Restricted Spatial Perception-Based Robotic Unloading System Using Dynamic Grasping Strategy

Unloading boxes from a container is daily operation in tobacco logistics warehouse which requires intensive manual labors. However, there are no existing proposals characteristic of versatile modules targeted for more general unloading scenarios. We present, in this paper, a more universal robotic manipulation system which can efficiently unload cardboard boxes from containers to subsequent storing cells. There are two novel contributions of our system. On one hand, we proposed an efficient restricted spatial perception framework which can precisely perceive, separate, and locate cartons randomly placed inside a container. A rotatable dual-camera setup provides a wide enough view window which assures our system could adapt to various containers of different internal dimensions. On the other hand, we designed a smart gripper with dual functionalities which can grasp a single or multiple cartons based on a novel reasoning and decision-making strategy. We deployed our system in a tobacco dock and carried out unloading tasks of more than 8000 cartons with no-prior unloading pattern, experiments not only indicated that our perception module could handle real-world scene in a precise way, but also showed a good result of grasping efficiency of more than 800 cases per hour.