Target Shape 1
Deformable linear object manipulation, particularly rope manipulation, is a vital yet challenging task in robotics due to the intricate dynamics, high degrees of freedom (DoF), and underactuated characteristics of soft objects. This project aims to develop effective strategies for controlling robotic arms with grippers to shape a rope into specified target poses. Our approach involves familiarizing ourselves with classical algorithms, evaluating existing methods in the field, and exploring potential improvements. The ultimate objective is to design a policy that not only accomplishes the rope manipulation tasks but also generalizes to new, unseen target poses, thereby enhancing the adaptability and versatility of robotic systems in handling deformable linear objects.
Generalize from the small training set to the level 1 test set.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Generalize from the small training set to the level 2 test set.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Generalize from the Large training set to the level 1 test set with task curriculum.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Generalize from the Large training set to the level 1 test set.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Generalize from the Large training set to the level 2 test set with task curriculum.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Generalize from the Large training set to the level 2 test set.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4
Texture Shift on twist and bend of the Rope.
Twist=2e4, Bend=2e5; Normal.
Twist=2e4, Bend=4e4; Deviant.
Twist=2e4, Bend=4e5; Tight.
Twist=5e4, Bend=4e4; Slipped.
Performance on 3D environment for testing our method.
Target Shape 1
Target Shape 2
Target Shape 3
Target Shape 4