Flexible manufacturing requires the industrial robots be able to accomplish different kinds of tasks in the complex industrial environment autonomously, for instance, to detect and localize the irregular work-pieces of different sizes and random placements, to flexibly manipulate different kinds of work-pieces and to perform the high-precision assembly tasks. The introduction of the visual system can enable the robot to understand the surrounding environment,and it has been applied to sorting and transporting the regular work-pieces. However, due to the environmental noises and the occlusion and projection deformation caused by the stacked work-pieces, it is difficult to localize the work-piece accurately. On the other hand, as a key process in most manufacturing, existing high-precision assembly strategies are based on the force sensors or flexible wrists. However, the force sensor is sensitive to the noise and its purchase and maintenance cost is high; the flexible wrist is not durable and has uncertainties during the operation. These shortcomings make the two methods be not widely used. In addition, existing parallel grippers are designed for some specific tasks, so they can only grasp a very few types of work-pieces. Therefore, more grippers are needed for manipulating different kinds of work-pieces. Meanwhile, the localization error of the work-piece and the incapability of the robot finger to reach the desirable place precisely, leads to the deviation between actual operation and ideal grasping program, which influences the stability of the grasping. Therefore, this thesis focused on the following issues: work-piece pose estimation in complex industrial environment, high-precision robotic assembly using limited information and the stable grasping strategy. The main research works and contributions of this thesis are summarized as follows: Firstly, according to the fact that multiple factors are coupled to affect the work-piece appearance, this paper presented a work-piece pose estimation framework based on the separation of the appearance factors. A Chamfer distance transform is used to detect the work-pieces while a bilinear model is used to separate the pose variables. The proposed method improves the accuracy and speed of the pose estimation. Secondly, according to the environmental noise and the occlusions caused by the stacked placement of the work-pieces, this paper presented a sub-pattern bilinear model which separates the influencing ...
修改评论