A Compacted Representation Method of Action Configuration Knowledge for a Robotic Arm Based on Manifold

Aiming at the problems of different shape/size of objects with high cost on robotic arm sorting task, a manifold-based fast sorting method for a robotic arm was proposed. A simple experimental device was designed and used to simulate the experiment instead of the robotic arm. The manifold space of similar objects with different size was segmented and the stable point set was wrapped in the form of a convex hull by compressing the high dimensional data and combining it with the fast convex hull solving algorithm. The experimental results showed that the highly stable manifold subspaces of similar objects of different size were consistent. By scaling down the manifold substructure of an object, the proposed method could obtain a highly stable sorting region for similar objects of different sizes. It could be used to generate an efficient and reliable robotic arm 6D pose configuration for sorting tasks, and to effectively improve the working efficiency of the sorting operation.