Particle filter applied to a manipulator with two flexible links

Traditional methods estimate and control the manipulator end-pointpsilas position indirectly by using the models of its kinematics and jointspsila measurements. If the measurement of end-point is available directly, and use extended Kalman filtering to estimate the state vector of the model dynamically, we can obtain more accurate value about the end-pointpsilas position than the traditional method. But extended Kalman filtering has linearization error, and the noise must be Gaussian. Particle filtering is applicable for strong nonlinear and non-Gasussian system. Itpsilas no need to linearize the model, and can get the approximate optimal estimation value based on the physical model. This paper first introduces the dynamic model of a planar manipulator with two flexible links and particle filtering algorithm. Then the Particle filtering is applied to estimate the state vector of manipulator model when jointspsila angles and end-pointpsila position are observations. Under the same non-Gaussian simulation environment, the comparisons between particle filtering and extended Kalman filtering about estimation accuracy of the end-pointpsilas position indicate that the former have the better performance.