Predicting the Motion of the End Effector in a Flexible Interconnected Manipulator with Neural Networks.

The estimation of the position of flexible robotic manipulators is a challenging task, especially for parallel and interconnected robots. This paper proposes a method to estimate the position of the end-effector of a flexible interconnected manipulator based on a virtual sensor principle. By using MSC ADAMS software, we developed a virtual prototype of the flexible interconnected manipulator and devised all feasible neural networks that map nonlinear relationships between angles of the active/passive joints to the position of the end-effector. The results indicated that it is possible to use the neural network estimate the position of the end-effector with a single passive joint and with high accuracy in both training-testing, with Mean Squared Error in the scale of 10 ⁻³ m, and unseen environments, with error bounded by less than 0.2 mm. The obtained results show the proficiency of the nonlinear relationships based on neural networks to predict the motion/position of the flexible manipulator with a promising and desirable capability.