영 공간을 이용한 로봇 매니퓰레이터의 그룹 관성 파라미터의 결정을 위한 수치적 방법

Abstract

This paper proposes a numerical method to identify an independent set of inertial parameters to be used in parameter estimation of robot manipulators based on a linear regression model. The set determined by this method can find solution independently and uniquely. Using an ideal sampled data, this paper verifies that the set determined by the proposed algorithm has unique solution. And this paper compares difference between the estimation performance using the sampled data of two different trajectories since the estimation performance for the set of inertial parameters varies to a sampled data of a trajectory. This comparison describes that a condition number of a regressor matrix of the linear regression model is directly related to the estimation performance. To improve the estimation performance in this paper, the regressor matrix for only one joint is used sequentially since a condition number of the regressor matrix for only one joint is lower than a condition number of the regressor matrix for all joints. Using the regressor matrix which has a low condition number, this paper shows that the grouped inertial parameters is estimated precisely for the robot system without friction. This paper also introduces a method for approximate estimation of friction force using the least-square (LS) method. The friction has a large effect on control. If the friction is estimated and compensated, the control performance can be improved further. However, the estimation for the friction is not perfect since a friction model for estimation is different to a friction model for simulation and the dynamic phenomenon of a dynamic friction model are not considered. In particular, a difference between the friction model for simulation and for estimation declines in estimation accuracy of the grouped inertial parameters since the ignored dynamic phenomenon of a dynamic friction model are crucial for estimation of the grouped inertial parameters. Here, note that the dynamic phenomenon of dynamic friction model are crucial for motion section that angular velocity is close to zero. On particular motion section that angular velocity is not close to zero, a difference between the friction model for simulation and for estimation can be minimized. To improve estimation performance for the friction and grouped inertial parameter, the sampled data of particular motion section, angular velocity is not close to zero, is utilized. Using this sampled data, the friction and grouped inertial parameters are estimated almost precisely.