Fast estimation of the kinematics of the parallel modules of a variable-geometry-truss manipulator using neural networks

In this paper, the kinematics of the parallel modules of a variable-geometry-truss (VGT) manipulator is studied, at the position level, by a multilayer neural network and via the kinematics of an equivalent serial mechanism. We show that the trained network provides good estimate of the exact solution to the kinematics of each module. Moreover, this indirect approach allows us to efficiently project the practical limits on the displacement of actuators onto the backbone-curve posture. This feature of the proposed approach is very attractive for the real-time configuration control and motion planning of a VGT manipulator, that comprises a large number of parallel modules connected in series