RuN: Residual Policy for Natural Humanoid Locomotion

์ €์ž: Qingpeng Li, Chengrui Zhu, Yanming Wu, Xin Yuan, Zhen Zhang, Jian Yang, Yong Liu | ๋‚ ์งœ: 2025-09-25 | URL: https://arxiv.org/abs/2509.20696 📄 PDF


Essence

Figure 2

Fig. 2: Overview of the RuN framework. (a) Motion Retargeting: Raw human motions are converted into a kinematically feas

RuN์€ Conditional Motion Generator๋ฅผ ํ†ตํ•œ ์šด๋™ํ•™์  ๋ชจ์…˜ ํ”„๋ผ์ด์–ด์™€ ๊ฐ•ํ™”ํ•™์Šต ๊ธฐ๋ฐ˜ residual policy๋ฅผ ๋ถ„๋ฆฌํ•˜์—ฌ, ์ธํ˜•๋กœ๋ด‡์˜ ์ž์—ฐ์Šค๋Ÿฌ์šด ๋ณดํ–‰-๋‹ฌ๋ฆฌ๊ธฐ ์ „ํ™˜์„ ์‹คํ˜„ํ•˜๋Š” decoupled residual learning ํ”„๋ ˆ์ž„์›Œํฌ์ด๋‹ค.

Motivation

Achievement

Figure 5

Fig. 5: Performance comparison of different algorithms. This figure shows

How

Figure 2

Fig. 2: Overview of the RuN framework. (a) Motion Retargeting: Raw human motions are converted into a kinematically feas

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: RuN์€ humanoid locomotion ์ œ์–ด์˜ ๊ทผ๋ณธ์ ์ธ ๋ณต์žก์„ฑ์„ elegantํ•˜๊ฒŒ ํ•ด๊ฒฐํ•œ well-motivated ํ”„๋ ˆ์ž„์›Œํฌ๋กœ, decoupled residual learning ์ ‘๊ทผ์ด ํ•™์Šต ํšจ์œจ์„ฑ๊ณผ ์ตœ์ข… ์„ฑ๋Šฅ์„ ๋ชจ๋‘ ๊ฐœ์„ ํ•˜๋ฉฐ ์‹ค์ œ ๋กœ๋ด‡์—์„œ ๊ฒ€์ฆ๋œ ๊ฐ•๋ ฅํ•œ ๋ฐฉ๋ฒ•๋ก ์ด๋‹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˜์ •