Global motion planners have been proposed for closed-loop robot based on the same paradigm than has been proposed for serial chains. First a sparse representation of the configuration space of the robot is constructed as a set of nodes. This is somewhat more complicated than for serial chain as the closure equations of the mechanism should be satisfied. Then a motion planning query consists simply in connecting the start and goal points through an appropriate set of nodes (usually minimizing the length of the trajectory). But such motion planner should be complemented by a local motion planner that addresses the following issues: \begin{enumerate} \item ensure that two successive nodes belong to the same robot kinematic branch (otherwise connecting these nodes will require to disassemble the robot) \item verify that all poses between nodes satisfy the robot constraints (if possible taking into account the uncertainties in the robot modeling) \item eventually try to shorten the trajectory length \end{enumerate} We present such a local motion planner that addresses all three issues and illustrates its use on a Gough parallel robot.