Background
Motion capture technology is a key enabler for the digitalization of the human body in VR/AR/MR, film production, and sports rehabilitation, and has attracted broad interdisciplinary attention. However, current commercial systems—such as optical and inertial capture systems—still face significant limitations: cumbersome wearability, restricted usage scenarios, and interference with natural activities, which seriously affect user experience and hinder the practical implementation of the metaverse. To overcome environmental constraints, researchers have explored wearable solutions, including flexible strain sensors, torque sensors, and miniaturized inertial units. Nevertheless, most of these approaches still rely on full-body sensor suits, which are complex to wear and often unacceptable to users, while the wiring required for numerous sensors poses additional challenges to ergonomics and comfort.
Insoles, as a daily necessity for everyone, hold great promise as an ideal and imperceptible data input terminal for the metaverse. Many studies have attempted to use insole-embedded sensors to collect and analyze lower-limb motion data. However, most of these efforts have been limited to classifying finite motions and fall far short of enabling seamless interaction and natural control as demanded by the metaverse. To date, no study has reported accurate joint position estimation relying solely on insole sensors, leaving a clear research gap in this area.

Research Progress
As shown in Fig. 1, this study proposes a fabrication method for flexible pressure sensors based on laser-selective cutting and dip-coating processes. The sensor demonstrates an ultra-high working pressure range (3770.9 kPa), high sensitivity (2.68 kPa⁻¹), and excellent durability, maintaining stable performance after more than 4 million pressure cycles. Based on simulations of plantar pressure distribution across central and edge regions under different postures, a rational sensor array layout was designed to enable comprehensive pressure signal acquisition during motion.

As shown in Fig. 2, the authors developed a flexible pressure-sensing insole integrating 16 sensing units. Combined with a self-developed array data acquisition system, real-time plantar pressure data collection was achieved. With a trained convolutional neural network, the system classified 10 different static and dynamic postures with an accuracy of 95.5%.

The authors further investigated real-time monitoring for rehabilitation exercises. For patients with weakened or injured legs, lifting and squatting are common rehabilitation movements. However, effective remote monitoring tools are lacking, making it difficult to provide objective evaluation of movement range and accuracy. As illustrated in Fig. 4, participants simulated both normal leg lifts and insufficient lifting height while wearing the system. The flexible pressure sensor array captured plantar pressure distributions in real time and predicted joint states accordingly. By calculating the ratio between the actual thigh-lifting angle and the target angle, the system automatically determined whether the movement met the requirements and quantified deviations, thereby providing reliable support for evaluating rehabilitation training quality.

Future Prospects
This work proposes a novel lower-limb motion capture system that, for the first time, combines a flexible pressure sensor array with a Transformer-based temporal regression model. The system enables accurate estimation of lower-limb joint positions using only insole-embedded sensors. It offers metaverse researchers a new data input terminal for human lower-limb motion, effectively overcoming the dependence of existing capture technologies on complex sensor setups and environmental constraints, and thus may accelerate practical applications of the metaverse. Furthermore, this study pioneers a new approach for high-precision lower-limb posture estimation based solely on flexible pressure sensor arrays, opening a promising research direction for the flexible sensing community with broad application potential.

The complete study is accessible via DOI: 10.34133/research.0835