Design and Development of Low-Complex Methodology for sEMG-Based Hand Activity Identification and Actuation of Soft Robots for Rehabilitating and Amputated Patients

Advancements in automated rehabilitation and pow-ered prosthetics necessitate a holistic approach involving wearable sensor-driven activity identification followed by generation of appropriate control signals to drive user-specific assistive tools. Here, we demonstrate an end-to-end system facilitating low-complex energy-efficient surfaceElectroMyography (sEMG)-driven human hand activity classification using simple and accurate k-means clustering and subsequent control of in-house fabricated soft robotic actuator for rehabilitative and prosthetic assistance. A Delsys Trigno EMG device was used to capture EMG data from the left flexor carpi radialis muscle of 20 healthy individuals corresponding to seven different hand activities. Thereafter, a COordinate Rotation DIgital Computer (CORDIC) based k-means algorithm was implemented to classify the hand activities after the execution of CORDIC-based Ensemble Em-pirical Mode Decomposition (EEMD) and Fast Independent Component Analysis (FICA) on the acquired data, followed by, necessary segmentation and feature extraction. Simultaneously, a lightweight pneumatically actuated soft finger system fabricated out of silicone was developed with the aim of emulating the classified hand activities. The novelty of this soft actuator lies in its strategic design that enables large bending, even at pressure as low as 160 kPa yet restricts the movement in one plane with almost no bulging added with improved force exertion during contact. Accurate control signals representing each of the classified activities were realized on a Raspberry Pi 4 board with python environment. For a thorough analysis, the classification accuracy linked with the proposed algorithm was evaluated for different feature sets and an accuracy as high as 96% was obtained. Thus, an end-to-end system was established for real-time control of the soft actuator with a latency of 3 seconds, making it suitable for rehabilitation and prosthetic control applications where low form-factor, fast response, and high accuracy are desired. The outcomes of this research hence motivate to consider System-on-Chip implementation as a future work towards the development of an effective rehabilitation management system.n