一种基于流形的机械臂动作构型知识压缩表达方法

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

针对在机械臂分拣任务中,存在物体形状各异、大小不一、训练神经网络成本过高的问题,提出一种基于流形空间的机械臂快速分拣方法。通过自主设计的一款简易实验装置模拟代替机械臂进行实验。对高维数据进行压缩,结合三维快速凸包求解算法,对体积大小不同的同类物体的流形空间进行分割,以凸包形式将稳定性较高的点集包裹起来。实验结果表明,体积大小不同的同类物体的高稳定流形子空间是一致的。该方法可以通过对一种物体的流形子结构进行尺度放缩,得到不同大小的同类物体的高稳定分拣区域,用于生成高效、可靠的机械臂分拣任务中的6D位姿构型,以提高分拣作业的工作效率。

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.