Modeling bimanual coordination using back propagation neural network and radial basis function network

Previous studies demonstrated that bimanual coordination can assist to rehabilitation program for patients with hemiplegia by improving their motor functions. Moreover, in addition to the rehabilitation assistance, bimanual coordination can also be used for prosthesis users to improve the usability of prosthesis. Intention detection and motion control algorithms for one hand case have been investigated in many studies in the literature. On the other hand, only few studies have focused on the model of bimanual coordination, and these studies are lack of the sufficient investigation of kinematic and dynamic parameters of the models. The purpose of this study was to model the bimanual coordination for developing rehabilitation support robots and prosthetic arms. In this study, artificial neural networks (ANN) were employed to examine the kinematic and dynamic parameters by taking them as feature vectors to ANN for defined bimanual tasks. For this purpose, two different ANN algorithms were selected; i) Back Propagation Neural Network (BPNN), and ii) Radial Basis Function Network (RBFN). The parameters were calculated from the results of a set of experiment, in which 3 different bimanual coordination tasks were recorded. Training time, test time, and error rate were used as evaluation criteria for performance analysis and comparison of the models. As a result, it was made clear that, bimanually coordinated behavior could be predicted at a certain level of error rate, within acceptable computation time, moreover, the trajectory input is important to predict the behavior of the bimanual coordination, furthermore, BPNN is more robust for non-periodic movement than RBFN.