What Matters in Building Vision-Language-Action Models for Generalist Robots

์ €์ž: Xinghang Li, Peiyan Li, Long Qian, Minghuan Liu, Dong Wang, Jirong Liu, Bingyi Kang, Xiao Ma, Xinlong Wang, Di Guo, Tao Kong, Hanbo Zhang, Huaping Liu | ๋‚ ์งœ: 2024-12-18 | URL: https://arxiv.org/abs/2412.14058 📄 PDF


Essence

Figure 2

Fig. 2: This work mainly considers three key ingredients for building VLAs based on VLMs: How to formulate the problem

Vision-Language-Action (VLA) ๋ชจ๋ธ ๊ฐœ๋ฐœ ์‹œ VLM ๋ฐฑ๋ณธ ์„ ํƒ, ์•„ํ‚คํ…์ฒ˜ ์„ค๊ณ„, ๋ฐ์ดํ„ฐ ํ™œ์šฉ ์‹œ์ ์ด๋ผ๋Š” ์„ธ ๊ฐ€์ง€ ํ•ต์‹ฌ ์š”์†Œ๋ฅผ ์ฒด๊ณ„์ ์œผ๋กœ ๋ถ„์„ํ•˜๊ณ , ์ด๋ฅผ ํ†ตํ•ด RoboVLMs ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ œ์•ˆํ•˜์—ฌ ๋กœ๋ด‡ ์กฐ์ž‘ ์ž‘์—…์—์„œ ์ตœ๊ณ  ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•œ๋‹ค.

Motivation

Achievement

Figure 4

Fig. 4: The experimental results for RoboVLMs in Simulations and real world.

How

Figure 2

Fig. 2: This work mainly considers three key ingredients for building VLAs based on VLMs: How to formulate the problem

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: VLA ๊ฐœ๋ฐœ์˜ ํ•ต์‹ฌ ์„ค๊ณ„ ์š”์†Œ๋ฅผ ์ฒด๊ณ„์ ์œผ๋กœ ๋ถ„์„ํ•œ ์ค‘์š”ํ•œ ๋ฉ”ํƒ€ ์—ฐ๊ตฌ๋กœ, ๊ด‘๋ฒ”์œ„ํ•œ ์‹ค์ฆ ์‹คํ—˜์„ ํ†ตํ•ด ์‹ค์งˆ์ ์ธ ๊ฐ€์ด๋“œ๋ผ์ธ์„ ์ œ์‹œํ•˜๊ณ  ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ œ๊ณตํ•จ์œผ๋กœ์จ ๋กœ๋ด‡ ๊ธฐ์ดˆ ๋ชจ๋ธ ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ์ƒ๋‹นํ•œ ๊ธฐ์—ฌ๋ฅผ ํ•  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋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˜์ •