Open Architecture for Vision-Based Robot Motion Planning and Control

Abstract

This chapter introduces a methodology for the vision-based motion control of robot manipulators. The motion control problem is decomposed into three computational stages: motion planning, trajectory generation and trajectory tracking. While the two latter activities are always executed in real time, motion is planned in traditional robot systems off line, by learning robot points or by using numerical output data from programs that plan minimal paths, avoid obstacles, etc. Guidance vision is introduced as an advanced motion control method, which provides flexibility when integrating industrial robots in computer-controlled manufacturing structures. A dynamic look-and-move system architecture is discussed, as a robot-vision system which is closed at task level. An open architecture is proposed as implementing solution for vision-based scene management and robot guidance, which integrates any types of robot controllers and image processing libraries. The chapter also presents a motion control algorithm for robots which are required to pick objects randomly moving on conveyor belts. The algorithm for visual tracking of conveyor belts for “on–the-fly” object grasping is partitioned in two stages: (i) visual planning of the instantaneous destination of the robot, (ii) dynamic re-planning of the robot’s destination while tracking the object moving on the conveyor belt. The ensemble [conveyor belt \(+\) actuator \(+\) sensor] is configured as a single-axis Cartesian robot, leading to a cooperation problem between robot manipulators subject to multitasking control. Experimental results are finally reported in what concerns the statistics of object locating errors and motion planning errors function of the size of the objects of the belt speed and of the light strobe.