A SUPPORT-VECTOR RULE BASED METHOD FOR THE CONSTRUCTION OF MOTION CONTROLLERS VIA NATURAL LANGUAGE TRAINING

This paper presents a support-vector rule based method for the construction of motion controllers via natural language training. With support vector machine theory, the support vector rules are extracted from fuzzy control rules composed of natural language instruction, and then a motion controller based on them according to a given control accuracy is obtained. In this way, both the number of control rules and the structure of controller are reduced and simplified, which makes the controller constructed by natural language training more applicable in practice. Experiments of natural language training for motion primitives control have been done on a wheeled robot.