Deformable Object Manipulation

Deformable Object Manipulation studies how robots can perceive and control objects whose shape changes during contact, from cloth folding and cable routing to soft-tissue retraction. Approaches include physics models such as mass-spring systems, Cosserat rods, and material point methods; learned dynamics and world models; and control methods including model-predictive control, reinforcement learning, and imitation learning. Work increasingly combines simulation with real-robot data and uses visual, tactile, or force feedback to handle occlusion, unknown materials, and contact changes. A central challenge is making models both fast and physically reliable: real objects vary across trials, and simulation-to-real transfer, safe interaction, and evaluation beyond task completion remain difficult.