Parts picking in disordered environment

In some tasks such as bin-picking, robots are required to pick up one part from among a number of piled-up scattered parts. To perform such tasks, the recognition of parts using a vision system is indispensable. The paper proposes approaches to reduce the processing load on vision systems and to detect accidents using a force/torque sensor in real time. For the former approach, the concepts of parts isolation factor and robot manipulability at the grasping position are introduced to formulate a parts-picking strategy for executing the task while maintaining the initial environment. The picking operation is executed in order of highest isolation factor and manipulability. An accident recovery method in which force and torque are monitored during operation is also presented