A Simulation-based Feasibility Study of a Proprioception-inspired Sensing Framework for a Multi-DoF Shoulder Exosuit

The compliant nature of exosuits makes them ideal for providing assistance to complex joints like the shoulder. Exosuits require soft, compact and accurate sensing units for reliable feedback control. In this work, we introduce an OpenSim simulation-based prototype of a proprioception-inspired sensing framework for a multi-DoF shoulder exosuit. The prototype is used to study the feasibility of the sensing system concept to accurately track multiple degrees of freedom (DoFs) of the shoulder simultaneously. The sensing system fuses data from 4 custom string poten-tiometers (SPs), that work together to sense the joint angles at the shoulder. The tendon-routing of the SP modules in the exosuit is proprioception-inspired and based on the organization of the muscles influencing shoulder movement. The sensor fusion/mapping of the simulation data from multi-sensor space to joint space is a multivariate multiple regression problem and was solved using Multi-Layer Perceptron (MLP) & Long Short-Term Memory (LSTM) neural networks. A simulation of the framework in OpenSim on 200,000 random shoulder movements achieved a root mean square error (RMSE) of $\approx 0.2^{\mathrm{o}}$ when trained on 70,000 random movements and tested on 130,000 random movements in both DoFs simultaneously.