EXTENDED KALMAN FILTER ESTIMATION OF HUMAN ARM CONFIGURATION

The Kalman Filter (KF) is a mathematical tool well suited for an algorithmic implementation that estimates the state of a linear dynamic system by exploiting the knowledge of the system’s dynamics, its statistics and a set of measurements. In practice however, the linearity conditions are often not satisfied, but some extensions of the filter algorithm can be implemented in order to overcome this problem. One possible extension is represented by the Extended Kalman Filter (EKF) which is one of the best known non-linear filter versions. The human arm is a highly non-linear system. Hence, the EKF algorithm can be exploited to estimate its joint-space configuration through few motion capture markers placed along the human arm. Other approaches, such as Least Squares (LS) and Linearized Kalman Filter (LKF) are tested to show how the EKF leads in general to better joint variable estimates. The design and validation of such estimation procedures are performed in a simulation environment, offered by Simulink and Simscape platforms in Matlab. Here it is possible to develop the 7 Degrees of freedom (DoF) robot manipulator (RM) which models the human arm and to find a suitable combination of the number of markers on the shoulder, forearm and hand for which it is possible to compute the state estimates, namely for which singular configurations are avoided. The aforementioned estimation algorithms are implemented in the real system and they are closely compared.