CReF: Cross-modal and Recurrent Fusion for Depth-conditioned Humanoid Locomotion

์ €์ž: | ๋‚ ์งœ: 2026-03-31 | URL: https://arxiv.org/abs/2603.29452 📄 PDF


Essence

Figure 2

Fig. 2.

CReF๋Š” cross-modal attention๊ณผ gated residual fusion์„ ํ™œ์šฉํ•˜์—ฌ raw depth ์ž…๋ ฅ์œผ๋กœ๋ถ€ํ„ฐ ์ง์ ‘ locomotion-relevant ํŠน์ง•์„ ํ•™์Šตํ•˜๋Š” ๋‹จ์ผ ๋‹จ๊ณ„ depth-conditioned humanoid locomotion ํ”„๋ ˆ์ž„์›Œํฌ๋กœ, ๋ช…์‹œ์  ๊ธฐํ•˜ํ•™์  ์ค‘๊ฐ„ ํ‘œํ˜„ ์—†์ด zero-shot sim-to-real transfer๋ฅผ ๋‹ฌ์„ฑํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Fig. 1.

How

Figure 2

Fig. 2.

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: CReF๋Š” ๋ช…์‹œ์  ๊ธฐํ•˜ํ•™์  ์ค‘๊ฐ„ ํ‘œํ˜„์„ ์ œ๊ฑฐํ•˜๊ณ  cross-modal attention๊ณผ gated recurrent fusion์„ ํ†ตํ•ด raw depth๋กœ๋ถ€ํ„ฐ ์ง์ ‘ locomotion-relevant features๋ฅผ ํ•™์Šตํ•˜๋Š” ํ˜์‹ ์  ์ ‘๊ทผ๋ฒ•์œผ๋กœ, zero-shot sim-to-real transfer์™€ ๋‹ค์–‘ํ•œ ์‹ค์ œ ํ™˜๊ฒฝ์—์„œ์˜ ๊ฐ•๊ฑดํ•œ ์„ฑ๋Šฅ์„ ํ†ตํ•ด humanoid locomotion ๋ถ„์•ผ์— significant contribution์„ ์ 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˜์ •