EgoHumanoid: Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration

์ €์ž: Modi Shi, Shijia Peng, Jin Chen, Haoran Jiang, Yinghui Li, Di Huang, Ping Luo, Hongyang Li, Li Chen | ๋‚ ์งœ: 2026-02-10 | DOI: 10.48550/arXiv.2602.10106 📄 PDF


Essence

Figure 1

Fig. 1: Introducing EGOHUMANOID, the first investigation on human-to-humanoid transfer for whole-body loco-manipulation.

EgoHumanoid๋Š” ๋กœ๋ด‡ ์—†์ด ์ˆ˜์ง‘ํ•œ ๋Œ€๊ทœ๋ชจ ์ธ๊ฐ„ egocentric ์‹œ์—ฐ๊ณผ ์ œํ•œ๋œ ๋กœ๋ด‡ ๋ฐ์ดํ„ฐ๋ฅผ co-trainํ•˜์—ฌ ํœด๋จธ๋…ธ์ด๋“œ ๋กœ๋ด‡์ด ๋‹ค์–‘ํ•œ ํ˜„์‹ค ํ™˜๊ฒฝ์—์„œ loco-manipulation์„ ์ˆ˜ํ–‰ํ•˜๋„๋ก ํ•˜๋Š” ์ฒซ ๋ฒˆ์งธ ํ”„๋ ˆ์ž„์›Œํฌ์ด๋‹ค. View alignment์™€ action alignment๋กœ ๊ตฌ์„ฑ๋œ embodiment ์ •๋ ฌ ํŒŒ์ดํ”„๋ผ์ธ์„ ํ†ตํ•ด ์ธ๊ฐ„-๋กœ๋ด‡ ๊ฐ„์˜ ์‹ ์ฒด ํ˜•ํƒœ, ๊ด€์ , ๋™์—ญํ•™์˜ ์ฐจ์ด๋ฅผ ๊ทน๋ณตํ•œ๋‹ค.

Motivation

Achievement

Figure 5

Fig. 5: Performance of human-robot data co-training with EGOHUMANOID. Our pipeline achieves unanimous improvements

How

Figure 3

Fig. 3: Pipeline of human-to-humanoid alignment. (a) View Alignment: Egocentric images are transformed to approximate

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: EgoHumanoid๋Š” ํœด๋จธ๋…ธ์ด๋“œ loco-manipulation ๋ถ„์•ผ์—์„œ human egocentric data ํ™œ์šฉ์˜ ์ƒˆ๋กœ์šด ๊ฐ€๋Šฅ์„ฑ์„ ์ฒด๊ณ„์ ์œผ๋กœ ๋ณด์—ฌ์ฃผ๋Š” ํš๊ธฐ์ ์ธ ์ž‘์—…์ด๋‹ค. Practical embodiment alignment pipeline, ํ˜„์‹ค ํ™˜๊ฒฝ์—์„œ์˜ ๊ฐ•๋ ฅํ•œ ์„ฑ๋Šฅ ๊ฐœ์„ (51%), ๊ทธ๋ฆฌ๊ณ  scalability ๋ถ„์„์€ ํ–ฅํ›„ humanoid ๋กœ๋ด‡ ํ•™์Šต์˜ ์ค‘์š”ํ•œ ๋ฐฉํ–ฅ์„ ์ œ์‹œํ•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˜์ •