Robotic Manipulation of Deformable Objects

Robotic manipulation of deformable objects addresses the complexities of interacting with items that change shape under force, such as cables, textiles, biological tissues and foodstuffs. Unlike rigid-body manipulation, deformable-object handling demands continuous estimation of shape, material properties and contact forces, often in high-dimensional state spaces. Research spans four core areas: perception, which involves sensing and reconstructing object geometry; modelling, where analytical, data-driven or hybrid representations predict deformation; state estimation, which fuses sensory input and model predictions; and planning and control, which generate trajectories or force profiles to achieve desired configurations. Advances in computer vision and machine learning have enabled real-time instance segmentation and latent-space dynamics modelling, while physics-based simulations and finite-element methods support model calibration and optimisation. Practical applications range from wire routing in manufacturing, cloth folding in logistics and robot-assisted dressing and surgery in healthcare, to precision harvesting in agriculture. Cross-disciplinary efforts leverage robotics, materials science and control theory to develop adaptive controllers and simulation-to-real transfer strategies. As robots increasingly work alongside humans and in unstructured environments, robust manipulation of deformable objects is crucial for automation in small-batch production, personalised assistance and safe human-robot interaction.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A comprehensive tutorial and survey has synthesised foundational models of shape and deformation dynamics, and reviewed techniques for learning object properties, state estimation and motion planning for desired deformations. This work highlights the value of combining analytical mechanics with data-driven identification to handle a broad class of non-rigid objects, from sheets to soft tissues.

A novel fast instance-segmentation method isolates deformable linear objects in real time by combining deep convolutional networks with a skeletonisation post-processing step. The approach constructs a graph representation of centre-line nodes and applies topological reasoning to separate intertwined elements, yielding robust 2D coordinate sequences suitable for spline-based manipulation and supporting inference rates above 20 frames per second.

An online model-based framework integrates a neural network conditioned on object parameters with gradient-based optimisation to perform shape control of linear flexible objects. Simultaneous estimation of model parameters and control inputs enables adaptation to unknown material properties and geometric variations. Experiments demonstrate efficient real-world manipulation of cables and ropes on diverse surfaces, validating the method’s accuracy and resilience compared to existing baselines.

Robotic Manipulation of Deformable Objects publication trend

The graph below shows the total number of articles in robotic manipulation of deformable objects across all publications each year (not limited to Nature Index journals).

Technical terms

Deformable Linear Object (DLO): A one-dimensional flexible item, such as a cable or rope, whose shape changes under external forces.

Instance Segmentation: The computer-vision task of detecting and delineating each object instance in an image.

Skeletonization: A process that reduces a shape to its medial axis or centre-line for simplified representation.

Gradient-based Optimisation: A numerical method that iteratively adjusts parameters by following gradients of a cost function.

Latent Space: A lower-dimensional representation learned by a neural network that captures essential object state features.

References

1. Modeling of Deformable Objects for Robotic Manipulation: A Tutorial and Review. Frontiers in Robotics and AI (2020).
2. FASTDLO: Fast Deformable Linear Objects Instance Segmentation. IEEE Robotics and Automation Letters (2022).
3. Deformable Linear Objects Manipulation With Online Model Parameters Estimation. IEEE Robotics and Automation Letters (2024).