Same Action, Different Outcome: Variability in Dynamic Cloth Manipulation

Mahed Dadgostar; Guillem Alenyà; Júlia Borràs

cs.ROOct 7, 2026

Same Action, Different Outcome: Variability in Dynamic Cloth Manipulation

Authors: Mahed Dadgostar, Guillem Alenyà, Júlia Borràs

Organizations: Institut de Robòtica i Informàtica Industrial, CSIC-UPC, Spain

Abstract

Although cloth is known to exhibit different outcomes under repeated fast dynamic motions, even when the same trajectory is applied, this variability has not yet been systematically characterized. Quantifying it is essential to assess the reliability of learned manipulation policies and the extent to which simulation can reproduce real-world behavior. To study this, we execute the same trajectory ten times across four dynamic tasks, two of which are novel, each tested with three cloths of very different properties and at up to three execution speeds, with a total of 269 recorded rollouts. For all of them, we record small marker positions on the cloth and synchronized stereo camera. We then formalize different metrics to quantify variability, and our results show how it is significant in every test condition, in most cases one to three orders of magnitude above the repeatability inherent to the robot and sensing noise. Our results also show variability is driven mainly by the cloth physical properties but also grows with speed. By replaying all the rollouts in four calibrated modern cloth simulators, we show that none of them can reproduce the variability magnitude we observed in real cloth, nor its ordering. Together with our analysis, we publish the dataset with synchronized OptiTrack, vision and robot logs and their corresponding simulator twins.

Figures & tables

Fig. 1 : Terminal cloth outlines for ten identical policy rollouts, shown on one frame, for 2 tasks: Swing to place (right) and gripper inward motion (left).

Fig. 2 : (Left) Bimanual FR3 cell with OptiTrack cameras to track retroreflective markers on the cloth while the two arms execute the grasp poses. (Right) Marker distribution.

	brown wool	green satin	red chiffon
mass ( $\mathrm{g}$ )	139	35	27
areal density ( $\mathrm{g}\text{\,}{\mathrm{m}}^{-2}$ )	400	100	76
friction coefficient $\mu$	0.68	0.48	0.40
elasticity † ( $\mathrm{\%}$ )	8 / 1 / 17	1 / 12 / 18	20 / 38 / 13
drape stiffness ( $\mathrm{\%}$ )	9.5	5.9	3.1

TABLE I : Fabric properties following [ 25 ]

Fig. 3 : Each row shows 5 frames of the recorded tasks. The markers used to study the variability for each task are colored and tracked in the camera view. Other relevant markers are in gray.

#	Task	Object †	Timescale $\mathrm{ts}$	Regrasp
1	Swing place	1 cloth (b)	0.8, 1, 1.2	No
		2 cloths (g, r)	0.65, 1, 1.35	No
2	Fold	3 cloths	1, 1.4	No
3	Linear join	3 cloths	1	No
		1 cloth (g)	1	Yes
4	Swing join	3 cloths	1, 1.5	No

TABLE II : The four tasks and their 27 configurations.

Fig. 4 : We show the variability amount for all tasks and their configurations with the whisker boxes of all the pair distances

D_{c}(i,j)

in all rollouts per configuration. The variability is orders of magnitude different depending on the task; that is why the y-axis is on a logarithmic scale.

Fig. 5 : Resting poses of the markers of the bottom edge for the swing place task (Task 1). It shows increasing variability for lighter cloths as well as with higher speeds.

Fig. 6 : Resting poses of the free marker m5, with respect to the target pose for the task of dynamic folding (task 2). Variability of the lighter cloths are significantly bigger than for the thick brown cloth.

Fig. 7 : For tasks 3 and 4, this graphic shows the amount of buckling (distance to line m5–m15 held by the grippers) described in Eq. ( 7 ) for the visible top cloth edge markers. The numbers at the top and bottom count the buckling side, using Eq. ( 8 ).

Fig. 8 : Comparison of the variability amount in each simulator compared to the OptiTrack ground truth. Points closer to the diagonal correspond to more accurate simulations. None of the simulators captures the real variability.

Fig. 9 : Variability using OptiTrack vs. using the markerless vision estimation. Although the trends are similar, vision does not capture the same variability values, nor the ordering.

Explore similar work

Enabling Robust Cloth Manipulation via Inference-Time Simulator-in-the-Loop Refinement
Jun 23, 2026Xin Liu, Yulin Li, Ziming Li +7Robotics Simulation Deformable Object Manipulation
National University of Singapore · Shanghai Jiao Tong University
A Simple Gripper Interface for Simulator-Agnostic Cloth Manipulation
Sep 24, 2026Abhilash Nayak, Franco Coltraro, Maria Alberich-Carramiñana +1Robotic Grasping Robotics Simulation
Institut de Robòtica i Informàtica Industrial, CSIC-UPC, Barcelona, Spain · Departament de Matemàtiques, Universitat Politècnica de Catalunya, Barcelona, Spain
FoldNet++: a Large-Scale Synthetic Dataset for Robotic T-Shirt Folding and Unfolding
Sep 14, 2026Yuxing Chen, Zhiyuan Wei, Bowen Xiao +2Generalization in Robotic Manipulation Synthetic Data Generation for Robotics
CFCS, Peking University, China · Galbot, China

Jun 23, 2026, cs.ROJ/K move · Enter open · S save

Enabling Robust Cloth Manipulation via Inference-Time Simulator-in-the-Loop Refinement

Xin Liu, Yulin Li, Ziming Li, Pengyu Jing+6

National University of Singapore · Shanghai Jiao Tong University

Robotics Simulation Deformable Object Manipulation Robotics

Open paper AskPDF arXiv

Simulator-in-the-loop optimization offers a promising inference-time mechanism for robot manipulation. It uses a physical simulator as a backend rollout engine to evaluate candidate trajectories in parallel and refine nominal actions online, a paradigm shown to be effective in rigid-body manipulation where state and contact are relatively tractable. We bring this paradigm to real-world cloth manipulation from a single RGB input through three pillars. (i) We design a scalable synthetic-data generation and inference-time rollout pipeline built on FLASH, a deformable-object simulator that provides a practical balance among physical fidelity, numerical stability, and rollout efficiency. (ii) We develop a real-to-sim module, trained purely on synthetic data, that maps a single RGB observation to simulation-compatible cloth state by fusing pretrained visual features with learnable canonical tokens. (iii) We perform online planning by coupling a sparse-mesh rollout backend with prior-guided MPPI, anchored at an offline-distilled policy trajectory, preserving manipulation-relevant deformation and contact while enabling sufficient parallel rollout batches. Real-robot experiments show higher success rates than baseline methods and closed-loop correction under mid-fold perturbations. Project page: https://silr-cloth.github.io/

Abstract

Explore similar work

Enabling Robust Cloth Manipulation via Inference-Time Simulator-in-the-Loop Refinement

A Simple Gripper Interface for Simulator-Agnostic Cloth Manipulation

FoldNet++: a Large-Scale Synthetic Dataset for Robotic T-Shirt Folding and Unfolding