Zero-Shot Whole-Body Humanoid Control via Behavioral Foundation Models

์ €์ž: Andrea Tirinzoni, Ahmed Touati, Jesse Farebrother, Mateusz Guzek, Anssi Kanervisto, Yingchen Xu, Alessandro Lazaric, Matteo Pirotta | ๋‚ ์งœ: 2025-04-15 | URL: https://arxiv.org/abs/2504.11054 📄 PDF


Essence

Figure 1

Figure 1 META MOTIVO is the first behavioral foundation model for humanoid agents that can solve whole-body control task

Forward-Backward representations with Conditional-Policy Regularization (FB-CPR)์„ ํ†ตํ•ด unlabeled behavior dataset์œผ๋กœ unsupervised RL์„ ์ •๊ทœํ™”ํ•˜์—ฌ, humanoid agent์˜ zero-shot whole-body control์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜๋Š” behavioral foundation model Meta Motivo๋ฅผ ๊ฐœ๋ฐœํ–ˆ๋‹ค.

Motivation

Achievement

Figure 3

Figure 3 Human-evaluation. Left figure reports the percentage of times a behavior solved a reward-based (blue) or a goal

How

Figure 2

Figure 2 Illustration of the main components of FB-CPR: the discriminator is trained to estimate the ratio between the l

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: FB-CPR์€ unsupervised RL์˜ exploration ํ•œ๊ณ„๋ฅผ behavior dataset ์ •๊ทœํ™”๋กœ ์ฐฝ์˜์ ์œผ๋กœ ํ•ด๊ฒฐํ•˜๊ณ , ๋ณต์žกํ•œ humanoid ์ œ์–ด์—์„œ zero-shot generalization์„ ๋‹ฌ์„ฑํ•œ ๊ธฐ์ˆ ์ ์œผ๋กœ ๊ฒฌ์‹คํ•˜๊ณ  ์˜๋ฏธ ์žˆ๋Š” ์—ฐ๊ตฌ์ด๋‹ค. ์žฌํ˜„์„ฑ ๋ณด์žฅ๊ณผ ๋‹ค์–‘ํ•œ ํ‰๊ฐ€๋Š” ๊ฐ•์ ์ด๋‚˜, ๋ฐ์ดํ„ฐ์…‹ ์˜์กด์„ฑ๊ณผ ์‹ค์ œ ๋กœ๋ด‡ ๊ฒ€์ฆ ๋ถ€์žฌ๋Š” ํ–ฅํ›„ ๊ฐœ์„ ์ด ํ•„์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˜์ •