Genetic algorithms based identification method for a robot arm

Abstract

An identification method for inverse dynamics of a robot arm based on genetic algorithms (GA) is considered. It is shown that the GA are able to search robot parameters effectively and accurately even if the robot has low resolution position encoders. It is possible because the method requires only position feedback and there is no need to find out the speed and acceleration of the links that usually can be done only through finite difference calculations that cause dramatic errors during identification. The effectiveness of the algorithm is demonstrated on the example of parameter identification of the PUMA 560 robot (for second and third links). The computed torque control method using results of the identification provides desired control quality with tracking error of 1 encoder pulse.