Source-linked AI summary
A Simple Gripper Interface for Simulator-Agnostic Cloth Manipulation
Abhilash Nayak, Franco Coltraro, Maria Alberich-Carramiñana, Carme Torras
TL;DR
Cloth manipulation is difficult because deformable textiles are sensitive to grasping, while detailed contact models can be costly and simulator-dependent. This paper introduces a simple kinematic gripper interface based on selected cloth positions and demonstrates its use in simulated and robot-arm folding. The experiments show grasp support and orientation affect folding outcomes, and the real-cloth demonstration reproduces realistic folding behavior with approximate cloth-parameter identification.
Problem
Cloth manipulation is challenging because highly deformable textiles respond strongly to grasping, while realistic contact-rich gripper models can be computationally expensive and simulator-dependent.
Source: §1 · p. 1Method
A kinematic gripper selects discrete cloth positions inside an attached grasping volume, stores them in the gripper frame, progressively squeezes them, and transports them with gripper motion.
Source: §1 · p. 1, §3.2 · p. 3, §5 · p. 7Results
Corner-folding experiments show grasp support and gripper orientation can significantly affect final cloth configuration, and a Franka arm reproduced realistic folding behavior with approximate cloth-parameter identification.
Source: §5 · p. 7Takeaways & Limitations
The results support representing grasp support and gripper orientation explicitly in trajectory-level folding and control applications.
Source: §5 · p. 7The model omits external-geometry contact, pushing, sliding, frictional slip, pressure distribution, finger compliance, and tactile feedback, limiting its suitability for high-precision tasks governed by local contact mechanics.
Source: §5 · p. 7
Abstract
from arXiv · showhide
This paper presents a grasping model for cloth manipulation specifically tailored to ease the deployment of robotic control methods. The model is robust, fast and easy to implement avoiding at the same time contact and friction considerations between the gripper and the cloth in favor of simple positional constraints. The gripper is described by its pose, jaw state, and an attached grasping volume. Two kinds of grasping volumes are considered: an axis-aligned box to simulate a pinch grasping and a square pyramidal volume to simulate point grasping. When the gripper closes, the discrete cloth positions lying inside this volume are selected, stored in the local gripper frame, and then transported with the gripper motion. A simple squeezing step is also included to progressively move the selected cloth positions toward the center of the grasping region, avoiding an instantaneous displacement at closure. The model can be used in any simulator as it only requires access to discrete cloth positions and a mechanism for imposing target positions as constraints. We implement our grasping model in conjunction with a constraint-based inextensible cloth simulator, where grasping is implemented as moving positional equality constraints coupled with stretch, shear, collision, and table contact projection steps. The same gripper trajectory is applied on a robot arm to fold a real piece of cloth, serving as a simple bridge between simulation and physical cloth manipulation and showcasing the realism and practicality of our idealized grasping model.
1 Introduction
The introduction motivates a lightweight gripper interface because cloth manipulation is sensitive to grasping, while detailed contact models can be costly and simulator-dependent. It presents a kinematic model and demonstrates folding experiments in simulation and with a robot arm.
1 Introduction
- Cloth manipulation is difficult because textiles are highly deformable and sensitive to grasp points, support size, and grip orientation, while detailed contact models can be computationally expensive and simulator-dependent.Planning, control, learning, and sim-to-real transfer are all named as challenges.Source: §1 · p. 1
- The proposed kinematic gripper has a pose, jaw state, and attached grasping volume, and stores selected cloth positions locally for progressive squeezing and transport.The model is intended as a lightweight interface for control rather than a replacement for precise contact-rich simulation.Source: §1 · pp. 1–2, §1.1 · p. 2
- Its low-dimensional action permits explicit representation, while uncertain cloth parameters, unmodeled contact, or sim-to-real effects can be assigned to a data-driven residual model.Source: §1 · p. 2
1.1 Contributions
- The gripper can be coupled to constraint-based cloth simulators as moving positional equality constraints.Source: §1.1 · p. 2
- Eight corner-fold cases compare grasp support and gripper orientation, and a simulation-designed trajectory is also executed by a Franka arm.The experiments showcase the importance of grasp choice and the simplicity of transferring the trajectory to a robot.Source: §1.1 · p. 2, §1 · p. 2
2 Related works
Prior work ranges from grasp-centered analyses and point-based folding abstractions to learning and model-based control. This paper develops an explicit, lightweight physics-based gripper interface intended to bridge point-grasp abstractions and detailed contact models.
2 Related works
- Earlier work analyzes how local constraints deform cloth (Borràs et al., 2020) or abstracts folding interaction through selected points or small regions (Li et al., 2015).Source: §2 · p. 2
- Recent cloth-manipulation methods combine learning and model-based control, while physics-ML work includes learned residual correction and physical-property estimation (Liang et al., 2024; Jiang et al., 2025).Caldarelli et al. (2026) combine a physics-based simulator with a learned Koopman representation and model predictive control.Source: §2 · p. 2
- The paper focuses on a whitebox gripper interface with explicit grasp geometry, orientation, and motion, avoiding detailed contact and friction models while distinguishing grasp supports and orientations.The authors position it between idealized point grasps and more detailed gripper-cloth contact models.Source: §2 · p. 2
3 Gripper
The gripper uses pose and jaw state to define a gripper-attached volume that selects and transports cloth nodes. The section adds progressive squeezing, positional equality constraints, and a possible learned-residual extension.
3.1 Gripper Pose and Jaw State
- The gripper is described by pose and jaw state, with pose represented using position and quaternion and jaw state distinguishing open from closed.Source: §3.1 · pp. 2–3
3.2 Grasping region
- A gripper-attached axis-aligned box emulates pinch grasping, while a square pyramid emulates point grasping.Source: §3.2 · p. 3
3.3 Node Selection
- At closure, nodes inside the volume are selected and stored locally; edge- and face-center tests add incident nodes on coarse meshes, and support persists until opening.Source: §3.3 · p. 3, Figure 2 · p. 3
3.4 Squeezing model
- Closure progressively moves grasped nodes toward the jaw-closing center to avoid abrupt displacement, but approximates compression without pressure, friction, or local deformation.Source: §3.4 · pp. 3–4
3.5 Grasping as a Constraint
- Grasped nodes follow gripper-computed targets as moving positional equality constraints usable across solver types; added per-step cost scales as O(|I_g|).The experiments use a constraint-based inextensible cloth simulator.Source: §3.5 · p. 4
3.6 Gray-Box Extension for Hybrid Intelligence
- A gray-box extension could retain the physics prediction and learn residual mismatch or simulator-parameter corrections; current experiments evaluate only the physical model.The extension is intended to account for unmodeled deformation, contact, slip, or sim-to-real effects.Source: §3.6 · pp. 4–5
4 Cloth Folding Experiment
The experiments compare corner-grasp strategies under a shared simulated fold, then apply the gripper-level motion to a robot arm. In simulation, rotating the gripper and using a coherent multi-node support produce the best folds among the tested cases; robot transfer is qualitative under parameter uncertainty.
4.1 Comparison of Different Kinds of Grasping
- All eight simulated cases use the same cloth, simulator parameters, and pulling trajectory; their final normalized RMS errors are measured against C8’s reference fold.The cases vary grasp volume, selected-node arrangement, and whether the gripper rotates.Source: §4.1 · p. 5
- Cases C5–C7 produce the lowest errors, while C8’s four-node box grasp provides a coherent patch that transfers motion more consistently and reduces local twisting.C8 is used as the reference fold because its final configuration most closely matches the intended fold.Source: §4.1 · pp. 5–6
- Gripper-interface computation takes 0.17, 0.20, and 0.22 ms per step for 1, 2, and 4 grasped nodes, respectively, or 0.5–1.5% of simulation time.The measured overhead increases approximately linearly with the number of grasped nodes.Source: §4.1 · p. 6
4.2 Cloth Folding Using a Robot Arm
- The physical experiment uses partial system identification, so it should be interpreted under parameter uncertainty rather than as a fully calibrated simulator.Uncertainty in damping, bending stiffness, and friction affects dissipation, fold curvature, and contact behavior, respectively.Source: §4.2 · p. 7
- The robot experiment shows that the same gripper-level command works in simulation and on a Franka arm, but the reported results are qualitative.The robot end-effector follows a Cartesian trajectory while the jaw state is sent separately.Source: §4.2 · p. 7
5 Conclusions and Future Work
The gripper interface stores selected cloth positions in its local frame, progressively squeezes them, and transports them with gripper motion. Simulated and real folding experiments demonstrate its use while exposing the importance of grasp support and orientation; the model remains limited for tasks requiring local contact mechanics.
5 Conclusions and Future Work
- At jaw closure, selected cloth positions are stored in local coordinates, progressively squeezed, and transported with gripper motion.The gripper is defined by its pose, jaw state, and an attached grasping volume.Source: §5 · p. 7
- The interface requires only discrete cloth positions and a mechanism for imposing target positions or constraints, enabling connection to multiple cloth-simulation approaches.The paper identifies particle-based, FEM-based, projection-based, and constraint-based simulators as compatible approaches.Source: §5 · p. 7
- In corner-folding experiments, grasp support and gripper orientation significantly affect the final cloth configuration and should be represented explicitly.The model was demonstrated with a constraint-based inextensible cloth simulator.Source: §5 · p. 7
- A Franka arm reproduced realistic folding behavior with the idealized model when cloth parameters were approximately identified.Figure 6 shows executions with constant and changing gripper orientation, alongside the final simulated cloth configuration.Source: §5 · p. 7, Figure 6 · p. 8
- Because the gripper is not contact-rich, it cannot represent surface pushing, sliding contact, jaw frictional slip, pressure distribution, finger compliance, or tactile feedback.The authors consider these simplifications acceptable for trajectory-level folding, but not high-precision tasks where local contact mechanics determine success.Source: §5 · p. 7
- Future work includes using the interface for learning and control, and adding contact-rich interaction for tasks requiring high precision.The kinematic model could serve as a fast baseline while detailed contact solvers or tactile feedback are activated when local finger-cloth interaction is necessary.Source: §5 · p. 7