Direct dynamic visual servoing at 1 kHz by using the product as 1.5D encoder

Abstract

This paper focusses on direct dynamic visual servoing at high sampling rates for machines used for the production of products that inherently consist of equal features placed in a repetitive pattern. A mechanical system is controlled on the basis of vision only. In contrast to kinematic visual servoing approaches, we do not use a hierarchical control structure. More specifically, the motor inputs are driven directly by the vision controller without the intervention of low level joint controllers. The product in view consists of a repetitive pattern, which is used as a 1.5D encoder purely on the basis of vision. Using fast image processing and a prediction based on a steady-state Kalman filter a 1 kHz direct visual servoing setup is created capable of using the repetitive pattern as a 1.5D encoder with an accuracy of 2 ¿m. The design is validated on an experimental setup.