Model Learning with Local Gaussian Process Regression based on Overlapping Clustering
Seung-Yoon Cho, Ju-Jang Lee · 제어로봇시스템학회 각 지부별 자료집 · 2011
In robot manipulator position control problem, modeling of inverse dynamics is important because it can allow accurate robot control using model based control methods such as PD control with computed feedfoward. However, modeling rigid-body inverse dynamics is not accurate in some case, because of unmodeled nonlinearities such as hydraulic cable dynamics, complex friction or actuator dynamics. Instead of rigid-body dynamics, nonparametric regression such as Locally Weighted Projection Regression (LWPR), Gaussian Process Regression (GPR) is proposed as alternative. LWPR is fast, but it is difficult to tune because of many user-parameters. GPR has high accuracy but low computation speed. High complexity of computation is drawback of GPR. To improve the low computation speed, Local Gaussian Process Regression (LGPR) is proposed. In this paper, Modified Local Gaussian Process Regression (MLGPR) is suggested for improving accuracy and computation time of Local Gaussian Process Regression (LGPR). In MLPGR, overlapping method is used for partitioning the training data. MLGPR uses overlapping clustering method, depending on the similarity measure. Proposed method is demonstrated by 2-dimension regression example and learning inverse dynamics of SARCOS arm.