LEARNING CONTROLLING METHOD AND LEARNING CONTROLLER

摘要

PROBLEM TO BE SOLVED: To learn an orbit follow-up control that is a control object with high performance. SOLUTION: This controller is provided with many control modules 11, 12,... in which have a forward model 2, a backward model 3 and a responsibility signal predictor 4. Here the prediction error of each forward model 2 is calculated by comparing the state prediction of each mode 2 with an actual state that is feedbacked, a responsibility signal estimating instrument 5 calculates a responsibility signal based on the prediction error and an output of the predictor 4 and a final movement command is calculated by weighting an output of each model 3 with the responsibility signal and is given to a control object 9.