GMT: General Motion Tracking for Humanoid Whole-Body Control

์ €์ž: Zixuan Chen, Mazeyu Ji, Xuxin Cheng, Xuanbin Peng, Xue Bin Peng, Xiaolong Wang | ๋‚ ์งœ: 2025-06-17 | URL: https://arxiv.org/abs/2506.14770 📄 PDF


Essence

Figure 3

Figure 3: An overview of GMT. Here gt denotes the motion target frame, ot denotes proprioceptive

GMT๋Š” humanoid ๋กœ๋ด‡์ด ๋‹ค์–‘ํ•œ ์ „์‹  ๋ชจ์…˜์„ ์ถ”์ ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ํ†ตํ•ฉ ์ •์ฑ…์„ ํ•™์Šตํ•˜๋Š” ํ”„๋ ˆ์ž„์›Œํฌ๋กœ, Adaptive Sampling ์ „๋žต๊ณผ Motion Mixture-of-Experts ์•„ํ‚คํ…์ฒ˜๋ฅผ ํ•ต์‹ฌ ์š”์†Œ๋กœ ์ œ์•ˆํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Figure 1: We deploy the general unified motion tracking policy on a medium-sized humanoid robot.

How

Figure 3

Figure 3: An overview of GMT. Here gt denotes the motion target frame, ot denotes proprioceptive

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: GMT๋Š” humanoid ๋กœ๋ด‡์˜ general motion tracking์— ๋Œ€ํ•œ ์‹ค์งˆ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•˜๋ฉฐ, Adaptive Sampling๊ณผ Motion MoE๋ผ๋Š” ๋‘ ๊ฐ€์ง€ ์‹ค์šฉ์  ๊ธฐ๋ฒ•์œผ๋กœ ๊ธฐ์กด์˜ ์‚ฐ๋ฐœ์  ์ ‘๊ทผ๋“ค์„ ํ†ตํ•ฉํ•œ ์šฐ์ˆ˜ํ•œ ์—ฐ๊ตฌ์ด๋‹ค. ์‹ค์ œ ๋กœ๋ด‡ ๋ฐฐํฌ ์„ฑ๊ณต๊ณผ ์ƒํƒœ-์ตœ์ฒจ๋‹จ ์„ฑ๋Šฅ์€ ๋†’์€ ๊ฐ€์น˜๋ฅผ ์ œ์‹œํ•˜์ง€๋งŒ, ๋” ๊ด‘๋ฒ”์œ„ํ•œ ํ•˜๋“œ์›จ์–ด ๊ฒ€์ฆ๊ณผ ์ด๋ก ์  ๋ถ„์„ ๊ฐ•ํ™”๊ฐ€ ํ•„์š”ํ•˜๋‹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˜์ •