Estimation and anticipation of elbow joint angle from shoulder data during planar movements

This paper describes the use of a feed-forward neural network for estimating and anticipating elbow joint angle. The method is based on mapping between six different combinations of muscles electromyographic signals (EMG) along with kinematics of the shoulder joint and the flexion/extension angle of elbow joint in four planar movements. Mean square error and cross correlation were used as quantitative criteria to reflect the performance of the method. We succeed to anticipate the future elbow angle up to 150 ms which is doing for the first time. For the most complete input combination which had also the best results, the cross correlation criterion between desired and anticipated splines for four movements respectively was %99.87, %99.90, %98.10 and %99.95.