FAME: Force-Adaptive RL for Expanding the Manipulation Envelope of a Full-Scale Humanoid

์ €์ž: Niraj Pudasaini, Yutong Zhang, Jensen Lavering, Alessandro Roncone, Nikolaus Correll | ๋‚ ์งœ: 2026-03-09 | URL: https://arxiv.org/abs/2603.08961 📄 PDF


Essence

Figure 2

Fig. 2: Overview of the proposed standing framework. During training (top), an upper-body dynamics encoder processes

FAME๋Š” ์–‘ํŒ” ์กฐ์ž‘ ์‹œ ์™ธ๋ถ€ ์† ํž˜์œผ๋กœ ์ธํ•œ ๊ท ํ˜• ๊ต๋ž€์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด, ์ƒ์ฒด ๊ด€์ ˆ ๊ตฌ์„ฑ๊ณผ ์–‘ํŒ” ์ƒํ˜ธ์ž‘์šฉ ํž˜์„ ์ธ์ฝ”๋”ฉํ•˜๋Š” latent context์— ์กฐ๊ฑดํ™”๋œ RL ์ •์ฑ…์„ ํ•™์Šตํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Fig. 1: FAME overview and real demonstration. Left: FAME conditions a standing policy on an upper-body context encoder t

How

Figure 2

Fig. 2: Overview of the proposed standing framework. During training (top), an upper-body dynamics encoder processes

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: FAME๋Š” latent context adaptation์„ ์–‘ํŒ” ์กฐ์ž‘ ์ค‘ balance ๋ฌธ์ œ์— ์ฐฝ์˜์ ์œผ๋กœ ์ ์šฉํ•˜๋ฉฐ, ์„ผ์„œ ๋ถˆํ•„์š” ๋ฐฐํฌ์™€ ์‹ค์„ธ๊ณ„ ๊ฒ€์ฆ์œผ๋กœ ์‹ค์šฉ์  ๊ธฐ์—ฌ๋ฅผ ํ•œ๋‹ค. ๋‹ค๋งŒ sim-to-real ๊ฒฉ์ฐจ์™€ ํž˜ ์ถ”์ • ์ •ํ™•๋„ ๋ถ„์„์ด ๋ณด๊ฐ•๋˜๋ฉด ๋”์šฑ ๊ฐ•๋ ฅํ•ด์งˆ ๊ฒƒ์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˜์ •