SEED-UMI

KO | EN

gitlite — search

#shell #postgresql #graphql #reactjs #cli #framework #computer-vision #deep-learning #mcp #electron

(★ 9)

[CoRL26] SEED-UMI: Sharing the Exoskeleton between human and robot for onE-to-one Dexterous demonstration

파일 탐색기

ZIP 다운로드

- arxiv-badge.svg
- figure2_shared_exoskeleton.png
- main.pdf
- paper-badge.svg
- project-page-badge.svg
- - figure.svg
  - figure1-teaser.png
  - figure2-hardware.png
  - figure3-pipeline.png
  - figure4-design.png
  - figure5-rollouts.png
  - figure6-comparison.png
  - figure7-efficiency.png
  - hero-shared-exoskeleton.png
  - SEED-UMI.pdf
- .nojekyll
- favicon.svg
- index.html
- - globals.css
  - layout.tsx
  - page.tsx
- - project-page.tsx
  - research-content.tsx
- - use-mobile.ts
- - site.ts
  - utils.ts
- - .nojekyll
  - favicon.svg
- - export-pages.mjs
- .gitignore
- .oxfmtrc.json
- .oxlintrc.json
- components.json
- next.config.ts
- package-lock.json
- package.json
- README.md
- tsconfig.json
- vite.config.ts
.gitignore
README.md

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

Tengbo Yu*,1,2 &nbsp; Jiahao Wu*,2 &nbsp; Daohan Li*,2 &nbsp; Bingxu Chen2 &nbsp; Hao Liu2 &nbsp; Xiaojian Ma2 &nbsp; Hangxin Liu1,†

1 State Key Laboratory of General Artificial Intelligence, 
 School of Intelligence Science and Technology, Peking University 
 2 Delta Intelligence 
 * Equal contribution &nbsp; † Corresponding author


 <a href="https://tengbo-yu.github.io/SEED-UMI/">
 <img src="assets/project-page-badge.svg" alt="Project Page: SEED-UMI" height="28">
 </a>
 &nbsp;
 <a href="assets/main.pdf">
 <img src="assets/paper-badge.svg" alt="Paper: PDF" height="28">
 </a>
 &nbsp;
 <a href="https://arxiv.org/abs/2609.11753">
 <img src="assets/arxiv-badge.svg" alt="arXiv: Preprint" title="Read the arXiv preprint" height="28">
 </a>

## Overview

**SEED-UMI** is a demonstration-collection and policy-learning framework for dexterous manipulation in which **both the human and the robot wear the same exoskeleton**. The shared structure provides corresponding joint measurements and matched wrist-camera observations, connecting natural human demonstrations to robot execution.

By recording encoder trajectories on both embodiments, SEED-UMI turns motion retargeting into paired cross-embodiment supervision. Policies learn directly from human-collected demonstrations and raw exoskeleton-centric wrist images.

## Key Features

- **Shared hardware and sensing.** A 20-DoF exoskeleton preserves the outer contact geometry, joint correspondence, and camera placement across the human and robot sides, with cuffs and mounts adapted to each embodiment.
- **Contact-aware motion mapping.** Robot motor babbling initializes the encoder-to-command mapping; paired human demonstrations and robot replay refine it under contact.
- **Direct visual policy learning.** Matched wrist views support learning from raw images without hand segmentation or generative inpainting. The paper evaluates ACT, Diffusion Policy, and π₀.₅.
- **Natural demonstration collection.** Operators interact directly with objects while wearing the exoskeleton, decoupling human data collection from real-time robot teleoperation.

## Citation

```bibtex
@misc{yu2026seedumisharingexoskeletonhuman,
  title={SEED-UMI: Sharing the Exoskeleton between human and robot for onE-to-one Dexterous demonstration},
  author={Tengbo Yu and Jiahao Wu and Daohan Li and Bingxu Chen and Hao Liu and Xiaojian Ma and Hangxin Liu},
  year={2026},
  eprint={2609.11753},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2609.11753},
}
```

SEED-UMI

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

SEED-UMI

오너의 다른 레포

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README