Pose accuracy compensation of parallel robots using RBF neural network

In designing and controlling a parallel robot, pose accuracy is one of the most important factors to be considered. Pose achieved by controlling joint values obtained from controller will, in general, deviate from the desired pose due to inaccuracies in the inverse kinematic model. In order to improve pose accuracy an approach using radial based function (RBF) neural network has been developed to calculate and interpolate joint correction of the joint space signal generated by controller using nominal parameters. The RBF neural network is trained on a database from pose measurement using coordinate measuring machine. After the learning phase, the network is tested on poses which were not part of the training data. The trained RBF neural network can be used to performed on-line pose accuracy compensation in task. Simulation and experiment results for a parallel robot are presented to show the effectiveness of the compensation method based on RBF neural network.