Visual Imitation Enables Contextual Humanoid Control

์ €์ž: Arthur Allshire, Hongsuk Choi, Junyi Zhang, David McAllister, Anthony Zhang, Chung Min Kim, Trevor Darrell, Pieter Abbeel, Jitendra Malik, Angjoo Kanazawa | ๋‚ ์งœ: 2025-05-06 | URL: https://arxiv.org/abs/2505.03729 📄 PDF


Essence

Figure 2

Figure 2: VideoMimic Real-to-Sim. A casually captured phone video provides the only input. We first

VIDEOMIMIC๋Š” ๋‹จ์ˆœํ•œ ํœด๋Œ€ํฐ ์˜์ƒ์—์„œ ์ธ๊ฐ„-ํ™˜๊ฒฝ 4D ๊ธฐํ•˜ํ•™์„ ๊ณต๋™ ์žฌ๊ตฌ์„ฑํ•˜๊ณ , ์ด๋ฅผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ RL ์ •์ฑ…์œผ๋กœ ํ•™์Šตํ•œ ํ›„ ์‹ค์ œ ํœด๋จธ๋…ธ์ด๋“œ ๋กœ๋ด‡์— ๋ฐฐํฌํ•˜๋Š” real-to-sim-to-real ํŒŒ์ดํ”„๋ผ์ธ์ด๋‹ค.

Motivation

Achievement

Figure 5

Figure 5: The policy performing various skills on the real robot: traversing complex terrain, standing, and

How

Figure 2

Figure 2: VideoMimic Real-to-Sim. A casually captured phone video provides the only input. We first

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: ์ด ๋…ผ๋ฌธ์€ ์ผ์ƒ ์˜์ƒ์œผ๋กœ๋ถ€ํ„ฐ ํœด๋จธ๋…ธ์ด๋“œ ๋กœ๋ด‡์˜ ๋ฌธ๋งฅ-์ธ์‹ ์ œ์–ด๋ฅผ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜๋Š” ์‹ค์šฉ์ ์ด๊ณ  ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ํŒŒ์ดํ”„๋ผ์ธ์„ ์ œ์‹œํ•˜๋ฉฐ, ๊ณต๋™ 4D ์žฌ๊ตฌ์„ฑ๊ณผ RL ๊ธฐ๋ฐ˜ ์ •์ฑ… ์ฆ๋ฅ˜์˜ ์กฐํ•ฉ์œผ๋กœ ๋†’์€ ๋…์ฐฝ์„ฑ์„ ๋ณด์ธ๋‹ค. ์‹ค์ œ ๋กœ๋ด‡ ๋ฐฐํฌ ์„ฑ๊ณต์€ ์—ฐ๊ตฌ์˜ ๊ฐ€์น˜๋ฅผ ํฌ๊ฒŒ ๋†’์ด๋‚˜, ํ™˜๊ฒฝ ํ‘œํ˜„์˜ ์ œํ•œ์„ฑ๊ณผ ๋™์—ญํ•™ ์ •ํ™•๋„ ์ธก๋ฉด์—์„œ ๊ฐœ์„  ์—ฌ์ง€๊ฐ€ ์žˆ๋‹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˜์ •