AI-Driven solid-state device to enable natural control of upper-extremity robotic exoskeletons

Abstract

The paper introduces the ‘Smart Handle’, a compact, lightweight, fully solid-state control device based on machine learning (ML) models, enabling intuitive manipulation across all three degrees-of-freedom (DOF) of translation and orientation. This handle-type control device utilizes low-cost, easily accessible sensors and materials. It addresses the need for natural control mechanisms in upper-extremity exoskeletons used in rehabilitation and occupational settings, offering position and orientation trajectories crucial for kinematic models. Additionally, individuals with motor difficulties using mobility chairs could benefit from its ability to learn and interpret diverse inputs as control commands. Through empirical study, it was demonstrated that the device adeptly learns and outputs continuous translation and rotation information from training sets focused on single DOF motions. Personalized calibration enables consistent movement classification with over 95% accuracy. Successful proof-of-concept experiments were conducted in exoskeleton and wheeled robot control applications, showcasing the Smart Handle's adaptability and reliability in diverse settings. This research underscores the device's potential across multiple domains, providing a foundation for enhanced intuitive control mechanisms in various technological applications.

Keywords

• Robotic exoskeletons
• control
• machine learning