Sequential updating of minimum set of dynamics parameters by stochastic identification

A model-based controller is effective for highly accurate and fast control of a robot. The dynamics model is derived from its equations of motion, and the minimum set of the dynamics parameters are experimentally identified. However, the experimental data includes the influence of both noise and un-modeled dynamics, the obtained model will be one of approximated solutions. From these considerations, the approximated model has to well represent the robot dynamics around the reference motion, and for high accuracy, new data needs to be added every time an experiment is conducted, which causes a lot of computation. The authors have proposed stochastic identification method to obtain suitable parameters for control system design. However, in this method, because of computational complexity of weighted least square mean, it is difficult to add new motion data. In this paper, we propose a sequentially updating parameter identification method based on statistical properties of the conventional method. By synthesizing the covariance of the parameter error, the nominal parameters are updated. The proposed method is performed to experiments using a planar 3-link manipulator, and its effectiveness is evaluated.