T-DOM: A Taxonomy for Robotic Manipulation of Deformable Objects

TL;DR

This work introduces T-DOM, a taxonomy that extends robotic manipulation understanding from rigid to deformable objects by jointly categorising object deformation, robot motion, and interaction types. It defines deformation using compression, tension, bending (structured and unstructured), torsion, and shear, augmented by a structured framework for bending states and a distinction between prehensile and non-prehensile interactions, including a dedicated contact sliding category. The authors validate the taxonomy on a 10-task deformable-object dataset featuring garments, meat-like phantoms, bags, and surgical items, and demonstrate that deformation-aware classifications better differentiate manipulation skills than prior taxonomies. They also discuss applications to gripper design, skill learning, and general manipulation policies, outlining future directions such as extending deformation types and improving deformation measurement for practical robotic deployment.

Abstract

Robotic grasp and manipulation taxonomies, inspired by observing human manipulation strategies, can provide key guidance for tasks ranging from robotic gripper design to the development of manipulation algorithms. The existing grasp and manipulation taxonomies, however, often assume object rigidity, which limits their ability to reason about the complex interactions in the robotic manipulation of deformable objects. Hence, to assist in tasks involving deformable objects, taxonomies need to capture more comprehensively the interactions inherent in deformable object manipulation. To this end, we introduce T-DOM, a taxonomy that analyses key aspects involved in the manipulation of deformable objects, such as robot motion, forces, prehensile and non-prehensile interactions and, for the first time, a detailed classification of object deformations. To evaluate T-DOM, we curate a dataset of ten tasks involving a variety of deformable objects, such as garments, ropes, and surgical gloves, as well as diverse types of deformations. We analyse the proposed tasks comparing the T-DOM taxonomy with previous well established manipulation taxonomies. Our analysis demonstrates that T-DOM can effectively distinguish between manipulation skills that were not identified in other taxonomies, across different deformable objects and manipulation actions, offering new categories to characterize a skill. The proposed taxonomy significantly extends past work, providing a more fine-grained classification that can be used to describe the robotic manipulation of deformable objects. This work establishes a foundation for advancing deformable object manipulation, bridging theoretical understanding and practical implementation in robotic systems.

Figures (5)

• Figure 1: Robotic manipulation of multiple deformable objects showing compression, tension, bending, torsion and shear deformation. Shear deformation examples include a cloth shear deformation from the open-source dataset provided by wang_2011_arcsim_cloth, and a sponge manipulated by a human.
• Figure 2: Qualitative classification levels for bending deformation as structured or unstructured for 1D and 2D deformable objects. The structured level is classified by loops and g-foldsvandenBerg2011_gfold, for 1D and 2D objects, respectively. The unstructured level is classified by knots for 1D objects, and as the number of accessible corners for a cloth flattening task.
• Figure 3: The proposed taxonomy is composed of deformation (D), motion (M), and interactions. The interactions are classified as prehensile grasp (G) and non-prehensile interactions (NP), and contact sliding (CS) as a special case of interaction that can take place in both prehensile grasps as well as each sub-category of non-prehensile interactions, shown by a dashed line connecting the CS block. Each sub-category is shown with its associated short tag.
• Figure 4: Analysis of the transitions of Task 1, Task 2, and Task 4 using the proposed taxonomy, taxonomy, as well as paulius_2020_manipulation_tax_embedding and bullock2012hand taxonomies, showing only the categories that show any change. For all tasks the structural outcome according to paulius_2020_manipulation_tax_embedding taxonomy is deforming temporarily, and therefore it is omitted for simplicity.
• Figure 5: Graph clustering of manipulation actions in the proposed dataset for a) taxonomy, b) taxonomy without deformation, c) paulius_2020_manipulation_tax_embedding taxonomy, and d) bullock2012hand taxonomy. The nodes represent the action-IDs and the edges action-IDs which share the same tag according to the evaluated taxonomy. The colours of the nodes represent the deformation of the object. The edges and clusters colours indicate node connectivity. The markers represent no bending (circle), structured bending (triangle up), unstructured bending (triangle down), and both structured and unstructured bending (square). Note that all graphs are complete graphs but only one edge is shown for clarity.