Gait analysis with a custom Instrumented Insole with Embedded Machine Learning: Feasibility Study

This work presents the development and evaluation of a wireless instrumented insole for human gait real-time analysis. Equipped with an embedded processor, the device estimates the Center of Pressure (CoP) and identifies eight gait phases (8GP) in real-time. It can be used in both clinical and non-clinical settings via a simple Wi-Fi-to-USB receiver connected to a host PC.Validation involved three volunteers who walked 64 steps in a straight corridor and completed 16 laps in a room with a force platform. Machine learning models were developed, trained, and embedded in a 32-bit ESP32 microcontroller. Decision trees classified gait phases, while artificial neural networks (ANN) estimated xCoP, yCoP, and vertical ground reaction force (vGRF). Two modeling strategies were tested: subject-specific and cross-subject. The subject-specific models achieved an average F1-score of 0.74, while general models reached 0.61. For CoP and vGRF estimation, the models yielded R2 values of 0.94 (xCoP), 0.93 (yCoP), and 0.78 (vGRF). The proposed system estimates gait parameters at a 96 Hz sampling rate with fully embedded processing. Based on low-cost, open-source hardware, the insole offers an accessible solution for gait monitoring and integration with assistive rehabilitation technologies across diverse environments, including daily living contexts.