Validation of an Embedded Prototype for the Automatic Evaluation of Squats Using IMU and Binary Classification by AI

Quispe Chambi J.C.; Quispe Chambi J.M.H.; Acero Cáceres S.; Rojas Calla A.R. · Smart Innovation, Systems and Technologies (2026) · DOI: 10.1007/978-3-032-24262-4_30

Resumen

The present study addresses the validation of an embedded prototype designed for the automatic assessment of the squat, a fundamental strength training exercise. The main challenge lies in the lack of accessible, real-time, and objective tools for evaluating squat technique in unsupervised or real-world settings. Our proposal integrates an ESP32 microcontroller, an MPU6050 inertial sensor, and a Python-based graphical interface to measure tilt angles during movement execution. The novelty of the system lies in its compact, low-cost design that enables real-time binary classification of squat executions using a single inertial sensor and embedded artificial intelligence. A neural network model was trained on pre-labeled data (correct and incorrect squats), achieving a final accuracy of 87% and a loss of 0.25, demonstrating high reliability in distinguishing between proper and improper techniques. These results confirm the system’s feasibility for providing immediate feedback, contributing to the development of portable and affordable solutions applicable in sports training, physiotherapy, and injury prevention. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Instituciones: Universidad Nacional del Altiplano; Universidad Nacional de Juliaca