Visual tracking for a manipulator using fuzzy rule-based system

A vision-based object tracking algorithm is proposed for a robot manipulator with stereo cameras, where the fuzzy rule-based systems (FRBS) are employed to represent the mapping from image features for the end position of the robot and the target to joint angle space variables for each angle of the robot. Wang and Mendel's (1992) rule extraction method is used in such a way that the initial rule table is built up with irregularly generated input-output pairs in a certain work space and updated every vision sampling time with newly generated input-output pair corresponding to the end position of the robot during operation. Once the rules are constructed, estimated reference joint angles can then be obtained using FRBS where image coordinates of the target object are applied as input variables, and then a PD control scheme can be used to generate control inputs to move the robot. Computer simulations and experiments were performed to illustrate the effectiveness of the method proposed.