Dexterous in-hand manipulation refers to that the relative configuration between the object and the end-effector changes over time. Despite some significant advancements in this field, it remains a relatively understudied topic, especially for bin picking of thin objects. In this work, we propose three novel dexterous manipulation techniques for grasping thin objects: scooping, dig-grasping, and tilt-and-pivot manipulation.

Scooping manipulation is executed via motion control with a minimalist rigid hardware design: a two-fingered parallel-jaw gripper with a fixed-length finger and a variable-length thumb; underactuated control or a compliant mechanism is not required. The manipulation relies on mobility analysis and requires the coordination of the entire manipulation system: the base...[ Read more ]

Dexterous in-hand manipulation refers to that the relative configuration between the object and the end-effector changes over time. Despite some significant advancements in this field, it remains a relatively understudied topic, especially for bin picking of thin objects. In this work, we propose three novel dexterous manipulation techniques for grasping thin objects: scooping, dig-grasping, and tilt-and-pivot manipulation.

Scooping manipulation is executed via motion control with a minimalist rigid hardware design: a two-fingered parallel-jaw gripper with a fixed-length finger and a variable-length thumb; underactuated control or a compliant mechanism is not required. The manipulation relies on mobility analysis and requires the coordination of the entire manipulation system: the base gripper, the variable-length thumb, and a six-DOF manipulator. It suits both object picking from a flat surface and bin picking from a clutter. To circumvent the limitations of the model-based method such as the requirement for instance segmentation and pose estimation, we devise an instance-agnostic learning framework to directly predict the optimal pre-scoop gripper configuration with the RGB-D information of the bin scenario. The framework is hierarchical and triple-tiered, and is evaluated on heterogeneous clusters of both seen and unseen objects.

As a novel technique complementary to scooping with regard to complete bin picking tasks, dig-grasping only suits malleable cluster, but takes less executing time. The manipulation is realized through a two-fingered gripper with asymmetric finger lengths.

For picking rigid polygonal objects with large width, and those with width greater than the maximum gripper opening in particular, tilt-and-pivot manipulation was proved to be effective in achieving this. It requires the coordination of two manipulators: one is attached with a parallel-jaw gripper and the other with a fixture. We discuss the kinematics and planning of tilt-and-pivot, end-effector shape design, and the overall practicality of the manipulation technique.