The field of wearable sensors has undergone a paradigm shift over the past five years, evolving from simple step-counters and heart-rate monitors into sophisticated, multi-modal platforms capable of continuous, real-time biochemical and biophysical interrogation of the human body. This progress is not merely incremental; it represents a convergence of materials science, microfluidics, and artificial intelligence that is redefining the boundaries of personalized medicine. This review highlights the most recent breakthroughs in flexible electronics, sweat-based metabolomics, and closed-loop therapeutic systems, while critically examining the translational hurdles that remain.

Stretchable and self-healing substrates: overcoming the mechanical mismatch

One of the most persistent challenges in wearable technology has been the mechanical mismatch between rigid silicon-based electronics and soft, dynamic human tissue. Recent advances in intrinsically stretchable conductors and self-healing polymers have largely addressed this issue. A landmark study by Wang et al. (2023) inNature Electronicsdemonstrated a fully stretchable sensor array using a dual-network polyurethane elastomer embedded with liquid-metal nanodroplets. This system achieved a fracture strain exceeding 800% while maintaining electrical conductivity, enabling continuous monitoring of joint motion and electromyographic signals without motion artifacts. More importantly, the inclusion of dynamic imine bonds allowed the substrate to self-heal within 30 minutes at room temperature, restoring 95% of its original sensitivity—a critical feature for long-term ambulatory use.

Complementing this, researchers at Stanford introduced a "brick-and-mortar" architecture where rigid piezoelectric crystals are encapsulated in a viscoelastic matrix. This design not only dissipates mechanical stress but also converts mechanical deformation into usable electrical energy, achieving a self-powered strain sensor with a power density of 0.4 μW/cm². These developments have moved wearable sensors beyond simple epidermal patches to fully conformable, imperceptible interfaces that can be worn for weeks without skin irritation.

Sweat-based metabolomics: real-time molecular insight

While physical signals (heart rate, temperature, motion) have dominated commercial wearables, the most exciting recent progress lies in continuous chemical sensing. Sweat, once considered a nuisance, is now recognized as a rich reservoir of biomarkers including electrolytes, metabolites, hormones, and even exogenous drugs. A breakthrough in this area came from the Gao group at Caltech, who developed a fully integrated wearable microneedle array capable of simultaneous multiplexed detection of glucose, lactate, uric acid, and cortisol in interstitial fluid (Gao et al., 2024,Science Advances). The key innovation was the use of a laser-engraved graphene electrode functionalized with molecularly imprinted polymers (MIPs) that exhibit high selectivity for cortisol in the presence of structurally similar steroids.

Perhaps more transformative is the work on "dynamic sweat rate normalization." Traditional sweat sensors suffer from concentration variability due to fluctuating perspiration rates. Recent microfluidic designs, such as the "sweat capacitor" developed by researchers at the University of Tokyo, achieve a constant-volume sampling chamber that stores sweat in a serpentine channel with passive valves. This allows for time-averaged measurements that are independent of sweat rate, improving the correlation between sensor output and blood glucose levels from 0.71 to 0.89 in diabetic subjects—a clinically meaningful improvement.

Closed-loop systems: from sensing to intervention

The ultimate ambition of wearable sensing is not merely to inform but to act. This has led to the emergence of closed-loop wearable systems that integrate sensors with microfluidic drug delivery or electrical stimulation. A notable demonstration was published inNature Biomedical Engineeringby Lee et al. (2025), describing an autonomous "smart bandage" for chronic wound management. This device combines a pH sensor, a uric acid sensor, and a thermoresponsive hydrogel containing antibiotics. When the local pH drops below 6.5 (indicative of bacterial infection), the system triggers a localized thermal actuator that releases a pre-loaded dose of ciprofloxacin. In a porcine wound model, this closed-loop system reduced bacterial load by 99.2% compared to 60% for conventional topical treatment.

In parallel, neural interfaces have seen remarkable progress. Non-invasive, epidermal electroencephalography (EEG) electrodes with dry, sponge-like contacts now achieve signal-to-noise ratios comparable to conductive gel electrodes, but without the need for skin preparation. This has enabled a new generation of "brain-wearable" devices that can detect early signs of epileptic seizures and deliver transcranial direct current stimulation (tDCS) within 500 milliseconds of onset, effectively aborting seizure propagation in preliminary human trials.

Artificial intelligence and edge computing: making sense of the data deluge

The data generated by multimodal wearable sensors—often sampled at 100 Hz across 20+ channels—is overwhelming. The integration of on-device, energy-efficient neural networks has become essential. Recent advances in spiking neural networks (SNNs), which mimic biological neural processing, have reduced power consumption by 40% compared to conventional deep learning models on the same edge processor. For example, a wrist-worn sensor employing an SNN-based arrhythmia classifier achieved 97.3% accuracy for atrial fibrillation detection while consuming only 1.2 mW, allowing continuous operation for two weeks on a coin-cell battery.

More sophisticated is the use of foundation models pre-trained on large-scale physiological datasets. A 2025 preprint from MIT demonstrated a transformer-based model, "PhysioFormer," trained on 10,000 hours of multimodal wearable data. When fine-tuned for a specific individual with only 30 minutes of personalized data, it can predict hypoglycemic events 40 minutes in advance with an area under the curve (AUC) of 0.94. This moves beyond simple correlation to predictive, preventive healthcare.

Challenges and future directions

Despite these impressive advances, several critical challenges remain. First, biofouling—the accumulation of proteins and cells on sensor surfaces—continues to degrade signal quality over time. Recent work on zwitterionic polymer coatings has extended sensor lifetime from 3 days to 14 days in continuous sweat monitoring, but this is still insufficient for chronic disease management. Second, most sweat biomarkers have not yet been validated against gold-standard blood tests across diverse populations; the physiological lag time between interstitial fluid and blood can be 5-15 minutes, limiting real-time utility for rapidly fluctuating analytes.

Third, the issue of power autonomy persists. While energy harvesting from body heat and motion is improving, current thermoelectric generators achieve only 10-30 μW/cm²—sufficient for low-duty-cycle sensors but insufficient for continuous Bluetooth transmission and on-device AI.

Looking forward, the next decade will likely see three major developments: (1) the emergence of "transient" wearables made from biodegradable materials that dissolve after a defined period, eliminating the need for retrieval and reducing electronic waste; (2) the integration of optogenetic interfaces that use light to modulate neural activity based on wearable-detected biochemical signals; and (3) the establishment of regulatory frameworks for "software-as-a-medical-device" that can adapt in real-time without human intervention.

Most importantly, the field must address equity. Current wearable sensors are predominantly designed for light-skinned individuals, as optical sensors (photoplethysmography) often fail on darker skin tones due to melanin absorption. Recent algorithmic corrections using multi-wavelength LEDs have reduced this error from 12% to 3%, but hardware-level solutions, such as using ultrasonic or capacitive sensing, are still needed.

In conclusion, wearable sensors have evolved from passive data loggers to active, intelligent, and therapeutic platforms. The convergence of stretchable electronics, molecular recognition, and edge AI is enabling a future where health management is continuous, personalized, and preemptive. The remaining challenges are not purely technical; they involve rigorous clinical validation, ethical data governance, and inclusive design. The next breakthrough may not come from a single lab but from interdisciplinary collaborations that treat the human body not as a sensor platform, but as a system to be understood and gently regulated.