Visual servo dynamics for robots based on an extended Kohonen network

A concept is proposed for the visual servoing concerned with robot dynamics, in which friction is considered. In this developed concept, the visual servo dynamics is learned by an extended Kohonen network. Based on the learned dynamic model, a control synthesis is designed for the visual servo system. To improve the control performance of visual servoing, an interpolation method is applied to determine the corresponding values at the interneuron positions. The advantage of this concept is that the difficulty and complexity of the calculation and calibration are avoided for the visual servoing, because the highly nonlinear relationship between the control torques of the robot and the visual features is determined by training of the extended Kohonen network. The learning ability and adaptability of this concept make the visual servo system more flexible, so that the application of the visual servo system can be expanded for challenging tasks. The simulation results are satisfactory for the proposed control concept