Spatial Vision-Based Control of High-Speed Robot Arms

Industrial robots are known to execute given programs at high speed and at the same time with high repeatability. From non industrial scenarios as in (Nakabo et al., 2005) or (Ginhoux et al., 2004) we know that cameras can be used to execute high-speed motion even in those cases in which the desired path is not a priori given but online sensed. In general, such visual tracking tasks of following a randomly moving target by a camera are not accurate enough for industrial applications. But there the work-piece will scarcely move randomly. So in this chapter we concentrate on another type of visual ser-voing, in which the path that has to be followed is fixed in advance, but not given by a robot program. A robot-mounted camera is a universal sensor which is able to sense poses with high accuracy of typically less than a millimeter in the close-up range. At the same time with high shutter speed the robot may move fast, e.g. at 1 m/s. In contrast to expensive high-speed cameras, yielding a high frame rate of e.g. 1 kHz as in (Nakabo et al., 2005), we restrict on a standard CCIR camera, to meet the requirements of a cost-effective hardware. Off-the-shelf cameras are fundamental for the acceptance in industry. So an important feature of our method is an appropriate control architecture that tolerates low sampling rates of sensors. The camera is mounted at the robot arm and measures the work-piece pose (given by boundary lines) with respect to the tool center point (tcp) of the robot. More precisely, the camera is mounted laterally to provide enough space for a tool. So with constant moving sense we have a predictive sensor as in (Meta-Scout, 2006). With alternating moving sense the camera has to be tilted so that the corresponding nominal line point