Sign inGet started
Dataset · Hugging Face

tklee-3012/cross-embodiment

ManiSkill Cross-Embodiment 3×5 Demonstrations for a cross-embodiment transfer benchmark built on ManiSkill 3: three robot arms × five table-top tasks, with one (robot, task) cell per robot held out (no data). A policy trained on the 12 training cells is tested on the 3 held-out cells, where the task was only ever demonstrated by the other robots. PickCube PushCube PullCube StackCube PlaceSphere Panda train train train held-out train XArm6-Robotiq train held-out… See the full description on the dataset page: https://huggingface.co/datasets/tklee-3012/cross-embodiment.

Author / organizationtklee-3012
Downloads0 (as reported by Hugging Face)
Likes1
Created25 October 2026
Last updated at source25 October 2026
maniskillcross-embodimentimitation-learningvision-language-action
Embed this entry
<iframe src="https://silytics.si/embed/item/08df9c31-6592-4e5e-b20a-23e92d805a51" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Recent changes

  • 25 Oct 2026 Description: none → ManiSkill Cross-Embodiment 3×5 Demonstrations for a cross-embodiment transfer benchmark built on ManiSkill 3: three robot arms × five table-top tasks, with one (robot, task) cell per robot held out (no data). A policy trained on the 12 trai
  • 25 Oct 2026 Topics added task_categories:robotics, maniskill, cross-embodiment, imitation-learning, vision-language-action

Changes detected by Silytics between refreshes of the source.

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 25 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.