Main

Wearable bioelectronics hold great promise for continuous health monitoring in real-life settings1,2,3,4,5,6. Dynamic environments introduce complex motion artefacts from muscle activity, body movement and physiological interference, severely degrading signal quality and limiting practical deployment7,8,9 (Supplementary Fig. 1). While materials engineering10,11, device design12,13 and algorithmic correction4,14 are mainstream strategies for mitigating motion artefacts, most methods target only a single artefact type or fixed frequency band, failing to address real-world complexities15,16,17,18. This challenge is especially evident in mental-health-related conditions, such as chronic fatigue, a multifactorial condition involving cognitive overload, circadian disruption and physical exertion, for which no single physiological marker is diagnostic19,20,21.

Chronic fatigue affects approximately 13.5% of adults in the USA and contributes to a substantial societal and economic burden, particularly in high-risk professions where fatigue increases the likelihood of human error22,23 (Fig. 1a and Supplementary Fig. 2). These realities underscore the pressing need for continuous, real-time fatigue assessment systems that operate reliably in daily settings characterized by motion and environmental variability. Among various biomarkers, such as brain activity24, biochemical markers25,26 and cardiovascular patterns27,28,29,30, wearable cardiovascular monitoring is particularly attractive for fatigue profiling owing to its portability and data processing efficiency. As cardiovascular states are regulated by the autonomic nervous system (Fig. 1b), electrocardiographic and haemodynamic signals serve as robust indicators of fatigue (Supplementary Note 1), provided motion artefacts are effectively mitigated.

Fig. 1: Overview of the MAP for fatigue profiling.
Fig. 1: Overview of the MAP for fatigue profiling.
Full size image

a, Impact of fatigue including economic loss, health issues and safety accidents, where the area associated with each factor represents its relative magnitude of impact. b, Illustration of the metahydrogel platform that collects electrophysiological and haemodynamic information and performs machine learning powered fatigue evaluation. c, Working mechanism of metahydrogel for collecting targeted signals and attenuating concomitant artefacts. Left: mechanical interference is blocked and dissipated by local resonance. Right: biopotential artefacts are filtered through frequency-selective ionic mobility, as solvation reconstruction dynamics do not respond at EMG frequencies. Schematics in b created in BioRender: brain, Tian, G. https://biorender.com/odm98mm (2026); spinal cord, Tian, G. https://biorender.com/ckttg2c (2026); heart, Tian, G. https://biorender.com/2540e4r (2026); vessel, Tian, G. https://biorender.com/hl83bbg (2026).

To address this challenge, we developed a meta-topological hydrogel that seamlessly integrates mechanical wave-modulating metastructures with topological-tuned ionic transport, facilitating programmable and targeted suppression of dynamically evolving motion artefacts. The coined metahydrogel artefact-mitigating platform (MAP) functions as an integrated noise cancellation interface for high-fidelity haemodynamic and electrophysiological signals acquisition for fatigue profiling and broader bioelectronics applications (Fig. 1c). Periodically assembled nanoparticles patterns within the hydrogel scatter mechanical waves to mitigate mechanical artefacts (Fig. 1c, left), while frequencies-selective ion migration, modulated through solvation structure design, mitigates biopotential interference (Fig. 1c, right). Structural and electrolyte optimization aligns the filtering behaviour with physiological noise spectra, and an autoencoder-based denoising algorithm suppresses transient, unstructured noise with preserving critical physiological features. Under dynamic conditions, this synergistic platform achieves an ECG signal-to-noise ratio (SNR) of 37.36 dB, while reducing systolic/diastolic blood pressure (SBP/DBP) measurement errors from 3.45 ± 10.99/2.02 ± 8.99 mm Hg to 0.37 ± 4.53/0.27 ± 3.71 mm Hg. By integrating extracted signal features with cardiovascular patterns, we developed a convolutional neural network and regression model to profile physiological patterns and classify fatigue levels, achieving an accuracy of 92.04% in daily fatigue monitoring. Beyond fatigue tracking, MAP generalizes to diverse biosignals, including heart sounds, respiratory sounds, voice, electroencephalography and electrooculogram. Hence, MAP establishes a scalable framework for high-fidelity, artefact-free bioelectronics suitable for deployment in real-world settings.

Mechanical damping effect of metastructured hydrogel

Unlike conventional damping materials that lack frequency selectivity and attenuate mechanical signals indiscriminately, a periodic precipitation pattern (Liesegang pattern) was introduced into a polyacrylamide (PAAm) hydrogel to construct a metastructured interface material capable of selectively decoupling mechanical waves (Supplementary Video 1). This self-assembled, ordered pattern was prepared through coupled reactant transport and precipitation kinetics (Fig. 2a and Supplementary Fig. 3). Cross-sectional imaging reveals a periodic CuCl2/K2CrO4 precipitation pattern (Fig. 2b), composed of CuCrO4 nanoparticles separated by depleted regions (Supplementary Figs. 4 and 5). The prenucleation mechanism was confirmed by experimental structural analysis and growth simulation (Supplementary Figs. 6 and 7 and Supplementary Note 2). Based on this mechanism, a stamping method was developed for large-scale fabrication of independent, periodic precipitation structures, highlighting its scalability (Fig. 2c and Supplementary Fig. 8).

Fig. 2: Mitigation of mechanical artefact by the metahydrogel platform.
Fig. 2: Mitigation of mechanical artefact by the metahydrogel platform.
Full size image

a, Schematic nucleation and periodic pattern growth model based on supersaturation theory. As the ion concentration (cab) exceeds the critical supersaturation (Ksp), nucleation and pattern growth are initiated to form the first precipitation region. Back-diffusion of electrolyte within the matrix creates a depleted zone, followed by iterative reactions and ionic diffusion. b,c, Photographs of cross-sectional morphology of CuCl2/K2CrO4 periodic pattern (b) and the large-scale fabrication via stamp method (c). Scale bars, 2 mm (b) and 3 cm (c). Experiments were independently repeated at least three times (n = 3) with consistent results. d, Simulated mechanical wave propagation in hydrogel matrix and metahydrogel. Right curves present the amplitude as marked by the dashed white line. e, Ashby plot of loss factor (η) and modulus (E) of metahydrogel and other damping materials. The dashed lines represent the materials with different ηE products. f, Damping frequency versus structural factor of metahydrogel. The inset shows the definition of structural factor P = xn+1/xn, where xn denotes the distance of the nth band measured from the reaction centre. g, Evaluation of resistance to three-dimensional force interference. The inset schematically shows the test set-up, and the projection on the bottom plane indicates the signal intensity detected by the sensor. h, Four-quadrant plot and CRs for evaluating BP measurement accuracy using a commercial sphygmomanometer and a piezoelectric sensor with or without the metahydrogel. Diagonal black lines indicate the ±5 mm Hg error bars, and the grey area represents the ±3 mm Hg exclusion zone.

Source data

Mechanical wave propagation tests confirmed that the periodic precipitation bands interrupted wave propagation (Extended Data Fig. 1). Simulations of the wave propagation and flow-field distribution further validated that the bulk of mechanical energy was absorbed by these nanoparticle structures31 (Fig. 2d and Supplementary Figs. 9 and 10). When arranged periodically, these resonators form a continuous frequency band, creating a frequency-selective phononic bandgap that blocks mechanical wave within defined ranges32. Denoising performance was assessed by comparing the fidelity of output signals under noise interference conditions, where the metahydrogel improved signal fidelity from 0.14 to 0.72 (Supplementary Fig. 11). Dynamic rheological measurements of metahydrogel exhibit a high loss factor of 4.18 at 0.1 Hz, indicating efficient mechanical energy dissipation (Supplementary Fig. 12). To benchmark performance, we compared the metahydrogel with conventional damping materials using an Ashby plot (Fig. 2e and Supplementary Table 1). To quantify the allowable dissipated energy, we introduced the loss-to-modulus product (ηE). With a loss factor of 4.18 and a modulus of 0.8 MPa (ηE = 3.34 MPa), the metahydorgel outperforms most conventional materials and hydrogels. Unlike traditional damping hydrogels constrained by an inherent trade-off between compliance and energy dissipation, the local resonance mechanism decouples softness from damping, surpassing the Ashby limit for soft matter33. Moreover, its modulus closely matches that of biological tissue, enabling conformal interface contact for accurate physiological signal monitoring34.

Mechanical artefact-free haemodynamic monitoring

Compared with conventional hydrogel damping materials that dissipate energy through non-selective structural deformation or bond reconfiguration, the metahydrogel forms a phononic bandgap via its periodic precipitation architecture, enabling frequency-selective damping. To validate tunability, metahydrogels with varying structural parameters were synthesized by modulating ion diffusion and the resulting precipitation patterns (Supplementary Fig. 13). Vibration transmission measurements and transfer-function analyses confirm structure-dependent characteristic damping frequencies, (Fig. 2f and Supplementary Fig. 14), enabling the metahydrogel to be tailored for specific application scenarios. We further systematically explored the effects of precursor thickness, rheology and gelation rate on pattern formation and the resulting damping performance (Extended Data Figs. 2 and 3). The modulus, size and mass of the precipitated particles, as well as the overall dimensions of the patterned region and the number of precipitation bands, were found to exert non-negligible effects on mechanical attenuation performance (Supplementary Fig. 15). Considering both the effective damping range and the characteristic frequency of motion artefacts, the metahydrogel with a thickness of 2 mm and a structure factor (P) of 1.187 was selected for subsequent applications.

Metahydrogels composed of different materials were synthesized to validate the universality of the damping effect, as distinct diffusion coefficients and reaction kinetics generated varied precipitation patterns (Extended Data Fig. 4a and Supplementary Note 3). Simulations confirmed that these material systems exhibit specific bandgaps across different frequency bands (Extended Data Fig. 4b,c), indicating that damping behaviour can be precisely tuned through material and structural design to cover frequencies relevant to human motion (0.5–30 Hz). Under multidirectional noise interference, the metahydrogel preserved vertical signal integrity by attenuating in-plane wave (Fig. 2g). The coefficient of variation in signal intensity is 0.03 for metahydrogel compared with 0.14 for polydimethylsiloxane (PDMS), highlighting superior noise-mitigation performance. Stable signal acquisition was maintained under impulsive disturbances as well (Supplementary Fig. 16). In addition, a comparison of the SNR of transmitted perpendicular vibration amplitude through metahydrogel and other commercial damping materials confirms its selective sensitivity to low-frequency physiological signals (Supplementary Figs. 17 and 18).

To determine suitability for skin-interfacing bioelectronics, the biocompatibility, permeability and adhesivity of metahydrogels were assessed. Due to the encapsulation of nanoparticles within the hydrogel matrix and negligible ionic leakage, there is no observable cellular damage when cocultured with epithelial cells (Supplementary Fig. 19). The porous network enables a water vapour transmission rate of 2.43 kg m−2 day−1, exceeding the average human skin permeability (~0.6 kg m−2 day−1)35, thus facilitating efficient moisture exchange (Supplementary Fig. 20). Furthermore, incorporation of gelatin enables temperature-triggered reversible adhesion, minimizing the risk of tissue damage (Supplementary Fig. 21). Together with the moisture-retaining properties arising from the electrolyte solution within the hydrogel, it can maintain internal water balance under prolonged use and extreme environmental conditions, thereby ensuring stable mechanical, electrical and sensing performance (Supplementary Figs. 22–24). Mechanical durability tests showed no interfacial fatigue or structural failure under cyclic stretching or high strain (Supplementary Fig. 25), and notched tensile tests confirmed efficient energy dissipation by the metastructure (Supplementary Video 2). We further compared the damping and mechanical performance of the metahydrogel against other commercial polymeric dampers (Supplementary Fig. 26), where it demonstrated superiority across all evaluated metrics. Leveraging this bioelectronic platform, a commercial piezoelectric film sensor was integrated to capture artefact-free haemodynamic data.

Because the arterial pressure signal is weak (<10 kPa) and highly susceptible to mechanical artefacts36,37, we assessed the sensor sensitivity to confirm its suitability for blood pressure (BP) monitoring. The integrated sensor achieves a sensitivity of 63.3 mV kPa−1 (Supplementary Fig. 27), and wrist-mounted measurements further show effective artefact suppression compared with direct sensor recording (Supplementary Figs. 28 and 29 and Supplementary Note 4). Bland–Altman analysis reveals that the metahydrogel reduces measurement errors from 2.53 ± 13.83/2.14 ± 10.05 mm Hg to 0.24 ± 4.35/0.16 ± 4.64 mm Hg for SBP/DBP, respectively (Supplementary Fig. 30). Four-quadrant plot analysis further confirmed the improved consistency of the metahydrogel (86.7% for SBP and 90.3% for DBP) relative to direct measurements (55.2% for SBP and 52.2% for DBP; Fig. 2h). These results establish the metahydrogel as a robust foundation for accurate BP monitoring and subsequent physiological assessments.

Biopotential artefact mitigation for ECG monitoring

Biopotential artefacts stem from the electrochemical activities in nervous and muscular cells, with electromyography (EMG) being the dominant interference source in ECG monitoring38. Because EMG amplitudes are comparable to or even exceed ECG signals, spectral overlap between ECG (0.1–100 Hz) and EMG (20–500 Hz) can severely distort recordings. Frequency energy analysis (Extended Data Fig. 5) shows that ECG energy is concentrated primarily within 0.1–30 Hz (~97%), while more than 90% of the EMG energy lies above 30 Hz, supporting the feasibility of selectively attenuating EMG without compromising ECG signals (Fig. 3a). This spectral separation enables frequency-selective biopotential filtering via ionic transport engineering. Based on the vehicular transport mechanism of ions in the electrolyte solution (Supplementary Fig. 31), the ion migration is strongly associated with the bonding interactions between ions and solvation molecules, as well as the overall size of the solvation structure39. Optimizing the solvation structure of ionic hydrogel electrodes enables frequency-selective bioelectric sensing by regulating ion transport.

Fig. 3: Frequency-selective ionic mobility by solvation structure design for mitigating for biopotential artefacts.
Fig. 3: Frequency-selective ionic mobility by solvation structure design for mitigating for biopotential artefacts.
Full size image

a, Frequency-dependent cumulative energy spectra of ECG and EMG signals. The insets show their typical waveforms and peaks. b, Molecular dynamic simulation of solvation configuration and its MSD. c, Electrical loss spectrum versus frequency of hydrogel varying with glycerol contents. The inset presents the mobility response of ions under electric field. A loss factor (LF) greater than 1 indicates that ions cannot respond rapidly enough to the applied electric field. d, Evaluation of interfacial stability between metahydrogel and commercial Ag/AgCl electrodes. The inset presents the equivalent circuit of the tissue–electrode interface. e,f, Comparison of single-period ECG signals with EMG noise collected by commercial Ag/AgCl electrode (e) and the metahydrogel (f). The bottom parts show the wavelet transform spectra, demonstrating that high-frequency noise is mitigated by the metahydrogel. Tissue schematic in d created in BioRender; Tian, G. https://biorender.com/rj7b9v3 (2026).

Source data

Given that CuCl2 and K2CrO4 reactants were used to prepare the metastructure precipitation, the transport of K+ and Cl− predominantly determines the conductivity of hydrogel system. Hydroxyl-containing (–OH) molecules were introduced to modify the ion migration pathways, the energy landscape and the hydrogen-bond network within the electrolyte. Bonding energy analysis shows that increasing –OH groups strengthens the binding energy (Supplementary Fig. 32), restricting solvation-shell reorientation and reducing ionic mobility. A biocompatible glycerol–H2O mixed solvent was therefore selected to regulate ion transport behaviour. Incorporation of glycerol molecules reshapes the solvation-shell configuration and reduces the effective dielectric constant of the medium, strengthening ion–solvent interactions and limiting energetically assessable ion migration pathways. Molecular dynamic simulations confirmed reduced mean squared displacement (MSD) and diffusion coefficient with glycerol incorporation (Fig. 3b and Supplementary Fig. 33), consistent with experimental ionic diffusion measurements (Supplementary Fig. 34 and Supplementary Note 5). Control experiments using high-viscosity polyethylene oxide electrolytes showed uniform attenuation without frequency selectivity (Supplementary Fig. 35), demonstrating that viscosity alone does not account for the selective damping. Therefore, only solvents that reshape the solvation shell and modulate ion–solvent interactions can effectively tune ionic relaxation dynamics.

Dielectric-loss spectra revealed that increasing glycerol content shifted the ionic response window towards lower frequencies (Fig. 3c). At high frequencies (>100 Hz), delayed ionic relaxation caused conductive loss, whereas at low frequencies (<1 Hz), sufficient ion migration leads to increased loss. Electrochemical impedance spectra further demonstrated a shift of the ionic relaxation peak to lower frequencies with increasing glycerol content (Supplementary Fig. 36). Optimizing the water-to-glycerol ratio (50:50 v/v) produced a response window of ~1–100 Hz, matching ECG bandwidth while attenuating high-frequency EMG noise. Before practical application, we evaluated the interfacial impedance stability of the metahydrogel compared with a commercial Ag/AgCl electrode (Fig. 3d). The metahydrogel exhibits superior stability under dynamic conditions due to stable adhesion and the metastructural damping. Frequency sweep tests confirm selective signal transmission, whereas the Ag/AgCl electrode transmits electrical signals indiscriminately (Supplementary Fig. 37). Electrolyte-dependent transfer functions exhibit composition-dependent frequency shifts (Supplementary Fig. 38). When the glycerol content reaches 50%, the characteristic response frequency (~0.4–40 Hz) closely aligns with the ECG bandwidth. Under dynamic upper-limb motion, the Ag/AgCl electrode produced an SNR of 3.33 dB, while the metahydrogel achieved 15.72 dB (Fig. 3e and Supplementary Figs. 39 and 40).

Monitoring of the cardiovascular system in a dynamic environment

Having demonstrated the capability of our MAP in mitigating both mechanical and biopotential artefacts, we then explored its application in practical dynamic scenario. To assess the robustness of MAP, arterial pulse (AP) signals were analysed over a 10-min period (Supplementary Fig. 41). Without MAP, motion artefacts distorted waveforms and produced scattered Poincaré plots unsuitable for heart rate variability (HRV) analysis. By contrast, AP signals collected with MAP remained stable, yielding tightly clustered Poincaré plot and stable SBP and DBP waveforms (Fig. 4a and Supplementary Figs. 42 and 43), reflecting consistent autonomic nervous system regulation. ECG consistency relative to AP signals was assessed using Bland–Altman analysis, with 97.1% of data points within the 95% agreement limits (Fig. 4b and Supplementary Fig. 44), confirming MAP’s reliability for physiological monitoring. Recurrence plots under dynamic motion showed large non-recurrent regions for the Ag/AgCl electrode, whereas MAP produced structured periodic patterns (Fig. 4c). Analysis of recurrence rate and determinism further confirms that MAP provides a clear representation of the autonomic nervous system activity (Supplementary Fig. 45), making it promising for fatigue assessment.

Fig. 4: Demonstration of mechanical and biopotential artefacts mitigating by metahydrogel.
Fig. 4: Demonstration of mechanical and biopotential artefacts mitigating by metahydrogel.
Full size image

a, Distribution of SBP and DBP data collected over 2 min under dynamic conditions with and without the use of metahydrogel. Data were obtained from one independent participant. The boxes represent the interquartile range (25th–75th percentile), while the whiskers indicate the mean ± standard deviation (s.d.). b, Bland–Altman plot showing HR measurement agreement between the metahydrogel ECG electrode and piezoelectric artery sensor. c, Stability evaluation of collected BP signals by the recurrent analysis of collected ECG signals in time domain. d, Schematic of the autoencoder framework for denoising. The autoencoder model denoises the data by mapping measurement-corrupted data points back to the data manifold (black arrows). The bottom part presents typical AP and ECG signals before and after the denoising process. e, Calculated recognition accuracy of artery pulse and ECG signal versus SNR. The bar plots show SNRs of sensing signals with different denoising process. f, Frequency–power spectra of ECG signals collected with and without the use of metahydrogel after processing with the autoencoder-based denoising algorithm. g, Stability evaluation of BP signals collected over 2 min during a kinetic activity. Data were obtained from one independent participant. Box plots show the median (central line), interquartile range (25th–75th percentile, box) and whiskers extending to the minimum and maximum values. h, Accuracy comparison of measured DBP achieved with and without metahydrogel (red star and red dot; the arrow serves as a guide to the eye) and other approaches. The grey region shows that these data were collected under static state, and the background shading indicates the IEEE accuracy categories (grades A, B and C) for wearable BP monitoring.

Source data

Although MAP effectively mitigates motion artefacts, residual physiological and unstructured interferences can still degrade signal quality7. Traditional filtering or smoothing algorithms indiscriminately remove noise, potentially discarding key physiological features. We therefore developed an autoencoder-based denoising algorithm designed to learn the structure of noise-free physiological signals (Supplementary Note 6), preserving critical features while suppressing artefactual signals (Fig. 4d, Extended Data Fig. 6 and Supplementary Fig. 46). The autoencoder alone increases the ECG SNR to 13.87 dB but fails to fully restore waveform features under severe motion-induced distortion (Supplementary Fig. 47). The metahydrogel alone improved the SNR to 15.75 dB, whereas combined processing reached 37.36 dB (Fig. 4e), demonstrating synergistic enhancement. This improved SNR substantially enhances feature peak identification, raising the detection accuracy from 52% to 93% for ECG waveforms.

To showcase the applicability of MAP in real-world dynamic environments, we monitored both ECG and BP signals from a moving participant, while simultaneously recording upper limb motion and acceleration using an attitude sensor (Supplementary Fig. 48). Despite strong motion artefacts, MAP combined with autoencoder processing effectively suppressed high-frequency EMG noise in the ECG spectrum (Fig. 4f). Similarly, BP measurements show improved agreement with a commercial sphygmomanometer, with absolute mean errors reduced from 2.02 to 0.27 and 3.45 to 0.37 mm Hg for DBP and SBP, respectively (Fig. 4g). Following the Institute of Electrical and Electronics Engineers (IEEE) standards for wearable electronics in BP monitoring, we benchmarked the accuracy of MAP against previously reported results40,41. Most existing methods require measurements in a static state (grey region), whereas for kinetic conditions, our MAP combined with the denoising algorithm reduced standard deviation of DBP from 10.99 to 4.53 mm Hg (Fig. 4h), achieving comparable performance to that of a wearable ultrasound patch42 (3.90 mm Hg). Moreover, we benchmarked the performance of our device using a clinical ISO 81060-2 gold-standard protocol43 (Supplementary Note 7 and Extended Data Fig. 7). MAP achieved mean errors of only 1.51/1.88 mm Hg for SBP/DBP with corresponding standard deviations of 2.99/2.86 mm Hg, well within the acceptance boundary. Overall, these results demonstrate the strong potential of MAP as a robust solution for haemodynamic monitoring in motion-intensive environments.

Demonstration of MAP’s robustness in dynamic environments for daily fatigue profiling

To validate the robustness and practicality of MAP in real-world scenarios, a fully integrated, flexible and wearable MAP system was fabricated for continuous, on-body physiological state monitoring during daily activities (Fig. 5a, Supplementary Fig. 49 and Supplementary Video 3). Commercial signal acquisition and wireless transmission modules enable arterial and electrophysiological signals transmission for processing (Supplementary Fig. 50). Frequency-dependent transfer-function analyses confirmed long-term stability of mechanical and electrical artefact mitigation (Supplementary Fig. 51). Moreover, controlled experiments were conducted on five participants, during which AP, BP and ECG signals were recorded under daily conditions. Comparisons of the coefficient of variation between MAP and commercial sensors demonstrated MAP’s effective artefact mitigation (Fig. 5b). Fatigue levels for participants were simultaneously assessed using the fatigue assessment scale (FAS) questionnaire as a quasi-gold standard, with scores ranging from 10 to 50 points (Supplementary Note 8 and Supplementary Table 2). Given that the FAS is a unidimensional index, scores from 10 to 50 were interpreted as representing a progressive increase in fatigue44.

Fig. 5: Artefact-free fatigue profiling in daily life.
Fig. 5: Artefact-free fatigue profiling in daily life.
Full size image

a, Photograph of a participant wearing the MAP for real-time physiological signals monitoring. The mobile phone shows the user interface. Scale bar, 10 cm. b, Robustness evaluation of collected different physiological signals from MAP and a commercial sensor. Data were obtained from independent participants (n = 5). Bars indicate mean ± s.d. c, Frequency histogram of the difference of adjacent RR peaks (Ri+1–Ri) at normal and fatigue states. The narrower distribution indicates the reduced HRV. d, Continuous 3-day multimodal physiological signals monitoring during different daily activities on a participant. For each timepoint, and three segments within this window (over 2 min) were analysed. Error bars represent the mean ± s.d. calculated from these replicates (n = 3). e, SHAP summary plot for fatigue evaluation based on the dataset collected by the MAP. Each axis plots the distribution of SHAP value of the selected feature for predicted instance. f, Ture value versus assessed fatigue levels. Data are presented with a ±2-point buffer to account for potential errors in the FAS questionnaires. g, ROC curves evaluating the classification performance of the MAP system for fatigue state profiling. The curves are colour-coded from light to dark to represent increasing score scales on the fatigue assessment. h, t-SNE-based clustering of fatigue levels among different participants. A total of five participants were assessed, and their fatigue states were categorized into seven levels based on scores from the FAS. Activity icons in d created in BioRender; Tian, G. https://biorender.com/omp1s6v (2026).

Source data

Across participants, physiological signals show consistent fatigue-related trends during prolonged work without sufficient rest. Over the course of the day, short-term high intensity work or exercise increases energy consumption in the body and brain, triggering sympathetic nervous system (SNS) activation, thereby elevating heart rate (HR) and SBP45. To assess autonomic nervous system dynamics, we extracted the low-frequency to high-frequency ratio (LF/HF) from ECG signals, which reflects the balance between SNS and parasympathetic nervous system (PNS) activity (Supplementary Fig. 52). Increased SNS activity suppresses the activation of PNS, which impacts cardiac homeostasis, resulting in a decreased HRV (Fig. 5c and Supplementary Fig. 53). Long-term monitoring accounted for circadian fluctuations, revealing fatigue-induced deviations beyond normal physiological ranges (Supplementary Fig. 54). Analysis of LF/HF ratio, sample entropy (SE) and the standard deviation of normal-to-normal intervals (SDNN) across participants revealed patterns indicating a disruption beyond normal physiological rhythm balance, reflecting pronounced PNS dysregulation (Fig. 5d and Supplementary Fig. 55). These results confirm the robustness of MAP to perform continuous, artefact-resilient physiological monitoring and reliable biomarker tracking.

Proof of application in data-driven fatigue profiling and extended scenarios

To demonstrate the potential of the MAP system for data-driven physiological analysis and preemptive intervention, we developed a prototype fatigue profiling framework as a proof-of-concept application. A deep learning pipeline was developed to extract physiological features and correlate them with questionnaire-based fatigue classifications. Principal component analysis was used for data dimensionality reduction and to identify the most discriminative ECG features (Extended Data Fig. 8 and Supplementary Figs. 56), which were used to train a convolutional neural network for multilevel fatigue classification (Supplementary Fig. 57). Feature space visualization using t-distributed stochastic neighbour embedding (t-SNE) revealed a clear clustering of fatigue states, achieving 92.0% classification accuracy, indicating that MAP provides high-quality data sufficient for capturing meaningful physiological variations (Supplementary Fig. 58).

We further explored the feasibility of fatigue level estimation using combined electrophysiological and haemodynamic features. Shapley additive explanation (SHAP) and permutation importance analyses confirmed the physiological interpretability of the learned features (Fig. 5e and Supplementary Fig. 59). A random forest regression model trained on FAS scores was constructed to assess fatigue levels with high confidence, achieving a coefficient of determination of 0.91 (Fig. 5f and Supplementary Fig. 60). Comparative analysis showed that MAP-enhanced classification achieved an area under the curve (AUC) of 0.93 in the receiver operating characteristic (ROC) analysis (Fig. 5g), substantially outperforming direct monitoring without MAP (AUC 0.64). These results demonstrated that the MAP-enabled signals retain meaningful physiological information suitable for modelling fatigue-related trends. Expanded cohort validation shows strong agreement between predicted and subjective FAS scores (92.04%; Fig. 5h and Supplementary Fig. 61), supporting model robustness across individuals.

To further validate the applicability of the MAP, we extended its use to various physiological monitoring modalities, including heart sounds, breath sounds, voice, electroencephalography and electrooculogram signals (Supplementary Figs. 62–64). During signal acquisition, MAP effectively suppressed diverse motion artefacts such as thoracic vibrations from breathing, laryngeal deformation during vocalization, and noise from body movements. By mitigating these artefacts, the system enables clearer and more accurate signal capture, facilitating more reliable disease diagnosis, speech recognition and brain activity sensing. These proof-of-concept demonstrations highlight the MAP system’s capability to support artefact-free, multimodal physiological monitoring under realistic conditions, positioning the platform as a foundation for future large-scale, data-driven health analytics. Ultimately, the MAP system presents a promising framework for objective fatigue evaluation and real-time physiological monitoring in ambulatory settings and warrants broader validation across diverse populations and use cases.

Conclusions and outlook

We report a meta-topological hydrogel bioelectronic platform designed to enable high-fidelity, multisource, artefact-resilient physiological signal acquisition in motion-intensive environments. By integrating phononic metastructure damping and ionic solvation engineering, this platform allows high-fidelity monitoring with targeted suppression of mechanical and biopotential artefacts, achieving low BP mean error of 0.37 and 0.27 mm Hg for SBP and DBP and high SNR of 37.36 dB for ECG in unstructured environments. To explore the feasibility of physiological pattern recognition, we conducted a proof-of-concept demonstration of fatigue profiling. By further processing these high-fidelity multimodal physiological features using deep learning and regression models, fatigue levels could be evaluated with an accuracy exceeding 92.04%, demonstrating the platform’s potential for early fatigue diagnosis and prevention. While fatigue monitoring serves as a representative application, we further validated the platform’s robustness in other physiological monitoring tasks, with MAP shown to effectively mitigate motion-induced artefacts across multiple biosignal modalities. These results validate the feasibility of objective fatigue assessment and highlight MAP’s utility in supporting data-driven health insights. Overall, this work establishes a versatile hydrogel interface for next-generation wearable bioelectronics, with wide-ranging applicability in real-world settings. This platform lays the groundwork for future advances in continuous, artefact-resilient biosensing and intelligent physiological monitoring, paving the way for personalized digital and connected healthcare.

Methods

All testing involving human participants was approved by the institutional review board of Southwest Jiaotong University (reference no. SWJTU-23012-NSFC(133)).

Acrylamide (AAm) monomer (>98% purity) was purchased from Tokyo Chemical Industry. N,Nʹ-methylene(bis)acrylamide (MBAA), potassium peroxydisulfate (KPS), N,N,Nʹ,Nʹ-tetramethylene-diamine (TEMED), copper chloride (CuCl2) and gelatin (porcine skin, gel strength 300) were purchased from Sigma-Aldrich. Potassium chromate (K2CrO4, >99.0% purity) was purchased from Alfa Aesar. Agarose powder (biotechnology grade with a low electroendosmosis) was purchased from Axil Scientific. PDMS (Sylgard 184) and its crosslinker were purchased from Dow Corning. The sources of commercial damper materials are as follows: Alpha-gel (GT-4, Taica), plasticine (Plasticine-200, Seow Khim Polythelene), D3O (D3O-Elbow, D3O) and starch (Sigma-Aldrich). Unless otherwise specified, all chemicals were used without further purification.

Fabrication

Preparation of metahydrogel

The PAAm hydrogel precursor was prepared by dissolving 1.446 g of AAm monomer and 0.016 g of MBAA in 9.2 ml deionized water with vigorous stirring for 10 min. Air bubbles in the solution were removed by vacuum degassing. Then, 0.02 g of KPS and 0.04 g of K2CrO4 were carefully added without introducing additional oxygen. Afterwards, 20 μl of TEMED was immediately added, and the mixture was stirred in an ice bath for 5 min to ensure uniform mixing. The resulting precursor was poured into an acrylic mould (5 × 5 × 0.2 cm3), and placed in an oven at 60 °C for 2 h to conduct prepolymerization, with the mould covered to prevent solvent evaporation. This was followed by a secondary polymerization at room temperature for 4 h, during which saturated KCl solution was used to maintain a relative humidity above 80%. For preparing the outer CuCl2 source, 1.8 g of agarose was first dissolved in 28.2 ml deionized water by stirring at 90 °C, and the solution was then poured into a cylindrical acrylic mould (diameter 1.0 cm, height 2.5 cm) and placed in a refrigerator to solidify (4 °C, 20 min). The solidified agarose stamp was then immersed in 1 M CuCl2 solution for 40 min to absorb the electrolyte. The agarose stamp was placed at the centre of the PAAm hydrogel, allowing CuCrO4 precipitate to gradually form through ion diffusion and reaction in the following 48 h. The morphology of the resulting metastructure was controlled by adjusting the crosslinking density of the PAAm hydrogel and the placement of the agarose stamp. The preparation of other metastructural pattern systems is detailed in Supplementary Note 3. To modulate the hydrophobic properties of the metahydrogel surface, it was treated using fluorine plasma. First, the metahydrogel was freeze-dried to remove moisture, and then placed in an inductively coupled plasma etching system (DICS-ICP-601, Chuangshi Weina Technology) for treatment. The working gas was CHF3, with a chamber pressure of 2 Pa, a source power of 200 W and a bias power of 100 W.

Preparation of other damping materials

The reference PDMS elastomer was prepared by mixing the base and curing agent at a mass ratio of 20:1. After thorough mechanical stirring, the mixture was degassed and poured into a mould, then cured at 80° C for 4 h. For the starch damper sample, a 55 wt% starch aqueous solution was prepared and poured into a cylindrical mould. The mould was sealed on both sides using polyurethane films to facilitate subsequent damping tests. For the plasticine sample, 5 wt% of water was added and mixed thoroughly, and the material was then shaped into cylindrical forms for testing.

Fabrication of sensing devices and integrated wireless systems

Platinum wire was used as the electrode for the metahydrogel, effectively preventing ionic corrosion and electrochemical reactions at the electrode–hydrogel interface. A piezoelectric polymer film sensor was attached to the hydrogel matrix for pulse signal monitoring. The entire wireless signal acquisition system was fabricated on a flexible printed circuit board, integrating a Bluetooth-enabled microcontroller (ESP32 WROM 32), a piezoelectric signal acquisition and amplification module, a power supply module, an ECG acquisition module (AD8232) and a serial communication module. The connections among different components are presented in Supplementary Fig. 50. The key data-acquisition parameters of the integrated system are summarized as follows. ECG and AP signals are sampled at a rate of 500 Hz using a 12-bit analog-to-digital converter (ADC) embedded in the microcontroller, and the ECG acquisition module operated under a fixed gain of 100. Considering that the metahydrogel damper inherently suppresses mechanical and biopotential artefacts above 1 Hz, a hardware high-pass cut-off was implemented in the acquisition circuit to stabilize the baseline without interfering with the damping behaviours of hydrogel, and a 50-Hz notch filter was used to remove power-line interference. All raw data were recorded without compression to preserve signal integrity. To ensure precise temporal alignment across ECG and artery pulse channels, a time stamp was implemented based on the ESP32 system clock, and all acquisition modules were synchronized through a shared microcontroller-triggered signal.

Characterizations

Materials characterization

The morphology and EDS mapping of metahydrogel were characterized by field-emission scanning electron microscopy (JSM-7001F, JEOL). The hydrogel sample was prepared via freeze-drying to prevent the collapse of the porous structure. The optical photographs of surficial and cross-sectional morphologies of metahydrogel were taken and recorded via an optical microscope (BX53M, Olympus). The Raman mapping of CuCrO4 was collected via a Raman imaging microscope (Alpha 300R, WITec, Oxford Instruments) with a 532-nm laser at 2 mW power. The spatial resolution of the chemical mapping was 10 μm. The modulus of metahydrogel was measured by a universal testing machine (MultiTest 1, Mecmesin) at room temperature, where the tensile/compression rate was set as 50 mm min−1. The electrochemical properties of metahydrogel were measured by an electrochemical workstation (CHI660E, Chenhua Device Company), and the sample was fully swollen in the electrolyte solution before testing. The conductivity and viscosity of different solutions were measured by a conductivity meter (Seven Compact, Mettler Toledo) and an Ubbelohde viscometer, respectively.

Damping performance tests

Dynamic mechanical testing was performed by a rheometer (MCR 302, Anton Paar RheoCompass) using a measuring plate with a diameter of 20 mm; the frequency was swept from 100 to 0.1 Hz at room temperature. For testing the propagation of mechanical waves in the samples, a full-field laser scanning vibrometer was used (LV-SC400, Sunny Optical Technology). A vibration motor was fixed at one end of the sample, and a function generator was used to drive the vibration source. The laser system then scanned designated areas on the sample to measure the vibration response. To ensure measurement accuracy, a layer of reflective metallic particles was sprayed onto the sample surface to improve laser reflectivity. For the omnidirectional interference resistance test, the samples were fixed in a designed testing mould, with a piezoelectric sensor placed below it. A constant vertical force was applied to the sample using a tensile testing machine. During the test, disturbance forces were applied from various angles through the outer frame of the mould (illustrated in Fig. 2g). By comparing the force signals received by the piezoelectric sensor with the applied forces, the interference-resistance performance of the hydrogel was evaluated. The SNR of transmitted mechanical signals was obtained using the same set-up. The electrical damping performance was evaluated using a broadband dielectric spectroscopy (Concept 80, Novocontrol). Before testing, the sample surface was dried with nitrogen gas to eliminate the influence of surface water. During the test, a 1-V a.c. voltage was applied, and the frequency was swept from 10−2 Hz to 104 Hz. All measurements were conducted at room temperature.

Cell cytotoxic test

The fibroblast cell line (L929) served as the model to evaluate in vitro cytotoxicity of the metahydrogel. Cells were purchased from ATCC and authenticated by the supplier through quality control testing. The prepared metahydrogel and reference samples were immersed in a phosphate-buffered saline–penicillin mixture. After incubation for a specified time, the extraction media were collected and applied to cell culture wells for an additional 24 h of incubation. Subsequently, the cell activity was evaluated using the Cell Counting Kit-8 assay. In parallel, a live/dead staining assay was performed to evaluate cell proliferation. The cell seeding and culture procedure were the same as described above. On days 1 and 3 of coculture, the cells were washed with phosphate-buffered saline and stained using a Calcein-AM/PI double staining kit. The cells were then observed under a fluorescence microscope. Live cells emitted green fluorescence, while dead cells showed red fluorescence.

Performance test and application validation of the MAP

Reliability evaluation of collected BP signals

To evaluate the reliability of the MAP in BP monitoring, we used a commercial photoplethysmography device as a reference. Given that commercial devices are susceptible to motion artefacts, the referenced BP results were collected under resting conditions, with different methods performed sequentially within a short timeframe to minimize the influence of physiological fluctuations. During BP signal acquisition with or without the MAP system, only hand movements were introduced to simulate motion artefacts, while avoiding full-body motion to prevent systemic physiological changes. The acquired arterial piezoelectric pulse signals were converted into BP waveforms using the method described in Supplementary Note 4. These waveforms were subsequently calibrated, enabling accurate and continuous BP monitoring under motion conditions.

Physiological signal acquisition in daily-life environments

The developed flexible wearable system (Fig. 5a) was utilized for physiological signal acquisition under daily-life environment. The pressure sensor and hydrogel electrode were used to collect BP and ECG signals, respectively. The signals were processed by dedicated acquisition modules and wirelessly transmitted via Bluetooth to a mobile phone or computer terminal for visualization and subsequent analysis. For consistency, light and routine signal preprocessing was applied to the collected data, as the metahydrogel already performs the primary artefact suppression. Artery pulses obtained from the piezoelectric sensor were gently smoothed using a Savitzky–Golay filter (window length 5) without altering their intrinsic amplitude and morphology. All preprocessing procedures were intentionally kept simple and processed in software to maintain consistency and reproducibility across datasets. During motion-state monitoring, a commercial posture sensor was attached to the participant’s arm to synchronously record physical activity data. For long-term monitoring, to reduce data volume and processing complexity, the sampling interval was set to once every 30 min, with each sampling lasting for 2 min. The physiological metrics were calculated as the average values over each sampling period. To simulate fatigue conditions as realistically as possible, participants are required to engage in prolonged and high-intensity simulated driving games. Meanwhile, a fatigue assessment questionnaire was administered to evaluate and confirm the participant’s fatigue state.

Theoretical simulation

Simulation of metastructure filtering

The metahydrogel consists of a soft matrix embedded with periodic structures formed by nanoparticle precipitation. These rigid nanoparticle–soft matrix units act as local resonators, absorbing mechanical energy at specific frequencies. For pressure (P) fluctuations in the matrix, the mechanical wave propagation follows the scalar wave equation:

$${\nabla }^{2}P-\frac{1}{{c}^{2}}\frac{{\partial }^{2}P}{{\partial t}^{2}}=0,$$

where \(c=\sqrt{K/\rho }\) is the wave velocity determined by the effective medium modulus (K) and mass density (ρ). The local resonator can be approximated as a mass–spring system

$$m\ddot{x}+c\dot{x}+{kx}={F}_{0}{{\rm{e}}}^{i\omega t},$$

where m is the effective mass of the nanoparticle cluster, and k is the elastic stiffness of the surrounding hydrogel. Resonance damping occurs near the natural frequency \({\omega }_{0}=\sqrt{k/m}\), where the absorbed energy is maximized. The given mechanical wave inputs were applied to evaluate the wave transmission and damping performance of the metastructural hydrogel.

Simulation of solvation structure based on molecular dynamics

Molecular dynamics were performed to investigate the structural and dynamic behaviour of the solvation system in Materials Studio 2022 software (Accelrys Software), using the Forcite module with the ab-initio-optimized COMPASS III force field. Molecular models were first constructed and subjected to geometry optimization to mitigate unfavourable conformations, and the canonical ensemble (N, V, T) simulations were carried out under periodic boundary conditions to mimic bulk-like conditions. The timestep was set to 1.0 fs, and the non-bonded interaction cut-off distance was set to 1.25 nm. The MSD of molecules was calculated as a function of time to analyse molecular mobility, and the diffusion coefficient (D) of molecules was obtained through the Einstein relation

$$D=\mathop{\mathrm{lim}}\limits_{t\to \infty }\frac{\left\langle {r}^{2}(t)\right\rangle }{{mt}},$$

where \(m=6\) for three-dimensional cases, and \({r}^{2}\) is the ensemble-averaged MSD. The radial distribution function g(r) was computed to quantify the probability of finding a given solvation pair at specific radial distance, offering insight into short-range ordering and solvation-shell structures. The binding energy (Ebind) between ion and solvent was calculated to evaluate the interaction strength as

$${E}_{\mathrm{bind}}={E}_{\mathrm{total}}-({E}_{\mathrm{ion}}+{E}_{\mathrm{solvent}}),$$

where Etotal is the energy of optimized solvation complex, and Eion and Esolvent are the energies of the isolated ion and solvent in the same environment, respectively. All simulation and analyses were conducted using default convergence criteria in Materials Studio.

Data processing and analysis

Concordance analysis of BP signals

The four-quadrant plot was constructed to assess the accuracy of SBP and DBP signals collected by different sensing systems (Fig. 2h). The horizontal and vertical axes represent the measurement difference and reference difference, respectively. The data points located in the first and third quadrants indicate consistent trends between the measured and reference BP. The concordance rate (CR) is calculated as

$$\mathrm{CR}=\,\frac{{N}_{{\rm{Q}}1}+{N}_{{\rm{Q}}3}}{{N}_{\mathrm{Total}}}\times 100 \% ,$$

where \({N}_{{\rm{Q}}1}\) and \({N}_{{\rm{Q}}3}\) are the number of points in the first and third quadrants, respectively, and \({N}_{\mathrm{Total}}\) is the total number of data points. To minimize the influence of measurement noise and physiological fluctuations, an exclusion zone of ±3 mm Hg was introduced in this analysis.

Recurrence analysis for ECG signal stability evaluation

The recurrence intensity mapping was performed to estimate the stability and dynamics evolution of ECG signals with different monitoring systems. For the collected ECG time series \(x\left(i\right)\), the phase space is reconstructed using time-delay embedding:

$$X\left(i\right)=\left[x\left(i\right),\,x\left(i+\tau \right),x\left(i+2\tau \right),\ldots ,x\left(i+(m-1)\tau \right)\right],$$

where \(\tau\) is the time delay and \(m\) is the embedding dimension. The similarity between any two-phase space vectors \(X\left(i\right)\) and \(X\left(j\right)\) was then computed on the basis of their Euclidean distance:

$${D}_{i,j}=\Vert X(i)-X(j)\Vert .$$

To improve interpretability, the distance \({D}_{i,j}\) was normalized and transformed into a recurrence intensity matrix (Fig. 4c), where colour gradients represent the similarity between different timepoints.

SNR analysis in the frequency domain of ECG signal

The power spectral density was conducted to estimate the SNR of ECG signal (Fig. 4f), the SNR at each frequency (\(f\)) was calculated as

$$\mathrm{SNR}\left(\,f\,\right)=10{\log }_{10}\left(\frac{{P}_{\mathrm{signal}}\left(f\,\right)}{{P}_{\mathrm{noise}}\left(f\,\right)}\right)\,(\mathrm{dB}).$$

This analysis allows a frequency-resolved assessment of signal quality, facilitating direct comparison between different sensing systems.

Classification accuracy assessment based on ROC

To assess the effectiveness and accuracy of fatigue evaluation, a comparative ROC analysis using two datasets: one collected with the MAP and the other collected without MAP. The extracted features were trained with a random forest classifier to predict subjective fatigue scales, which served as ground truth. ROC curves were generated by plotting the true positive rate (sensitivity) against the false positive rate (1 − specificity) across a range of classification thresholds. The AUC was calculated to quantify the discriminative power of each method.

Statistics and reproducibility

All statistical analyses were performed using OriginPro (version 2024; OriginLab) and MATLAB (R2022a). Detailed statistical methods and analysis procedures are described in the corresponding figure legends and Methods sections.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.