Motion-intensity estimation for upper limb exoskeletons: Optimizing support for dynamic assistance

Abstract

In response to the escalating concern regarding work-related injuries, particularly musculoskeletal disorders (MSDs) deriving from repetitive movements and strenuous postures like overhead activities, this Master's Thesis goes into the field of upper limb exoskeletons for shoulder support, focusing on determining the effort required for specific tasks, such as lifting objects, and providing dynamic assistance, based on the effort estimated, through an exoskeleton. The incidence of upper limb injuries has shown a concerning increase over the years, emphasizing the urgency for innovative solutions to mitigate the risks posed by overhead work and heavy lifting. The study utilizes surface electromyogram (sEMG) electrodes and inertial measurement units (IMUs) together with a neural network comprising a combination of a convolutional neural network (CNN) and long short-term memory (LSTM) cells. By leveraging the sEMG signal as the ground truth and IMU data as input, the neural network, consisting of CNN and LSTM, offers superior results in estimating effort levels. The placement of electrodes on the bicep, anterior deltoid, and middle deltoid, and of the IMUs on the lower abdomen, torso, upper arm, and forearm enables the retrieval of muscle activation and kinematic data, essential for determining the level of effort during tasks. The processed signals are used to adjust the exoskeleton parameters, ensuring appropriate assistance levels and enhancing operator safety. This research contributes to the optimization of support for dynamic assistance in upper limb exoskeletons, offering insights into personalized assistance based on user motion dynamics. The results conveyed from the analysis of the IMU data and muscle activation confirm what was already understood from the literature, putting it into the more specific context of lifting tasks. Moreover, the results obtained from the neural network and effort estimation algorithm represent a strong foundation for future development, being the first to explore the collaboration of IMUs and EMG sensors for effort estimation.