Wrist Angle Estimation with a Musculoskeletal Model Driven by Electrical Impedance Tomography Signals

Wrist kinematics estimation with muscle signals is a key issue in the field of wearable robots. In this study, we proposed a musculoskeletal-based-method driven by the Electrical Impedance Tomography (EIT) signals for continuously estimating wrist flexion/extension angles. The EIT-based interface can construct the conductivity distribution of the anatomical cross-sectional plane with a soft elastic sensing front-end, which is designed by our group. The estimation method took advantage of the flexor/extensor muscles' spatial information detected by the EIT-based interface to map the signals to the wrist angles. The whole model was designed with a musculoskeletal kinematic model, a muscular geometry model, and a mapping function between the EIT signals and the muscle morphological parameters. We validated the proposed method with intra-subject, inter-subject, and inter-posture cross-validations on 14 subjects in total. The results were compared with two data-driven algorithms (Lasso and kernel-based SVM). The muscle-model-based method was more robust to training data sizes than the other two methods. It achieved an average $R^{2}$ of 0.97 with 1:10 intra-subject CV and 0.91 with 2:12 inter-subject CV. The model also quickly overcame the effects of posture changes with a short-time feature update. The results of our study are comparable, if not better, to that of state-of-the-art. Future endeavors are worth being paid in this direction to get more promising outcomes.