Robot parameter identification in periodic motion for dynamic controllers and priority assessment of identified parameters based on sensitivity analysis

A robot motion tracking control system consists of feedback and feedforward controller. The former is realized by state feedback such as PD controller or a dynamic controller such as PID controller, while the latter is obtained by inverse dynamics analysis based on a robot dynamics model. To obtain the robot dynamics model, an approximate solution of minimum set of dynamics parameters is identified from experimental data using the least-squares method, etc. The accuracy of the approximation, i.e., which parameters have priority, is highly dependent on the robot motion and feedback controller used. We have proposed a stochastic parameter identification method using sensitivity analysis to determine the priority of the parameters. However, a state feedback controller is assumed to be used for this approach. In this paper, we derive a stochastic parameter identification method for small periodic motions of the end-effector, such as weaving motion of a welding robot, using a dynamic controller, and verify the effectiveness of the proposed method through experiments using a planar three-link manipulator. Based on the error ellipsoid of the parameters, we show that the priority of the parameters to be identified changes as the controller changes, and that the priority of the parameters that the controller will compensate decreases.