Statistical Methods for Estimating the Dynamical Parameters of Manipulators

The determination of the dynamical parameters of robot manipulators is crucial in many applications where model based control architectures are needed to match stringent performance requirements. Unfortunately only the kinematic model of the robot is usually available whereas the dynamical parameters are unknown and very difficult to compute by first principle or CAD analysis. In the last decades several algorithms have been proposed to identify these parameters mainly based on the least-square analysis. In this paper we present an identification method based on a statistical algorithm never used so far in robotics, which brings new insight into the understanding of the identified parameters and improves robustness of the computation. We think that this approach represents a significant improvement as compared to using standard statistical tools, as shown by results of the identification of the Puma 200 robot.