FolDeX
收藏资源简介:
FolDeX是首个专注于长时域可变形物体操作的实物基准数据集与评估平台,由复旦大学、美的集团人工智能研究中心及卡内基梅隆大学联合创建。数据集拥有超过2000小时的真实机器人操作经验,覆盖20余项任务及10余种机器人本体,以衣物折叠为核心任务。其数据通过真实机器人物理演示收集,并围绕恢复、任务、场景和本体四个维度组织,以探究异构数据的复用效能。该数据集旨在弥补现有基准在长时域可变形操作评估方面的不足,为具身智能模型提供标准化的物理世界测试环境,并促进数据高效复用策略的研究。
FolDeX is the first physical benchmark dataset and evaluation platform focused on long-horizon deformable object manipulation, co-developed by Fudan University, the AI Research Center of Midea Group, and Carnegie Mellon University. The dataset contains over 2000 hours of real-world robotic manipulation demonstrations, covering more than 20 tasks and over 10 robot embodiments, with cloth folding as its core task. Its data is collected via real robotic physical demonstrations and organized along four dimensions: recovery, task, scene, and embodiment, to explore the reuse efficiency of heterogeneous data. This dataset aims to address the shortcomings of existing benchmarks in long-horizon deformable manipulation evaluation, provide standardized physical-world test environments for embodied intelligence models, and promote research on efficient data reuse strategies.





