Research Analyzer
← Back ICRA 2026

Right-Side-Out: Learning Zero-Shot Sim-To-Real Garment Reversal

Chang Yu, Siyu Ma, Wenxin Du, Zeshun Zong, Han Xue, Wendi Chen, Cewu Lu, Yin Yang, Xuchen Han, Joseph Masterjohn, Alejandro Castro, Chenfanfu Jiang

PDF

AI summary

Key figure (auto-extracted from paper)
A zero-shot sim-to-real framework decomposes garment flipping into keypoint-conditioned primitives and leverages a high-fidelity GPU-parallel MPM simulator to achieve up to 81.3% success without real-world fine-tuning.
Garment manipulation Sim-to-real transfer Zero-shot learning Material Point Method Task decomposition Bimanual robotics

Problem

Flipping garments right-side out is a highly dynamic, contact-rich, and severely occluded manipulation task that remains largely unexplored in robotics due to the prohibitive cost of collecting teleoperation data and the complexity of modeling rapid contact changes.

Approach

The task is broken into three bimanual primitives (DRAG, FLING, INSERT&PULL) parameterized by keypoints, with policies trained entirely in a custom GPU-parallel MPM simulator using automated domain randomization and deployed directly on real hardware.

Key results

  • Introduced robotic garment reversal and a decomposition into DRAG, FLING, and INSERT&PULL primitives
  • Developed a high-fidelity, GPU-parallel MPM simulator with a fully automated data generation pipeline
  • Achieved up to 81.3% zero-shot sim-to-real success rate on diverse garments using only a depth camera
  • Enabled scalable, annotation-free data generation through geometric and material randomization

Why it matters

It provides a scalable, demonstration-free pathway for robots to perform fundamental household garment tasks, advancing practical sim-to-real transfer for complex deformable object manipulation.

Abstract

Turning garments right-side out is a challenging manipulation task: it is highly dynamic, entails rapid contact changes, and is subject to severe visual occlusion. We introduce Right-Side-Out, a zero-shot sim-to-real framework that effec- tively solves this challenge by exploiting task structures. We decompose the task into DRAG/FLING to create and stabilize an access opening, followed by INSERT&PULL to invert the gar- ment. Each step uses a depth-inferred, keypoint-parameterized bimanual primitive that sharply reduces the action space while preserving robustness. Efficient data generation is enabled by our custom-built, high-fidelity, GPU-parallel Material Point Method (MPM) simulator that models thin-shell deformation and provides robust and efficient contact handling for batched rollouts. Built on the simulator, our fully automated pipeline scales data generation by randomizing garment geometry, material parameters, and viewpoints, producing depth, masks, and per-primitive keypoint labels without any human anno- tations. With a single depth camera, policies trained entirely in simulation deploy zero-shot on real hardware, achieving up to 81.3% success rate. By employing task decomposition and high fidelity simulation, our framework enables tackling highly dynamic, severely occluded tasks without laborious human demonstrations. More details and supplementary material are on the website: https://right-side-out.github.io.

Index terms

Bimanual Manipulation Deep Learning in Grasping and Manipulation Simulation and Animation

Related papers