Skip to Main Content
This paper proposes the tree-structured knowledge approach for performing part recognition in controlling MEMS-arrayed manipulation surfaces. In this approach, a new data structure, a tree-structured array, is used to store knowledge about models of the objects at an offline stage and to accumulate and share knowledge among neighboring active cells about shapes of objects which must be reconstructed and differentiated on a MEMS-arrayed surface at the online stage. Comparing this approach with the previous matrix-based approach, which contained redundant information in each cell and communication, and demanded excessively frequent comparison in shape differentiation, the current tree-structured knowledge approach aims to use one model for a shape in database, reducing the memory footprint, and avoiding frequent comparison in the differentiation phase. In this paper, both approaches are analysed and compared. Though the current approach shows better performance in terms of a smaller memory footprint and lower communication cost, it trades off the reduction of memory footprint against the probability of the differentiating.