FolDeX: A Physical-World Benchmark for Long-Horizon Robotic Manipulation of Deformable Objects

Chenhuan Liu, Yi Xu, Feng Wu, Hanyang Wang, Wenxiao Kuai, Weihao Ding, Shan Wang, Yang Liu, Shuyong Gao, Wenqiang Zhang · N/A · 2026

Summary

Embodied AI, including vision-language-action and world-action models, must operate reliably in the physical world.

Abstract Summary

Embodied AI, including vision-language-action and world-action models, must operate reliably in the physical world. Yet methods that perform well in simulation can degrade substantially on real robots, especially in long-horizon deformable-object manipulation, where policies must track changing states and execute reliable multi-stage bimanual interactions. Existing real-robot benchmarks mainly focus on short-horizon rigid-object tasks and offer limited coverage of long-horizon deformable manipulation. We introduce FolDeX, a physical-world benchmark built entirely from real-robot data, with garment folding as its primary task. Since real-robot data collection is costly, FolDeX studies how heterogeneous physical experience can be reused efficiently. The benchmark is organized around four research axes: leveraging human intervention and recovery data collected during deployment; transferring data across tasks, including across garment categories and from rigid to deformable-object manipulation; reusing data across scenes with changes in lighting, background, and layout; and transferring data across robotic embodiments. FolDeX provides 2,000+ hours of real-robot data spanning 20+ tasks and 10+ embodiments. We also establish a fair real-robot evaluation platform for externally submitted policies, with standardized tasks, held-out physical objects, controlled initializations, and a unified execution protocol. The platform is publicly accessible at https://ai.midea.com/#/fold-challenge. We hope FolDeX will serve as a unified testbed for heterogeneous real-robot data reuse and reliable long-horizon deformable manipulation.

Key Points

Embodied AI, including vision-language-action and world-action models, must operate reliably in the physical world.
Yet methods that perform well in simulation can degrade substantially on real robots, especially in long-horizon deformable-object manipulation, where policies must track changing states and execute reliable multi-stage bimanual interactions.
Existing real-robot benchmarks mainly focus on short-horizon rigid-object tasks and offer limited coverage of long-horizon deformable manipulation.
We introduce FolDeX, a physical-world benchmark built entirely from real-robot data, with garment folding as its primary task.
Since real-robot data collection is costly, FolDeX studies how heterogeneous physical experience can be reused efficiently.

FolDeX

FolDeX is a physical-world benchmark and dataset built entirely from real-robot data for long-horizon deformable manipulation. It covers continuous deformation, multi-stage state transitions, contact-rich bimanual coordination, partial observability, and errors accumulating over long episodes.

Garment folding is the central task because it requires reasoning about garment structure and continuously changing physical states.

FolDeX — FolDeX: A Physical-World Benchmark for Long-Horizon Robotic Manipulation of Deformable Objects — Figure 2: Overview of FolDeX. Rows show, from top to bottom, the major stages of a long-horizon folding task, garment categories, other deformable- and rigid-object manipulation tasks, and the robotic embodiments and scenes.

Experiments

We conduct preliminary real-robot experiments corresponding to the four research questions defined by FolDeX: recovery-data utilization, cross-task transfer, cross-scene transfer, and cross-embodiment transfer. These experiments provide reference results and comparisons for future research.

We first establish multi-task reference baselines and compare unified multi-task training with task-specific training under standard and RTC-based training configurations.

Experiments — FolDeX: A Physical-World Benchmark for Long-Horizon Robotic Manipulation of Deformable Objects — Figure 3: FolDeX embodiments and data composition. The benchmark contains 2,000+ hours of real-robot experience spanning 20+ tasks and 10+ robotic embodiments. The statistics summarize the distribution of garment-manipulation tasks, other deformable-object tasks, rigid-object tasks, and robotic embodiments.

Sources and Demonstrations

Figures are reproduced from Liu et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.