Robots have been playing an important roles on our life. In various eld such as entertainment, military, space and medical elds, the precise control is required according to the increase of the importance on robot. In robot manipulator position control problem, modeling robot inverse dynamics is important because it can allow accurate robot control using computed torque control and PD control with computed feedback. However, modeling rigid-body inverse dynamics is not simple and 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. Locally Weighted Projection Regression is fast, but it is dicult to tune because of many user-parameters. Gaussian Process Regression has high accuracy but low computation speed. In other word, high complexity of computation is drawback of Gaussian process regression. In Gaussian Process Regression, the inverse of Gram matrix is a signicant problem and it dominates the computation time. To improve the low computation speed, there are many methods such as approximation method and Local Gaussian Process Regression (LGPR). In approximation method, the approximation of inverse of Gram matrix is proposed and in local Gaussian Process Regression, the training data is divided into local training data using Gaussian kernel. It generates M local models. After partitioned the training data, the local model is trained. When test data is given, each local model predicts the local prediction. The total prediction value is weighted average of M local prediction values. The weight is a similarity measure and it can be calculated by Gaussian kernel.
In this paper, Modifed Local Gaussian Process Regression (MLGPR) is suggested for improving accuracy of Local Gaussian Process Regression. Modied Local Gaussian Process Regression is used adaptive method for partitioning the training data. Modied Local Gaussian Process Regression uses multiple model generation threshold wgen values depending on the local target variance. Also MLGPR uses overlapping partition method instead of crisp partition. Proposed method is demonstrated by 2-dimension regression example, learning inverse dynamics of SARCOS arm and learning and control of simple two link planner arm. The result of simulations will be compared with other method such as Gaussian Process Regression, Local Gaussian Process Regression. As a result, the accuracy of Modfied Local Gaussian Process Regression is improved and computation time is increased slightly. The result represent Modied Local Gaussian Process Regression has high accuracy as compared with Local Gaussian Process Regression.
시스템의 역동역학(Inverse dynamics)을 잘 아는 경우, 모델기반의 제어방법을 사용하여 시스템을 정확하게 제어할 수 있다. 하지만 실제의 시스템의 경우 다양한 요소로 인해 모델링 에러가 발생한다. 기존의 사용되던 강체 동역학(Rigid body dynamics)이 아닌 이를 대체하기위한 방안으로 논파라메틱 회귀방법(Nonparametric regression)이 제시되었다. 본 논문에서는 기존에 사용되던 국지 가우시안 프로세스 회귀분석(Local Gaussian Process Regression)이 가지는 문제점을 찾고 이를 해결하기 위해 변형된 국지 가우시안 프로세스 회귀분석 방법(Modified Local Gaussian Process Regression)을 제시하였다. 기존의 국지 가우시안 프로세스와 같은경우 모델간의 경계에서 날카롭고 급격한 예측을 만들어내는 경계문제(Boundary problem)이 발생한다. 이는 기존의 방법이 트레이닝 데이터(Training data)를 구획(Partitioning)하는 과정에서 단일한 방법을 이용하여 데이터를 구획하기 때문이다. 따라서 제시한 방법에서는 경계문제를 해결하기 위해 단일한 구획방법이 아닌 중첩된 구획방법(Overlapping partitioning)을 사용하였다. 또한 이를 검증하기 위해 세가지 경우에 대한 시뮬레이션을 수행하였다. 결과적으로 제안한 변형된 국지 가우시안 프로세스 회귀분석 방법이 중첩된 구획방법을 이용하여 경계문제를 해결함으로써 좀 더 정확한 회귀분석이 가능함을 보여준다.
