In modern flexible lean production process, industrial robots usually need to operate irregular, randomly distributed work-pieces of different sizes. Vision systems can improve the active perception ability of industrial robots so as to guide the robots to perform operations more flexible. Recently, vision systems have been widely applied to many industrial processes, for instance, work-piece sorting and flaw detection. However, in some applications, such as grasping and assembly, work-pieces are often required to be identified and localized precisely in complex and changeable environments. Therefore, supported by a grant from National Science Fund for Distinguished Young Scholars entitled “Autonomous, high-precision assembly and grasping robot”, this paper concerns on the following key issues: vision based work-pieces localization as well as its applications in robotic grasping and assembly. Tasks of visual localization, robotic grasping and assembly in complex environments have difficulties from the following aspects: high-precision visual localization makes the searching process in six dimensional space complex; the stacked placement makes the work-pieces occlude each other and causes projection distortion in the images, so that the information from the observations are incomplete; the robotic grasping and assembly process are lack of the abilities of active cognitive and visual learning, thus it does not satisfy the requirements of the flexible manufacturing. Therefore, Evolutionary Algorithm, being an approaching optimization strategy, is introduced to: guarantee the fast convergence in the searching space of the work-piece pose; enable different components and features with different discrimination to self-adaptively adjust online, so as to handle the incomplete information from the observations; provide visual cues for high-precision grasping and assembly process. The main works and contributions of this paper are summarized as follows: By theoretically analyzing the relationship between time complexity of the Evolutionary Algorithm and the population size, a Differential Evolutionary Algorithm with changeable population size is proposed. The proposed algorithm can self-adaptively adjust the population size, generations and control parameters of the Evolutionary Algorithm so as to control the optimization direction and the shrinking of the population size. Therefore, the proposed algorithm keeps the searching space and improves the converg...
修改评论