Musculoskeletal disorders (MSDs) are a significant workplace concern, with laborers and material movers being particularly vulnerable due to repetitive actions like squatting and lifting. Wearable robots offer a potential solution by reducing physical effort, but effective personalization is crucial. Current methods for estimating metabolic cost, the gold standard for assessing physical effort, rely on indirect calorimetry with respiratory masks—bulky and cumbersome systems unsuitable for real-world applications. These systems are slow to determine physiological responses and are limited to lab settings due to their size and weight. While heart rate variability (HRV) measured via ECG shows promise for assessing work intensity, commercial HR straps are uncomfortable and introduce motion artifacts. This research proposes a soft, flexible biopatch system integrated with an ankle-foot orthosis (AFO) to overcome these limitations, enabling real-time metabolic cost estimation and personalized robot assistance.

Existing studies utilize commercially available devices for metabolic cost estimation, but these devices often involve rigid masks and complex antenna systems, hindering field deployability. These systems suffer from slow response times compared to ECG processing, making real-time feedback for wearable robot adjustment difficult. Other approaches leverage HRV-RMSSD from ECG to estimate cognitive effort, but commercial HR detection devices lack conformal contact with the skin, creating discomfort and motion artifacts. The complexity and cost of epidermal electrode fabrication also pose challenges to mass manufacturing. This paper aims to address these limitations by developing a comfortable, lightweight, and readily manufacturable bioelectronic system.

A soft flexible biopatch (SFB) was developed, comprising a flexible printed circuit board (fPCB), a battery assembly, and micro-manufactured laser-patterned gold electrodes on a stretchable substrate. The fPCB included an ECG analog-to-digital converter, microprocessor, IMU sensor, and Bluetooth antenna. The electrodes were fabricated using a high-precision femtosecond laser micro-machining system, allowing for efficient and cost-effective manufacturing. The SFB's mechanical reliability, including flexibility and stretchability, was assessed computationally via finite element analysis (FEA) and experimentally through cyclic uniaxial stretching tests. A two degrees of freedom ankle-foot orthosis (AFO) with active plantarflexion was used for squatting assistance. Six participants (four male, two female, aged 20-30) performed squatting, walking, and running experiments with the SFB and AFO while metabolic costs were measured using indirect respiratory calorimetry. ECG signal quality was evaluated using signal-to-noise ratio (SNR). HRV-RMSSD was calculated from the filtered HR data. Motion data from the IMU were used for activity classification using a convolutional neural network (CNN). The relationship between metabolic cost, HRV-RMSSD, and perceived exertion (PE) was analyzed.

The SFB demonstrated high SNR (>25 dB) during all activities, with ECG quality comparable to a commercial HR chest strap (R² = 0.961). The study found a strong negative correlation (R = -0.758, p-value = 1.2e-7) between normalized metabolic cost and normalized HRV-RMSSD during squatting. The CNN model achieved 88% accuracy in classifying six different motions (running, elevated running, walking, elevated walking, standing, and squatting). A moderate negative correlation (R = -0.689, p-value = 6.6e-6) was observed between normalized physical effort (PE) and normalized HRV-RMSSD. The SFB showed minimal resistance change (<0.1%) during cyclic loading, demonstrating its durability and suitability for wearable applications. The device exhibited sufficient breathability and adhesion properties. The battery provided over 9 hours of continuous operation.

The strong correlation between metabolic cost and HRV-RMSSD demonstrates the potential to use the SFB for real-time estimation of metabolic cost, obviating the need for bulky respiratory masks. The high accuracy of the motion classification algorithm, combined with the SFB's ability to capture HRV, opens up possibilities for personalized exoskeleton control based on the user's physical effort and activity. The findings contribute towards developing field-deployable exoskeletons capable of providing adaptive assistance, tailored to individual needs. The moderate correlation between perceived exertion and HRV-RMSSD highlights the subjective nature of perceived effort, warranting further investigation into refining the assessment of physical effort.

This study successfully developed and validated a wearable bioelectronic system for real-time estimation of metabolic cost and physical effort. The miniaturized, comfortable, and durable SFB, in conjunction with the AFO and machine learning algorithms, provides a significant advancement in human-robot interaction. Future research will focus on large-scale studies incorporating additional physiological signals (muscle activity, temperature, and stress levels) for more comprehensive real-time optimization of exoskeleton assistance.

The study involved a relatively small sample size (six participants), limiting the generalizability of the findings. The correlation between PE and HRV-RMSSD was moderate, suggesting that subjective perception of effort needs further refinement. The study focused primarily on squatting; future research should expand to other activities and populations. The current AFO is a tethered system; future work should investigate the use of a wireless AFO.