel was subsequently cast using a PVA‐H₃PO₄ solution where the introduction of phosphoric acid significantly enhanced the ionic conductivity. Lastly, the bottom flat electrode was prepared using the same PVA transfer process. Crucially, this study employed a homogeneous structural matrix comprising the top PVA substrate, the middle PVA‐H₃PO₄ ionogel layer, and the bottom PVA substrate. This integrated strategy effectively mitigates the severe modulus mismatch commonly encountered at the interfaces between previously reported elastomers and ionogels. As highlighted by recent research advances [50], this homogeneous material strategy effectively prevents local stress concentration and interfacial delamination during dynamic shear deformation. Furthermore, peel tests demonstrated that the adhesion strength between the metal electrodes and the PVA substrate in this architecture was significantly superior to that of PDMS and gold systems (Figure S2). This macroscopically enhanced mechanical stability is fundamentally governed by the high degree of molecular similarity among the constituent layers [51]. These molecular level interactions transform the multilayer stacked architecture into a mechanically continuous entity enabling the intermediate ionogel layer to achieve a high degree of structural coupling with the top and bottom layers. Ultimately, this integrated PVA strategy constructs a robust biomimetic sensing interface that ensures highly reliable dynamic responses even under conditions of repeated loading and dynamic friction.

2.2. Skin‐Inspired Sensing Principle

The structural design of the sensor is characterized by a biomimetic three‐layer architecture (Figure 2c). Specifically, the top metal micro‐protrusion electrode mimics the stress‐dispersion function of the human epidermis; the intermediate micro structured ionogel layer replicates the directional conductivity of the dermal collagen network under strain; and the bottom planar electrode serves as the signal acquisition substrate. To minimize the initial baseline capacitance and enhance structural integrity, hollow PDMS spacers are integrated at the device periphery, effectively preventing direct contact between the electrodes in the unloaded state. The multi‐interfacial physical interactions, governed by spatial dimensional mismatches, facilitate a continuous evolution of both mechanical deformation and morphology. The authentic three‐layer architecture and the preservation of micro‐air gaps in the assembled sensor were further validated via cross‐sectional SEM (Figure S3), with detailed geometric dimensions and material specifications summarized in Table S1. The sensing mechanism of the device across the entire loading range is schematically illustrated in Figure S4. During Stage I of normal preload and initial contact differentiation, the application of normal pressure causes the peripheral spacers to yield and facilitates the closure of the internal architecture. Due to the intentional spatial dimensional mismatch between the top micro‐protrusion array and the ionogel microgroove array, the top interface exhibits non‐conformal contact and differentiates into two distinct states. A portion of the micro‐protrusions presses directly against the ridges of the ionogel to bear the primary normal load. Concurrently, misaligned protrusions enter the interior of the microgrooves. Since the protrusion height is less than the groove depth, these elements remain in a freely suspended state, maintaining significant micro‐scale air gaps with the groove base and sidewalls. This discrete multi‐point contact configuration establishes the baseline electric double‐layer (EDL) capacitance of the system, which increases synchronously with the applied pressure. This phase effectively simulates the encoding mechanism of SA‐I receptors for static pressure as governed by Equation (1).

$$\mathrm{C}={\mathrm{C}}_{\mathrm{p}}+1/\left(1/{\mathrm{C}}_{\mathrm{top}}+1/{\mathrm{C}}_{\mathrm{bottom}}\right)$$ (1)

Here, C_top and C_bottom represent the EDL capacitance values at the interfaces of the top and bottom electrodes with the gel layers, respectively, which depend on the contact area and applied pressure. C_p represents the coupling capacitance between the two electrodes (Figure S5) [52, 53, 54]

Subsequently, finite element simulations (Figure 2d) and experimental observations (Figure 2e) further elucidate the microscopic dynamic interaction mechanisms under shear force. In the initial phase of applying shear force, the system enters Stage II, Shear “Stick” and Geometric Approach. At this point, the protrusions pressing against the gel ridges provide the initial surface friction, maintaining the system in a mechanical “stick” state. Consequently, the soft gel matrix undergoes macroscopic shear tilting (bulk deformation), leading to a reduction in the effective thickness of the gel layer. Driven by this geometric coupling effect, the suspended protrusions within the grooves begin to displace, laterally approaching the sidewalls and longitudinally nearing the groove bottom. Once the shear stress exceeds the critical threshold of surface static friction, the system rapidly transitions to Stage III, Incipient Slip and Transient Geometric Collapse. In this stage, the previously suspended protrusions abruptly reach the bottom and forcefully collide with the soft groove sidewalls, instantaneously triggering lateral buckling and local geometric collapse of the sidewalls, thereby tightly enveloping the 3D surfaces of the protrusions. The comparison between Figures S6 and S7 further confirms that the increase in contact area under pure normal loading is extremely limited, whereas the coupling intervention of shear force significantly amplifies the interfacial deformation. This results in a non‐linear surge in the effective contact area, dominated by local collapse, at the sub‐millisecond level. It is precisely this “geometric amplification effect,” generated by the spatial dimensional mismatch combined with shear loading, that effectively transforms the microscale incipient slip into an intense transient capacitance spike signal, successfully emulating the ultra‐high sensitivity response of FA‐I mechanoreceptors to dynamic slip features.

Notably, this transient sensing mechanism based on local geometric abrupt changes exhibits high sensitivity to physical boundary conditions. The triggering of the transient spikes fundamentally relies on the unsaturated space within the microgrooves. Under extreme normal preload, the load‐bearing gel ridges undergo severe longitudinal compression. Based on the Poisson effect of incompressible materials, the compressed sidewalls subsequently exhibit significant lateral expansion. This expansion prematurely consumes the air gaps inside the grooves before shear slip occurs, effectively submerging the suspended micro‐protrusions in advance. When the internal spatial gaps reach a saturated state, the system completely loses the deformational degrees of freedom required for transient sidewall collision and geometric collapse, leading to the severe suppression of the slip spike signals under high preloads. Therefore, maintaining a moderate normal preload is the optimal condition to balance macroscopic geometric interlocking with the preservation of microscopic misaligned gaps. Furthermore, sufficient surface friction is crucial for accumulating the necessary elastic strain energy prior to slipping, while a higher shear rate imparts a greater strain rate to the system, making the collision between the protrusions and sidewalls more sudden and intense, which in turn generates a steeper capacitive pulse response. The optimization of these boundary conditions further enhances the sensor's early‐warning capabilities in practical grasping and slip detection applications.

2.3. Capacitive Sensing Characteristics of Normal Pressure Response

The figures of merit for pressure sensors include sensitivity, measurement range, response time, and detection limit. Sensitivity (S) is defined as

$$\mathrm{S}=\left(\mathrm{\Delta }\mathrm{C}/{\mathrm{C}}_0\right)/\mathrm{\Delta }\mathrm{P}$$ (2)

ΔC is the change in capacitance, C₀ is the initial capacitance, and ΔP is the applied pressure. To optimize the sensor's performance, the radius of the fixed top electrode micro‐protrusions is 150 µm, with a pitch of 150 µm. The study investigates the effect of the ionogel groove layer's radius and depth on sensitivity and cyclic stability. When the groove layer radius is 150 µm, sensors with different depths (70, 140, 210 µm) were selected for sensitivity and cyclic repeatability testing. The results show that as the depth increases, the sensitivity increases proportionally. When the depth is 210 µm, the sensitivity reaches its maximum value of 757 kPa^{− 1}. Similarly, when a groove radius of 250 µm is selected, sensitivity also increases with depth, and at a depth of 210 µm, the sensitivity reaches the highest value of 1873.83 kPa^{− 1} (Figure S8). Therefore, it can be observed that when the groove radius of the ion layer is greater than the radius of the protrusion of the upper electrode, the contact area during pressure loading increases, which enhances the pressure sensing performance.

The ionogel layer assembly, featuring grooves with a radius of 250 µm at various depths, also demonstrated excellent cycling performance under cyclic loading and unloading at 65 kPa (Figure S9). Consequently, the final selected structure comprised upper protruding electrodes with a radius of 150 µm and spacing of 150 µm, and an intermediate gel layer with grooves of 250 µm radius and 250 µm spacing. To establish the optimal electrochemical operating regime for this configuration, the frequency‐dependent response of the sensor was evaluated across a broad range from 100 Hz to 500 kHz, as illustrated in Figure S10. All subsequent performance characterizations were conducted at a measurement frequency of 1 kHz and an excitation amplitude of 1 V. This specific frequency regime was selected because it successfully minimizes electrode polarization while ensuring rapid ionic redistribution during transient microstructural deformation.

Furthermore, the effective sampling rate of the system was maintained at 43 Hz. To meticulously preserve the transient mechanical features generated by incipient slip for subsequent classification via the Random Forest algorithm, no additional low‐pass filtering was introduced at the hardware level. Under these optimized physical and electrical boundary conditions, the sensor exhibited the highly responsive pressure characteristics detailed below.

To objectively evaluate the static performance of the sensor, a continuous piecewise linear calibration was conducted across the entire measurement range, as illustrated by the solid red lines in Figure 3a. The normal sensitivities are 1175.16 kPa^{− 1} for the range of 0–9.34 kPa, 154.52 kPa^{− 1} for 9.34 to 90.23 kPa, and 32.81 kPa^{− 1} for 90.23 to 300 kPa. Even under substantial preload pressures of 20 or 180 kPa, the sensor generated clear responses to minute pressure increments of 0.05 and 0.1 kPa, as demonstrated in Figure 3b. Response time tests involving the loading and unloading of 10 and 20 g weights revealed loading times of approximately 22 ms and unloading times of approximately 11 ms, which are depicted in Figure 3c. Furthermore, the sensor could effectively detect normal pressures as low as 2 Pa, as shown in Figure 3d. It also exhibited excellent repeatability. Specifically, stability was maintained under 40 kPa cyclic loading and unloading tests, as presented in Figure 3e. While the spacing between cyclic curves increased slightly with rising pressure levels at 50, 100, 150, 200, and 250 kPa, stable response characteristics were still preserved at each pressure level during repeated testing, as indicated in Figure 3f. The response of the sensor under different environmental conditions is illustrated in Figure S11. While elevated temperature and humidity increase the sensitivity due to enhanced ionic mobility and material softening, the initial baseline remains stable. Additionally, Figure S12 shows the signal variations during bending and stretching, indicating that the device maintains regular and predictable performance even under complex mechanical deformations. To evaluate the reliability of the fabrication process, systematic statistical analyses were conducted on ten independent samples from the same batch and five independently prepared batches, with the results presented in Figure S13. For intra‐batch samples, the relative standard deviations (RSDs) for initial capacitance and response at 100 kPa were 0.62% and 1.20%, respectively, while inter‐batch evaluations yielded corresponding values of 0.87% and 2.37%. These deviations primarily originate from the non‐standardized manual assembly and structural alignment inherent in laboratory‐scale fabrication. While these fluctuations remain within acceptable limits for sensing applications, the use of high‐precision automated manufacturing in the future could further minimize process‐induced errors and optimize signal uniformity. The mechanical stability of the sensor was further evaluated through rigorous static and dynamic tests, as detailed in Figures S14 and S15. The device exhibits a highly stable static creep rate of 0.22% and excellent baseline recovery during dynamic cycling, effectively mitigating the inherent material hysteresis. Furthermore, the sensor maintained over 98% signal retention after 2,000 continuous loading cycles. While this 2,000‐cycle test serves as a representative demonstration rather than the ultimate operational limit, it thoroughly verifies the structural reliability of the underlying design principle. Further optimization of the assembly processes will provide additional room for improvement to meet the rigorous demands of future practical applications. The long‐term stability of the sensor over a period of 25 days in an open room‐temperature environment is characterized in Figure S16. The results indicate that the initial capacitance baseline remains highly stable, while the response amplitude shows a gradual decrease after 10 days. This amplitude degradation is fundamentally attributed to inherent water evaporation within the unencapsulated hydrogel system. While the current device demonstrates reliable short‐term dynamic capture capabilities, addressing this environmental sensitivity through ultra‐thin elastomer encapsulation or water‐retaining dopants remains a key direction for future practical deployments.

FIGURE 3.

Characterization of normal force sensing performance. (a) Relative capacitance response, piecewise linear fitting curves, and pressure‐dependent sensitivity variation over the 0–300 kPa range. (b) Capacitance response resolution for 0.05/0.1 kPa micro‐pressure under 20 and 180 kPa reference pressures. (c) Loading‐unloading response time at 1 kPa/2 kPa. (d) Capacitance response characteristics under ultralow pressures of 2 Pa/5 Pa. (e) 100‐cycle stability test at 40 kPa pressure. (f) Cyclic loading repeatability across multiple pressure levels (50–250 kPa).

To comprehensively evaluate the overall capabilities of the device, we have summarized the key performance parameters of the proposed sensor and benchmarked them against recent representative works in Table S2. To explicitly delineate our contribution beyond standard sensitivity metrics, the table highlights the distinct slip sensing strategies and advanced functional capabilities of these sensors. Distinct from pressure sensors that primarily focus on steady‐state measurements or rely on decoupled sensor arrays, our device is specifically designed for transient signal detection. By leveraging transient geometric collapse within a single sensing channel, it achieves exceptional dynamic signal resolution. This highly responsive characteristic establishes a robust technical foundation for capturing complex high‐frequency mechanical events in subsequent practical tasks, such as non‐visual liquid identification. Owing to its flexible and miniaturized design, the device shows significant potential for wearable physiological signal monitoring. When placed on the radial artery, the sensor clearly monitors the pulse at approximately 72 beats per minute. The amplified waveforms resolve distinct percussion, tidal, and diastolic waves, all of which are consistent with clinical cardiovascular features (Figure S17). The ability to distinguish these subtle features confirms the high signal‐to‐noise ratio of the sensor and its capacity to capture minute dynamic pressures with high fidelity [55, 56]. Positioned on the throat region, it regularly detected swallowing signals (Figure S18). When mounted on a finger joint, it exhibited stable responses during both dynamic bending and static posture holding, verifying its excellent mechanical flexibility and practical utility in motion tracking (Figure S19).

2.4. Study on the Characteristics of Incipient Slip Detection Induced by Shear Force Through Capacitance Variation

Although both normal and shear forces induce an increase in capacitance, their underlying mechanisms are intrinsically distinct. Under constant normal pressure, the shear force generates lateral stretching within the micro‐groove ion layer, thereby expanding the effective contact area through tension‐induced geometric deformation. This response manifests as a rapid capacitive jump without any actual macroscopic sliding.

In our experiments, a constant normal pressure was first applied to the sensor surface and maintained stably. Subsequently, a lateral shear force was applied to ensure no relative slippage, while keeping the device stationary to fully record the process of changes in the contact area between the device layers, from static friction to dynamic friction, and from incipient slip to macroscopic sliding (Figure 4a). As shown in Figure 4b, after applying the normal force (Stage I), the application of the lateral force does not induce slip, but the capacitance value increases rapidly and remains stable under the subsequent constant shear force (Stage II), with the transition from Stage I to Stage II taking 46 ms. This phenomenon is attributed to the stretching effect induced by the lateral shear force within the grooved ionogel layer, which causes a reduction in layer thickness and an expansion of the microgrooves, leading to deeper engagement between the upper electrode protrusions and the ion layer, thus increasing the EDL capacitance. Therefore, even without macroscopic sliding, the capacitance value increases significantly.

FIGURE 4.

Shear force and slip detection performance. (a) Schematic of the evolution of the electrode‐ion layer contact area caused by shear force under constant normal force. (b) Reproducible response (3 cycles) of shear force step loading under non‐slip conditions. (c) Capacitance transition characteristics under shear force loading at the slip critical state (3 repetitions). (d) Relative capacitance curves of shear force after applying 100/200/500 g downward load with fixed normal force. (e,f) Real‐time slip detection when gripping a cylindrical object/apple with the installed sensor on the index finger (3 repetitions). (g) Slip‐triggering curves on surfaces with 200 to 800 grit sandpaper under constant normal force. (h,i) Capacitance response curves and contact heatmaps during the slip process of hard ball/soft ball gripping.

To validate the physical necessity of the biomimetic architecture and reveal the core features governing the transient slip response, a series of structural ablation experiments are detailed in Figure S20. By comparing four distinct interface configurations (flat‐flat, flat‐groove, protrusion‐flat, and protrusion‐groove), the experiments confirm that unsaturated three‐dimensional air gaps and spatial dimensional mismatch are the core prerequisites for triggering the transient geometric collapse. Furthermore, to highlight the unique advantages of this specific structure, Table S3 presents a systematic comparison of initial capacitance and sensitivity among these ablated interface configurations, confirming the advantages of the dual‐microstructured design in avoiding contact saturation and enhancing dynamic resolution. Figure 4c further illustrates the process where the lateral shear force gradually increases after applying the normal pressure (Stage I) until slip is triggered. The onset of slip (Stage II) is marked by a prominent capacitance peak, which then changes as the shear force decreases (Stage III). Figure S21 examines the performance of the incipient slip signal under different conditions, and the results show that the signal characteristics remain consistent regardless of the direction of the lateral force. Meanwhile, under fixed upper electrode layer radius (150 µm) and spacing (150 µm), the capacitance response to increasing shear force was compared when the radius of the ionogel grooves was 0, 150, and 250 µm. When the radius is 0 µm (flat), the normal force response is rapid, but the capacitance change under shear force is almost negligible; when it is 150 µm, the capacitance change is small and slow as the shear force increases; when the groove radius is 250 µm, the capacitance response becomes dramatically faster and more pronounced under shear force loading (Figure S22).

The sensor was placed between the bracket holder and the bottle to apply a constant normal force, and 100, 200, and 500 g weights were cyclically loaded. The experimental results showed that under the constant normal force, the capacitance value increased with the weight, demonstrating that the device effectively responds to shear‐induced contact area expansion even under a steady normal preload (Figure 4d). The sensor was placed on the surface of the index finger, and a disc was grasped with the thumb. After applying a fixed normal gripping force during Stage I, a lateral force was applied without inducing macroscopic slip. The capacitance value initially increased rapidly and then stabilized in Stage II. Immediately afterward, the fingers were clenched to maintain only the normal force in Stage III. This process was repeated three times, indicating that on a smooth surface, the device could achieve stable grasping through the incipient slip signal as shown in Figure 4e. The same method was applied to an apple gripping experiment. Due to the reduced contact area inherent in curved surfaces, slip occurred more rapidly when shear force was applied. The experiment successfully captured the capacitance peak immediately at the slip in Stage II. By quickly tightening the grip to increase the normal force in Stage III, the contact was re‐stabilized, and this process was repeated three times as depicted in Figure 4f. To provide physical insight into the sensing mechanism, Figure S23 presents a direct comparison among the experimentally measured capacitance response, finite element analysis (FEA) simulations, and the corresponding cross‐sectional SEM observations. The results demonstrate a clear correspondence between the computational models and the physical device. Specifically, the simulation captures the dynamic expansion of the effective contact area under coupled normal and shear forces, where the progressive structural deformation directly correlates with the step‐like increments observed in the electrical signal. Further details regarding the structural evolution across the specific loading stages are documented in the Supporting Information.

Additionally, the ability of the device to identify surface macro‐textures was validated by evaluating its response to varying levels of surface roughness across 200‐, 400‐, 600‐, and 800‐grit sandpapers. To ensure accuracy, the Ra values of 8.1 and 6.4 µm were utilized as quantified benchmarks for the 600‐ and 800‐grit sandpapers respectively [57], and correspondingly, the Ra values for the coarse 200‐ and 400‐grit textures were approximately 16 and 11 µm. During the experiments, a constant normal force was applied, followed by a gradual increase in shear force until macro‐slip occurred, with simultaneous recording of capacitance changes. The results indicated that as the surface roughness decreased, the maximum capacitance value at the onset of slip diminished, and the capacitance variation during incipient slip became notably more gradual.

This phenomenon is fundamentally driven by frictional contact mechanics, where the higher friction coefficient of rough surfaces facilitates greater shear‐induced stretching deformation within the microstructures prior to macroscopic slip. During this initial phase, significant elastic strain energy accumulates within the device. Once the applied force exceeds the static friction limit, the abrupt release of this energy drives the micro‐protrusions to forcefully collide with the microgroove sidewalls at an elevated instantaneous velocity. Consequently, the rougher surfaces generate more pronounced fluctuations in the effective contact area at the instant of slip, which manifests as higher capacitive pulse amplitudes, as illustrated in Figure 4g. Heat map analysis further confirms that surfaces with lower roughness reach macroscopic slip more rapidly and exhibit accelerated response characteristics.

To evaluate the device's performance in practical scenarios, it was integrated onto the index finger of a robotic hand for grasping experiments. Figure 4h illustrates the results of the robotic hand using its index finger and thumb to grasp a rigid ball (blue). During stable grasping, the capacitance remained constant because the rigid ball underwent negligible deformation, maintaining a stable contact interface between the sensor electrodes and the ionogel layer. However, at the onset of slip, the capacitance signal exhibited distinct spikes, stemming from transient fluctuations in the contact area during sliding. The spectrogram below the signal reveals concentrated energy clusters, reflecting the device's sensitive response to dynamic contact transitions. For the deformable soft ball (Figure 4i), the capacitance baseline was notably higher than that of the rigid ball, primarily due to the higher normal force generated by its larger mass. Meanwhile, the continuous compliance of the soft ball led to a progressive increase in contact area, causing a gradual rise in capacitance and a broader slip‐induced peak. The time‐frequency spectrum of the soft ball displays a relatively uniform energy distribution, contrasting with the high‐frequency fluctuations observed for the rigid ball, thereby confirming the device's capability for tactile hardness discrimination.

2.5. Transient Slip Signal Applied to Robot Liquid Identification

To demonstrate the feasibility of the proposed flexible tactile sensing system for robotic liquid identification (Figure 5a), five typical liquids were selected. These liquids include honey (1), glycerol (2), diethylene glycol (3), ethylene glycol (4), and water (5), which cover a wide range of densities and viscosities detailed in Table S4. To minimize geometric interference and ensure the acquired signals were exclusively driven by internal fluid dynamics, these liquids were contained in standardized transparent polyethylene terephthalate  plastic bottles. The transparency was utilized strictly for visual ground‐truth labeling during data collection; the Random Forest classification algorithm relies entirely on the robust non‐visual electrical signals of the sensor. However, it should be explicitly noted that the current proof‐of‐concept fixes these structural variables to isolate inherent fluidic features. In broader practical applications, variations in container stiffness, wall thickness, and material properties will inevitably alter the mechanical transmission path from the fluid to the sensor interface, thereby affecting the signal‐to‐noise ratio. Achieving universal recognition across arbitrary sealed and opaque containers remains a critical direction for future research.

FIGURE 5.

Schematic of robotic liquid identification application. (a) Process of the multi‐liquid identification model based on the Random Forest algorithm. (b) Shaking process of the bottle loaded with water/honey. (c) Relative capacitance curves of five different liquids during shaking. (d) Signal amplification view of the response characteristic interval. (e) includes a comparative bar chart detailing the mean recognition accuracies and standard deviations across the three input conditions, alongside the percentage‐normalized confusion matrices for both the equal volume setup and the joint‐variable (equal mass and equal volume) fusion setup.

Under a standardized motion program, the robot agitated the bottles at a constant angular velocity. All specific operational parameters, including shaking angle, angular velocity, gripping force, and contact geometry, are summarized in Table S5. During this dynamic interaction, the sensor continuously collected capacitance signals. Furthermore, a constant fill fraction of 50% was employed across all main liquid identification experiments. Systematic evaluations under varying fill fractions, presented in Figure S24 and Table S6, indicate that moderate fill levels provide optimal signal strength and preserve critical high‐frequency transient features, whereas extreme filling conditions degrade sensing performance. At low fill levels, insufficient fluid mass fails to generate adequate interfacial shear deformation. Conversely, at high fill levels, restricted headspace leads to quasi‐rigid‐body motion that severely suppresses transient slip signals. Based on these quantitative analyses, we define the effective operating envelope for the system to be between a 30% and 80% fill fraction. The sensor is mounted at the robot's fingertip. During the shaking process, it perceives the liquid's inertia, viscous response, and geometric contact characteristic changes through a series of integrated mechanical interactions. The typical shaking action can be divided into three stages, with the processes of shaking water and honey shown in Figure 5b. First, in the vertical gripping stage (Stage I), the normal pressure generated by the static mass of the liquid under the shaking acceleration dominates the sensor output. It is observed that as the density increases, the normal force becomes stronger, leading to an increase in the steady‐state capacitance value, expressed as

$${\mathrm{F}}_{\mathrm{normal}}=\rho ·\mathrm{V}·{\mathrm{a}}_{\mathrm{shake}}$$ (3)

Here, ρ represents the liquid density, V is the liquid volume, and a_shake is the shaking acceleration. Subsequently, the bottle tilts toward the sensor side in Stage II, causing an instantaneous increase in the contact area due to the pressure from the bottle wall on the sensor, which triggers a sharp increase in capacitance. However, this abrupt signal primarily arises from changes in the geometric contact structure (Figure 5c). The key to liquid identification occurs during the rotational operation in the direction of the thumb. In this stage, the bottle undergoes angular acceleration along the transverse direction, marked as Stage III. Different liquids exhibit distinct behaviors due to their specific inertia and shear viscosity. In the initial rotational phase, the liquid lags the motion of the bottle, generating noticeable shear stress at the interface, expressed as

$$\tau =\mu ·\frac{\omega ·\mathrm{R}}{\mathrm{d}}$$ (4)

The shear stress, denoted as τ, is related to the liquid viscosity µ, the radius of the bottle R, the boundary layer thickness d, and the angular velocity of the shaking ω. Under the application of shear stress, the sensor capacitance increases rapidly, and then, as the bottle continues to rotate, the effective contact area between the sensor and the bottle gradually decreases, causing the capacitance signal to drop. As the bottle returns to its upright position, the normal pressure recovers, and due to the delayed backflow of high‐viscosity liquids, the capacitance recovery process is slower compared to low‐viscosity liquids. This results in a characteristic capacitance curve along the time axis showing an “increase‐decrease‐increase” pattern (Figure 5d). This coupled dynamic response can be described by the following model.

$$\mathrm{\Delta }\mathrm{C}\left(\mathrm{t}\right)={\mathrm{k}}_1\times \tau \left(\mathrm{t}\right)\times {\mathrm{A}}_{\mathrm{c}}\left(\mathrm{t}\right)+{\mathrm{k}}_2\times {\mathrm{F}}_{\mathrm{normal}}\left(\mathrm{t}\right)$$ (5)

where A_c(t) represents the contact area, and k₁ and k₂ are calibration constants. Due to the combined differences in density and viscosity, each liquid exhibits a unique capacitance time‐series signal during the shaking process.

It is worth noting that the presented equations are intended to serve as a qualitative analytical framework rather than a strict predictive mathematical model. Due to the inherent complexity of how internal fluid dynamics are transmitted through the container wall to the sensor interface, key parameters in the theoretical model, such as the effective thickness of the boundary layer and the dynamic contact area within the sensor, are exceedingly difficult to quantify directly and accurately. Consequently, these equations are primarily utilized to reveal the qualitative proportional trends and physical correlations among liquid viscosity, interfacial stress, and the output signals of the sensor. This approach provides the necessary theoretical support and conceptual foundation for the dynamic response patterns observed in the experiments.

To accurately capture the viscous damping, inertial impact, and dynamic rheological signatures of the fluid, 15 explicitly defined physical features were extracted and classified into four distinct physical categories. The amplitude and energy characteristics include the mean value, standard deviation, maximum peak, minimum static baseline, and root mean square (RMS) energy. The time‐domain morphology and Hjorth parameters comprise the wave duration, curve length, maximum gradient, Hjorth mobility, and Hjorth complexity. Furthermore, the statistical distribution features incorporate kurtosis, skewness, and crest factor. Lastly, the frequency‐domain and dynamic features are uniquely represented by the spectral centroid and decay ratio. This highly interpretable physical feature space closely correlates with the underlying fluid mechanics, ensuring full methodological reproducibility for the subsequent Random Forest classifier. To provide robust validation and ensure statistical significance, we established a comprehensive tactile dataset comprising between 363 and 438 independent transient waveforms per category, totaling 1950 samples. The entire dataset was subjected to a rigorous stratified 5‐fold cross‐validation strategy. Experimental results demonstrate a mean blind‐test recognition accuracy of 91.62% ± 1.78% under equal mass conditions and 95.23% ± 1.56% under equal volume setups, utilizing the independent 15‐dimensional physical feature sets. Notably, when equal mass and equal volume were introduced as joint variables, the system utilized a dual‐channel soft‐voting fusion mechanism on the 30‐dimensional fused feature vectors to capture the intrinsic physical nuances of fluid interactions, which solidified the comprehensive recognition accuracy at 99.04% ± 0.47%. Figure 5e presents the performance comparison bar chart and the detailed confusion matrices for the dual‐channel joint‐variable conditions, while the percentage‐normalized confusion matrices for the individual equal mass and equal volume testing scenarios are documented in Figures S25 and S26.

Furthermore, to strictly test the model's generalizability on entirely new datasets, a Leave‐One‐Bottle‐Out (LOBO) validation strategy was implemented under the equal volume configuration. The model was trained on interaction data from two bottles and blind‐tested on a completely unseen third bottle. Under this highly demanding condition, the model achieved a remarkable overall accuracy of 91.82% ± 2.56%, effectively demonstrating its capability to extract invariant physical features rather than memorizing localized stochastic fluctuations. The detailed performance metrics for this LOBO experiment are provided in Table S7, and the corresponding confusion matrix is presented in Figure S27.

To further evaluate the ability of the system to distinguish fluids with varying concentration levels, we conducted a proof‐of‐concept test using five glycerol‐water mixtures (10, 30, 50, 70, and 90 wt.%, Table S8). Although the high temporal similarity of their waveforms under stochastic shaking renders simple macroscopic flow‐speed differentiation ineffective, the model successfully leveraged multi‐dimensional physical features to achieve a robust classification accuracy of 94.19% ± 4.35% (Figures S28 and S29). This experiment effectively delineates the discrimination boundaries of the current system. When viscosity differences become so minute that the corresponding dynamic signal variations are overshadowed by the sensor's inherent background noise, the classification accuracy naturally declines. Nevertheless, there remains significant potential to further elevate this resolution boundary and expand the system's adaptability through the future implementation of advanced signal preprocessing techniques and hardware optimization. Furthermore, to strictly validate the system's reliability under drifting operational parameters, we examined the impact of a severe 30 mm grip offset from the central position. The experiments revealed a crucial physical mechanism: the high‐frequency dynamic signal variations caused by internal fluid sloshing are physically orthogonal to the baseline signal alterations induced by spatial offsets. By effectively decoupling the intrinsic fluidic signatures from these static gripping variations via our comprehensive 15‐dimensional physical feature space, the model achieved a highly robust mean recognition accuracy of 84.03% ± 2.38%, even under a strict bidirectional cross‐domain blind test (i.e., the model was trained exclusively on data from the original position and tested on the offset position, and vice versa). This confirms the system's strong generalization capability and operational robustness in real‐world scenarios where precise robotic alignment cannot always be guaranteed. The detailed cross‐domain evaluation protocols and the corresponding confusion matrices are provided in (Figure S30).

2.6. Sensing Strategy and Outlook

Having demonstrated the successful application of the sensor in robotic liquid identification tasks, it is essential to elucidate the underlying sensing mechanisms and the inherent advantages that enable such performance. A distinctive characteristic of this system is that the sensor prioritizes integrated signal acquisition over the pursuit of orthogonal decoupling between normal and shear forces at the hardware level. Instead, it exploits the physical coupling of these forces to capture the intricate dynamics of fluid motion. This strategy is inspired by biological tactile perception, where the nervous system naturally processes integrated stimuli and performs holistic pattern recognition rather than explicitly decoupling mechanical stress components at the physical interface.

In the liquid classification tasks investigated here, the sensor experiences complex combined forces during the reciprocal shaking of the container by the robotic hand. The fluid movement induced by periodic acceleration and deceleration generates dynamic normal impact forces against the container walls, while the inertial resistance caused by the mass shifting of the liquid generates tangential shear forces. The specific patterns in which these forces fluctuate and overlap are strictly governed by the rheological properties of the liquid, such as viscosity and density. Compared with using multiple electrode arrays to decouple these forces, the single‐channel architecture in this study preserves the continuously changing composite stress. This approach transforms the interaction into a coupled waveform containing rich, multi‐dimensional physical information, thereby enabling highly effective rheological sensing. This complexity forms the fundamental rationale for employing the physics‐informed Random Forest algorithm. When processing these coupled mechanical signals, the machine learning pipeline does not need to mathematically disentangle independent physical scalars. Instead, it relies on robust physical feature engineering to identify implicit rheological invariants from the integrated signal response, thereby achieving reliable fluid classification.

In practical applications, flexible sensors inevitably face cross‐interference from multiple physical fields such as temperature and humidity. To address this, recent studies have proposed effective multimodal decoupling strategies, including the integration of multilayer biomimetic architectures [58], corrugated symmetric structures [59], and dual‐channel reading mechanisms [60]. While these strategies excel at separating absolute physical quantities, their reliance on heterogeneous materials and multi‐channel wiring increases manufacturing complexity and may compromise device durability under continuous dynamic impacts. In contrast, recent research on biomimetic soft grippers suggests that the core of precisely capturing incipient slip lies in clever interfacial mechanical design to utilize local stress concentration effects, rather than merely stacking decoupling hardware [61].

Consistent with this practical approach, the architecture proposed in this work features mismatched dimensions between protrusions and grooves. This specific design amplifies high‐frequency slip signals through a simple geometric interlocking mechanism. Based on this rationale, this work forgoes complex hardware‐level decoupling and adopts a homogeneous sensing architecture to ensure long‐term structural stability. Environmental fluctuations in temperature and humidity manifest as ultra‐low‐frequency continuous drift, whereas liquid mass applies a distinct steady‐state load and slip events generate high‐frequency transients. By extracting multidimensional physical features from the raw signals, the system inherently isolates environmental noise while retaining essential liquid density and dynamic slip information. The subsequent Random Forest algorithm prioritizes these distinct temporal features for classification, which ensures the reliability of complex recognition tasks while significantly simplifying the physical complexity of the sensor.

This sensing strategy provides a practical solution where traditional fluid characterization methods fall short. Conventional instruments offer high precision but require invasive sampling, posing contamination risks and causing irreversible sample consumption, which is particularly problematic for precious or hazardous materials. Visual systems depend on container transparency, and most tactile sensors only detect macroscopic weight or steady pressure. Thus, capturing subtle fluid features through opaque walls remains challenging. In contrast, our tactile sensor uses its microstructures to convert internal rheological changes into externally measurable feedback. Integrated onto a robotic gripper, it captures transient force signals during shaking without opening the container or making direct fluid contact. This non‐invasive and non‐visual capability is crucial for automatically classifying photosensitive reagents and nondestructively inspecting hazardous sealed liquids in opaque packaging.

The current tests serve as a proof of concept for the detection method. For future practical applications, it is necessary to improve manufacturing consistency, upgrade hardware circuits, and optimize filtering algorithms. Furthermore, training robust machine learning models with more comprehensive datasets under various perturbations is expected to further enhance recognition accuracy for diverse liquids in complex practical scenarios.

Beyond liquid identification, this incipient slip capture mechanism offers potential for the stable manipulation of objects with shifting internal loads. While sensors often react only after macroscopic sliding occurs, our device detects minute relative displacements at the contact interface caused by fluid inertia. Such predictive feedback enables robotic hands to dynamically adjust their gripping force in response to internal fluid movement during the shaking process. These capabilities provide a foundation for future robotic systems to handle fluidic containers with refined perception.

3. Conclusion

Inspired by the synergistic mechanoreceptors in human skin, we developed a biomimetic ionic tactile sensor that achieves non‐invasive and non‐visual identification of liquids in sealed containers with an accuracy of 99.04% ± 0.47%. This performance is attained by capturing transient interfacial slip signals during shaking and decoding their extracted physical features via a Random Forest ensemble classifier, yielding a robust solution for the management of opaque or hazardous substances. This application‐level performance is fundamentally rooted in the dual‐structural design incorporating micro‐protrusions and grooves, which enables a single sensing channel to respond to two distinct stimuli. Specifically, the protrusion‐groove dimensional mismatch induces transient geometric collapse to achieve a rapid shear response of 46 ms, effectively mimicking the capability of FA‐I mechanoreceptors in capturing dynamic slip. Simultaneously, the same interlocking structure facilitates stress concentration to deliver an ultrahigh peak pressure sensitivity of up to 1873.83 kPa^{− 1}, paralleling the static encoding function of SA‐I receptors for pressure stimuli. Looking forward, the incipient slip detection mechanism validated in this study holds significant potential beyond liquid identification, providing a novel technological pathway for fields requiring refined tactile feedback, such as early slip warning in soft robotics, real‐time stable interaction in teleoperation, and the development of next‐generation multimodal electronic skins. These capabilities underscore broad application prospects for the future of intelligent robotic perception.

4. Methods

4.1. Materials

Polyvinyl alcohol (PVA), phosphoric acid (H₃PO₄), and distilled water were obtained from the laboratory. Glass slides (50 mm × 50 mm) were purchased from Guo‐Luo Glass Shop.

4.2. Preparation of PVA Solution

Initially, PVA (5 g) was added to deionized water (50 mL) and stirred at room temperature for 15 min. The mixture was then placed on a heating plate at 90°C and stirred for 1 h until the solution became transparent. Afterward, it was allowed to cool while stirring at room temperature. Subsequently, H₃PO₄ (4.2 mL) was added to the solution, which was further stirred for 40 min. The solution was then left to stand for 30 min until any bubbles disappeared.

4.3. Preparation of Glass Surface Template

A template pattern was designed using computer‐aided desigm (software, with circular radii of 150 and 250 µm, and corresponding spacings of 150 and 250 µm. A 38 µm thick photosensitive dry film was cut into 50 mm × 50 mm pieces and placed on the surface of a glass slide. The film and the glass slide were then exposed to strong light for 30 s. After exposure, the glass slide was immersed in a Na₂CO₃ (1 wt.%) solution to remove the exposed surface, leaving behind the patterned glass template with the desired circular features. The prepared templates were then set aside for further use.

4.4. Electrode Fabrication

PDMS (Sylgard 184, Dow Corning) and its curing agent were mixed in a 10:1 weight ratio. The mixed PDMS precursor was placed in a vacuum oven for 30 min to remove any trapped air bubbles. The PDMS mixture was then poured onto the template with a radius of 150 µm and spacing of 150 µm. After curing, the PDMS film was peeled off, and a 20 nm gold (Au) layer was deposited onto the surface using thermal evaporation. The PVA‐H₃PO₄ solution was then poured onto the gold‐coated PDMS surface, and vacuum was applied multiple times to ensure the solution fully entered the PDMS grooves. After the bubbles disappeared, the assembly was left to stand for 12 h, forming the PVA substrate with protruding gold electrodes as the top electrode. The bottom planar electrode was fabricated using an unpatterned PDMS template, onto which a gold layer was deposited and subsequently transferred to a flat PVA substrate.

4.5. Fabrication of the Intermediate Ion Layer

A template with a radius of 250 µm and a spacing of 250 µm was chosen. The PVA‐H₃PO₄ solution was poured onto the surface and left to stand for 12 h, resulting in the formation of the PVA‐H₃PO₄ ion layer with grooves.

4.6. Optimization of Performance

During performance optimization, ion groove layers with radii of 150 and 250 µm were fabricated using the method described above. The depth of the grooves was controlled by the number of photosensitive dry film layers used. A single layer of photosensitive dry film has a thickness of 38 µm. However, due to the inherent volumetric shrinkage of the PVA‐H₃PO₄ solution during the solvent evaporation and curing process, the effective depth of the replicated microstructures is proportionally reduced. Consequently, utilizing two, four, or six layers of the dry film template yielded actual measured groove depths of approximately 70, 140, and 210 µm, respectively. The PVA‐H₃PO₄ solution was then poured onto these templates, resulting in ion groove layers with different radii and depths.

4.7. Preparation of PDMS Spacer Layer

PDMS (Sylgard 184, Dow Corning) and curing agent were mixed in a 10:1 weight ratio. The mixture was placed in a vacuum oven for 30 min to remove any bubbles. The PDMS precursor was poured into a pre‐designed polytetrafluoroethylene  mold and cured at 80°C for 30 min.

4.8. Device Fabrication

The top and bottom electrodes, along with the ion layers, were cut into 1 cm × 1 cm squares. PDMS spacers were placed around the bottom electrode, and adhesive tape was used to assemble the device by connecting the top and bottom parts.

4.9. Characterization and Measurement

The prepared materials and sensors were characterized using a comprehensive suite of analytical techniques. Scanning electron microscopy (SEM) was employed to examine the surface microstructures and interfacial cross‐sections for high‐resolution structural verification. An inductance, capacitance, and resistance digital bridge (HG2810B, High Grace Electronics Co., Ltd.) was utilized to systematically characterize the capacitance and resistance signals of the device, providing insights into its dynamic electrical sensing behavior under mechanical stimuli. Finally, a digital single‐lens reflex (DSLR) camera was used to obtain macroscopic optical images of the materials and experimental setups, providing clear visual documentation of the structural appearance and interactive states.

Funding

This work was supported by the National Natural Science Foundation of China under the Basic Science Center Program for “Space Robot Intelligent Manipulation” (Grant No. T2388101) and National Natural Science Foundation of China (NSFC, No. 52372042).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: adma73955‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that supports the findings of this study are available in the supplementary material of this article.