A NEURAL NETWORK MODEL FOR COMPENSATING ROBOT KINEMATICS ERROR

Advanced industrial robots are commanded to accomplish different tasks with program that is executed in computer. The operating software provides users with information on position and orientation of theend effectors by computing them as functions of the joint variables. These functions are generally not exact enough such that differences between the computed and the actual position can be significant. The differenceas high as 10mm may be possible for the robot with roughly 0.1mm repeatability. Some other methods forcalibrating and compensating robot kinematic error need to analysis error sources,to parametrize the effectsof these error and to estimate these parameters,usually there can be up to 72 error parameters for a six-jointrobot.In this paper,a neural network model for compensating robot kinematic error is proposed.An im-proved BP learning algorithm is also presented.Experimental results in RM501,a five-degree-of freedom articulated robot,show that the proposed method is efficient and feasible.