A telerobotic manipulator is under study, which transmits to object under control an operator hand motion in three dimension (3D) space. The manipulator sensor is precise three axes MEMS-accelerometer. Under the manipulator development, a problem of filtering control signal arises, since a noise is produced by operator hand tremor process. The problem solves with the Bayesian filtering in use. So the Kalman smoother is applied with linear system parameter estimates by the Expectation-Maximization algorithm.