Manipulation Pipeline for Robotic Tight Packaging Using Variable Stiffness

Abstract

The automation of industrial scenarios by employing robots in logistic tasks such as packaging presents new challenges in robotics research, in terms of development of end-effectors for these applications, and of methodologies for the manipulation and the packaging of the items. The most commonly explored packaging problems are Bin Packing and Pick-and-Place. In these applications, the focus is mainly on checking if the item fits in the bin, and finding a placement for it; the packaging aspect is often as trivial as lowering the object above its intended position and releasing it. This thesis presents a more challenging but also more realistic application where the robot performs tight packaging, meaning that the container is filled with items so that they fit closely together, with minimum clearance between them. This setting provides a contact rich environment, with noise and disturbances coming from various sources of uncertainties, such as the unconstrained items in the container, the object localisation and the robot itself. This work investigates the effect of a reliable placing strategy, and of passive compliance in the system to improve robustness and success rate in such a task. It is achieved by exploiting the characteristics of a mechanically compliant hybrid gripper with variable stiffness mounted on a redundant robotic arm. Additionally, it explores the roles of the hand configuration and the stiffness level in the task execution. A complete pipeline is designed to execute the full operation, from the grasp of the items, through the plan of the packaging, to the placement of the object in the container, until the container is filled. The approach is tested, both in simulation and with the real robot, by evaluating it with a tight packaging and a stacking of building blocks, and with an emulated tight packaging task using cuboid boxes.