In the actuator state monitoring of a time varying human multi-joint arm dynamics, typical issues are compounded by problems related to the uncertainty factor consisting of measurement noises and modeling error of the rigid body dynamics. In general, the uncertainty factor is under the case of non-Gaussian noises. In this paper, for improving the monitoring, a robust filter system based on a score function approach is modified. The score function is associated with U_D factorization algorithm. The selection of the shape parameter in the monitoring filter is discussed. Examples of the proposed method for an experiment-based human arm model show better accuracy and robustness compared with standard Kalman filter.