Artificial Neural Network Prediction of Deflection Maps for Cable-Driven Robots

In this paper, the learning models of cable-driven robots are developed applying the artificial neural network (ANN). For known input and output data and known relationships (regression problem), the deflection maps of cable-driven parallel robots are predicted utilizing a multi-layer ANN. Two cable robots, a planar robot and a translational spatial robot, are examined to evaluate their models. The deflection maps of these cable robots are generated using the ANN and a non-linear optimization method. The predicted deflections of the ANN models, using much less number of poses for training, are highly satisfactory and comparable to the results obtained by a nonlinear optimization method throughout the pertinent discretized workspaces. In addition, ANN models could predict the deflections for poses that the nonlinear optimization methods may not. Moreover, with variations in robot/task parameters, such as payload, ANN models may predict accurate deflections.