Path detection, estimation and correction for robot guidance

Abstract

This thesis proposes new methods for robot path correction. These methods are targeting applications of visual beads on wing parts, such as doors, fender, trunk lids, etc. Since such applications require a high demand of accuracy next to quick cycle times, the methods presented are not only targeting the path correction alone, but also improvements of feature detection, calibration, and sensor images. Therefore, this thesis introduces a method for detecting noise points within the images of laser stripe sensor, by applying statistical methods based on Brownian motion with drift. Furthermore, in order to improve feature detection with an acceptable speed, we present a method that combines an Em-ICP with the Douglas Peuker algorithm to achieve much higher execution speed. Finally, the core of this thesis is an algorithm to determine the 6-dimensional correction of the application path of a robot based on sparse stripe scans. The final topic this thesis addresses is how to make these methods applicable within visual servoing applications. This includes a new approach to motionless calibration for laser stripe sensors and also an adaption of the path correction algorithm in order to provide relative measures.