Deformable Object Manipulation

Grasping and handling objects that bend, stretch, or change shape under force, such as clothes, cables, or food.

Deformable object manipulation covers objects whose shape changes noticeably under force: cloth and clothing, ropes and cables, bags, dough, food, even human tissue. Traditional grasping research treats objects as rigid bodies, needing only position and orientation — 6 degrees of freedom — to describe. Deformable objects instead have an enormous number of shape degrees of freedom, can fold and self-occlude, and deform nonlinearly under force, which makes them hard to model and simulate. A 2021 survey by Jihong Zhu and colleagues identifies three main technical difficulties — deformation is hard to perceive, the degrees of freedom are high, and deformation is nonlinear to model — and, based on a survey of peer researchers, concludes that perception is the area most worth investing in. Applications include folding laundry, industrial cable-harness assembly, harvesting fruit and vegetables, surgical suturing, and dressing assistance for the elderly.

ExampleFolding a shirt: the robot must first flatten a wrinkled T-shirt, then fold it along creases, and the shirt's shape changes after every step, so there is no fixed grasp point that can be marked in advance.

Also called

DOM

Related

Garment Manipulation · Deformable-Body Simulation · Cloth Simulation · Contact-rich Manipulation · Tactile Sensor · SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation

Sources

Challenges and Outlook in Robotic Manipulation of Deformable Objects (arXiv 2105.01767)