EGM: Efficiently Learning General Motion Tracking Policy for High Dynamic Humanoid Whole-Body Control

์ €์ž: Chao Yang, Yingkai Sun, Peng Ye, Xin Chen, Chong Yu, Tao Chen | ๋‚ ์งœ: 2025-12-22 | DOI: 10.48550/arXiv.2512.19043 📄 PDF


Essence

Figure 2

Figure 2: Overview of the EGM framework. First, large-scale Mocap datasets are retargeted to Humanoid, then a small

EGM์€ Bin-based Cross-motion Curriculum Adaptive Sampling๊ณผ Composite Decoupled Mixture-of-Experts ์•„ํ‚คํ…์ฒ˜๋ฅผ ํ†ตํ•ด 4.08์‹œ๊ฐ„์˜ ์†Œ๋Ÿ‰ ๋ฐ์ดํ„ฐ๋กœ 49.25์‹œ๊ฐ„์˜ ๋‹ค์–‘ํ•œ ๋ชจ์…˜์„ ํšจ์œจ์ ์œผ๋กœ ์ถ”์ ํ•˜๋Š” ์ผ๋ฐ˜ํ™”๋œ ํœด๋จธ๋…ธ์ด๋“œ ์ œ์–ด ์ •์ฑ…์„ ํ•™์Šตํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Figure 1: We deploy a unified student policy trained with EGM in the simulation environment, achieving high robust

How

Figure 2

Figure 2: Overview of the EGM framework. First, large-scale Mocap datasets are retargeted to Humanoid, then a small

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: EGM์€ Bin-based adaptive sampling๊ณผ CDMoE ์•„ํ‚คํ…์ฒ˜์˜ ์ƒˆ๋กœ์šด ์กฐํ•ฉ์œผ๋กœ humanoid motion tracking์˜ ๋ฐ์ดํ„ฐ ํšจ์œจ์„ฑ๊ณผ dynamic motion ์„ฑ๋Šฅ์„ ํš๊ธฐ์ ์œผ๋กœ ๊ฐœ์„ ํ•˜๋ฉฐ, ์†Œ๋Ÿ‰ ๋ฐ์ดํ„ฐ ํ•™์Šต์˜ ์‹ค์šฉ์„ฑ์„ ์ž…์ฆํ•˜๋Š” ๊ฐ•๋ ฅํ•œ ๊ธฐ์—ฌ๋ฅผ ์ 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˜์ •