MonoFold: Monocular Sim-to-Real Manipulation of Deformable Sheets With Compact Geometric Representations
IEEE Robotics and Automation Letters · 2026-12 · 11 (12) · 13869-13876
DOI: 10.1109/lra.2026.3746129 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Robotics and Automation Letters, 2026-12. Verfasst von Hamidreza Raei und 3 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: European Union's Horizon Europe Project TORNADO, Marie Skłodowska-Curie Actions.
Abstract (Originalsprache)
Deformable sheet manipulation is challenging due to high-dimensional dynamics, self-occlusion, partial observability, and contact-rich interaction. This work presents MonoFold, a monocular sim-to-real framework for fixed single-grasp manipulation of sheet-like objects using a single-arm robot with an eye-in-hand RGB camera. We study three corner-defined placement primitives: flat placement, diagonal folding, and horizontal folding. MonoFold uses a compact 9-D corner-relative representation that captures task-relevant sheet geometry without dense mesh, point-cloud, particle-level, or image-sequence reconstruction. The policy is trained in simulation with privileged corner states through a multi-fidelity reinforcement-learning pipeline, and deployed by mapping monocular RGB observations, via segmentation and depth-based regression, to the same 9-D state. This shared geometric interface aligns simulation training, real-world perception, and closed-loop control. Simulation results show that primitive-specialized policies improve folding over a shared MLP, while hardware experiments demonstrate reliable flat placement and promising folding across evaluated sheet materials and grasp locations. These results suggest that task-aligned compact representations can support monocular sim-to-real manipulation for corner-defined sheet placement and folding.
Förderung
- European Union's Horizon Europe Project TORNADO
- Marie Skłodowska-Curie Actions
Zitieren
APA
Raei, H., Zhao, J., Momi, E. D., & Ajoudani, A. (2026). MonoFold: Monocular Sim-to-Real Manipulation of Deformable Sheets With Compact Geometric Representations. IEEE Robotics and Automation Letters, 11(12), 13869-13876. https://doi.org/10.1109/lra.2026.3746129
Vancouver
Raei H, Zhao J, Momi ED, Ajoudani A. MonoFold: Monocular Sim-to-Real Manipulation of Deformable Sheets With Compact Geometric Representations. IEEE Robotics and Automation Letters. 2026;11(12):13869-13876. doi:10.1109/lra.2026.3746129
BibTeX
@article{Raei2026,
author = {Raei, Hamidreza and Zhao, Jianzhuang and Momi, Elena De and Ajoudani, Arash},
title = {MonoFold: Monocular Sim-to-Real Manipulation of Deformable Sheets With Compact Geometric Representations},
journal = {IEEE Robotics and Automation Letters},
year = {2026},
volume = {11},
number = {12},
pages = {13869--13876},
doi = {10.1109/lra.2026.3746129}
}
EUVIMED