Multimodal Estimation of Endpoint Force During Quasi-Dynamic and Dynamic Muscle Contractions Using Deep Learning

Accurate force/torque estimation is essential for applications such as powered exoskeletons, robotics, and rehabilitation. However, under dynamic conditions a challenging due to changing joint angles, force levels, muscle lengths, movement speeds. We propose novel method accurately model the generated isotonic, isokinetic (quasi-dynamic), fully conditions. Our solution uses deep multimodal CNN learn from EMG-IMU data estimate elbow flexion extension, both intra- inter-subject schemes. The proposed extracts representations EMG (in time frequency domains) IMU domain) aggregates them obtain an effective embedding estimation. describe new dataset containing EMG, IMU, output data, collected number of different experimental conditions, use this evaluate our method. results show robustness approach in comparison other baseline methods well those literature, setups validation obtained R2 values are 0.91±0.034, 0.87±0.041, 0.81±0.037 intra-subject 0.81±0.048, 0.64±0.037, 0.59±0.042 scheme, during isokinetic, contractions, respectively. Additionally, indicate that improves significantly when kinematic information (IMU data) included. Average improvements 13.95%, 118.18%, 50.0% (intra-subject) 28.98%, 41.18%, 137.93% (inter-subject) contractions respectively achieved.