HUMAN ACTIVITY RECOGNITION AND OPTIMIZATION OF BIPED EXOSKELETES THROUGH ARTIFICIAL INTELLIGENCE: AN INTEGRATED APPROACH

This paper explores the integration of inertial sensor-based human activity recognition (HAR) with the optimization of bipedal exoskeletons using artificial intelligence (AI) techniques. The motivation for the study stems from the need to improve the adaptability and energy efficiency of exoskeletons for practical applications. The specific hypothesis is that combining HAR with reinforcement learning (RL) can lead to personalized and efficient control strategies. The aim of the research is to develop a robust HAR system for classifying activities such as normal walking, stair climbing/climbing and sitting/standing, and to optimize exoskeleton control through AI-based simulations. The methodology involves preprocessing sensor data (accelerometer and gyroscope) by segmentation and feature extraction, followed by supervised classification with Support Vector Machines (SVM) and Random Forest, and RL optimization in simulated environments such as Webots. Preliminary results indicate an HAR accuracy of 92% and a 15% reduction in metabolic cost by RL, improving stability and user comfort. This innovative approach contributes to exoskeleton design by reducing manual adjustments, with potential applications in rehabilitation and physical augmentation.