OmniRetarget: Interaction-Preserving Data Generation for Humanoid Whole-Body Loco-Manipulation and Scene Interaction

์ €์ž: Lujie Yang, Xiaoyu Huang, Zhen Wu, Angjoo Kanazawa, Pieter Abbeel, Carmelo Sferrazza, C. Karen Liu, Rocky Duan, Guanya Shi | ๋‚ ์งœ: 2025-10-08 | DOI: 10.48550/arXiv.2509.26633 📄 PDF


Essence

Figure 2

Fig. 2: OMNIRETARGET overview. Human demonstrations are retargeted to the robot via interaction-meshโ€“based

OmniRetarget์€ interaction mesh ๊ธฐ๋ฐ˜์˜ ์ œ์•ฝ ์ตœ์ ํ™”๋ฅผ ํ†ตํ•ด human motion์„ humanoid robot์„ ์œ„ํ•œ ๊ณ ํ’ˆ์งˆ kinematic reference๋กœ retargetํ•˜๋ฉฐ, ์ƒํ˜ธ์ž‘์šฉ์„ ๋ณด์กดํ•˜๋ฉด์„œ ๋‹จ์ผ ์‹œ์—ฐ์œผ๋กœ๋ถ€ํ„ฐ ๋‹ค์–‘ํ•œ ๋กœ๋ด‡ ๊ตฌ์ฒดํ™”, ์ง€ํ˜•, ๋ฌผ์ฒด ์„ค์ •์œผ๋กœ ํšจ์œจ์ ์ธ data augmentation์„ ์ˆ˜ํ–‰ํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Fig. 1:

How

Figure 2

Fig. 2: OMNIRETARGET overview. Human demonstrations are retargeted to the robot via interaction-meshโ€“based

Originality

Limitation & Further Study

Evaluation

Novelty: 4/5 Technical Soundness: 3/5 Significance: 4/5 Clarity: 4/5 Overall: 4/5

์ดํ‰: OmniRetarget์€ interaction-preserving motion retargeting๊ณผ ์ฒด๊ณ„์  data augmentation์„ ํ†ตํ•ด humanoid robot ์ œ์–ด์˜ ๋ฐ์ดํ„ฐ ๋ณ‘๋ชฉ์„ ํ•ด๊ฒฐํ•˜๋Š” ์‹ค์งˆ์ ์ด๊ณ  ์˜ํ–ฅ๋ ฅ ์žˆ๋Š” ๊ธฐ์—ฌ์ด๋ฉฐ, ์ตœ์†Œํ•œ์˜ reward engineering์œผ๋กœ complex whole-body loco-manipulation ๊ธฐ์ˆ ์˜ zero-shot sim-to-real transfer๋ฅผ ์„ฑ๊ณต์ ์œผ๋กœ ์ž…์ฆํ•˜์—ฌ ๋กœ๋ณดํ‹ฑ์Šค ์ปค๋ฎค๋‹ˆํ‹ฐ์— ๋งค์šฐ ์œ ์šฉํ•œ ๊ณต๊ฐœ ๋„๊ตฌ ๋ฐ ๋ฐ์ดํ„ฐ์…‹์„ ์ œ๊ณตํ•œ๋‹ค.

← ๋ชฉ๋ก์œผ๋กœ ๋Œ์•„๊ฐ€๊ธฐ

๐ŸŽง Audio Overview

์ด ๋…ผ๋ฌธ ๋ฆฌ๋ทฐ๋ฅผ ํŒŸ์บ์ŠคํŠธํ˜• ์˜ค๋””์˜ค๋กœ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. (Gemini ยท ํ‚ค๋Š” ๋ธŒ๋ผ์šฐ์ €์—๋งŒ ์ €์žฅ ยท ์™„์„ฑ๋ณธ์€ ์ด๋ฉ”์ผ๋กœ๋„ ์ „์†ก)
โ–ธ ๊ณ ๊ธ‰: ๊ตฌ์„ฑ ๋ฐฉํ–ฅ(๋Œ€๋ณธ ์ž‘์„ฑ ์ง€์นจ) ์ง์ ‘ ์ˆ˜์ •