Estimation of three-dimensional ground reaction forces using low-cost smart insoles

Estimating three-dimensional ground reaction forces (GRFs) is critical for health monitoring, exercise science and rehabilitation. Traditional methods such as force plates and instrumented treadmills are often expensive, cumbersome and limited to laboratory settings. While inexpensive portable sensing devices are promising to address these limitations, they require advanced algorithms to ensure the reliability of the estimation results. This study introduces a dual-stream attention model for estimating 3D GRFs via low-cost, wearable sensing insoles. The proposed model dynamically distributes weights between the CapSense and IMU data streams, effectively capturing the complementary strengths of the two sensors to significantly improve the accuracy and robustness of the GRFs prediction across a wide range of physical activities. This study highlights the experimental setup and exercise protocol designed for efficient and reliable data collection. The proposed machine learning model was validated against gold standard force plate measurements, which achieved the lowest normalised root mean squared error (NRMSE) prediction error of 4.1 %. This insole-based measurement of GRFs provides a practical and cost-effective solution for continuous health monitoring, personalised exercise training and advanced rehabilitation, making it an important addition to wearable technology and biomechanics.