基於物件角度分類演算法之機械手臂夾取策略

Abstract

none

In recent years, robotics has been studied a lot, especially for the grasping of robotic arms, because of the impact of industrial automation and industry 4.0 requirements. In most applications, the grasping pose is defined manually. To save labor costs and time, we need to develop reliable grasping strategies for robotic arms automatically. Although some tools had been developed to help produce grasping pose candidates, there is still much room for improvement because sufficient robustness is required to achieve a successful grasping. To address this problem, this paper proposes grasping strategies that produce reliable grasping poses without human labeling. The grasping strategies can be applied to objects of any shape with complicated geometry. We perform all the training and testing processes in a simulation environment, but the simulation results can be transferred to robotic arms in the real-world environment to reproduce the action. We then further evaluate the quality of the grasping strategies produced by a large amount of object grasping attempts in which some disturbances are added. The high success rate of grasping shows its potential applications in industrial production lines, providing a promising way to help robotic arms perform high-quality grasping with picking or similar tasks and verify if the planned trajectory is safe and smooth.