Embodied-R1: Reinforced Embodied Reasoning for General Robotic Manipulation

์ €์ž: Yifu Yuan, Haiqin Cui, Yaoting Huang, Yibin Chen, Fei Ni, Zibin Dong, Pengyi Li, Yan Zheng, Hongyao Tang, Jianye Hao | ๋‚ ์งœ: 2025-08-19 | URL: https://arxiv.org/abs/2508.13998 📄 PDF


Essence

Figure 1

Figure 1 Overview of the Embodied-R1 framework and its zero-shot manipulation performance.

Embodied-R1์€ 'ํฌ์ธํŒ…'์„ ํ†ต์ผ๋œ embodiment-agnostic ์ค‘๊ฐ„ ํ‘œํ˜„์œผ๋กœ ์ •์˜ํ•˜๊ณ , Reinforced Fine-tuning(RFT)์œผ๋กœ ํ›ˆ๋ จ๋œ 3B VLM์œผ๋กœ์„œ ๋กœ๋ด‡ ์กฐ์ž‘์˜ perception-action gap์„ ํšจ๊ณผ์ ์œผ๋กœ ๊ทน๋ณตํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Figure 1 Overview of the Embodied-R1 framework and its zero-shot manipulation performance.

How

Figure 3

Figure 3 Overview of training data: In stage 1, we focus on improving the modelโ€™s spatial reasoning capability,

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: Embodied-R1์€ ํฌ์ธํŒ…์ด๋ผ๋Š” ๋ช…ํ™•ํ•œ ์ค‘๊ฐ„ ํ‘œํ˜„๊ณผ RFT ๊ธฐ๋ฐ˜ ํ›ˆ๋ จ ๋ฐฉ์‹์œผ๋กœ embodied AI์˜ ์˜ค๋ž˜๋œ perception-action gap ๋ฌธ์ œ์— ์šฐ์•„ํ•œ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•˜๋ฉฐ, ์‹ค์ œ ๋กœ๋ด‡์—์„œ์˜ ๊ฐ•๋ ฅํ•œ zero-shot ์„ฑ๋Šฅ์œผ๋กœ ๊ทธ ์‹ค์งˆ์  ๊ฐ€์น˜๋ฅผ ์ž…์ฆํ•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˜์ •