SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model

์ €์ž: Delin Qu, Haoming Song, Qizhi Chen, Yuanqi Yao, Xinyi Ye, Yan Ding, Zhigang Wang, JiaYuan Gu, Bin Zhao, Dong Wang, Xuelong Li | ๋‚ ์งœ: 2025-01-27 | URL: https://arxiv.org/abs/2501.15830 📄 PDF


Essence

Figure 2

Fig. 2: Overview of SpatialVLA. Given an image observation ot and a task instruction L, the model processes the image

๋กœ๋ด‡ ์กฐ์ž‘์„ ์œ„ํ•œ 3D ๊ณต๊ฐ„ ์ดํ•ด๋ฅผ ๊ฐ•ํ™”ํ•œ VLA ๋ชจ๋ธ SpatialVLA๋ฅผ ์ œ์•ˆํ•˜๋ฉฐ, Ego3D Position Encoding๊ณผ Adaptive Action Grids๋ฅผ ํ†ตํ•ด ์ด์งˆ์ ์ธ ๋กœ๋ด‡ ๊ฐ„ ์ผ๋ฐ˜ํ™” ๊ฐ€๋Šฅํ•œ ๊ณต๊ฐ„ ํ‘œํ˜„์„ ํ•™์Šตํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Fig. 1: We present SpatialVLA, a spatial-enhanced vision-language-action model that is trained on 1.1 Million real robot

How

Figure 2

Fig. 2: Overview of SpatialVLA. Given an image observation ot and a task instruction L, the model processes the image

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: ๋ณธ ๋…ผ๋ฌธ์€ VLA ๋ชจ๋ธ์— ์ฒด๊ณ„์ ์ธ 3D ๊ณต๊ฐ„ ์ดํ•ด๋ฅผ ๋„์ž…ํ•˜๊ณ  ์ด์งˆ์  ๋กœ๋ด‡ ๊ฐ„ ์ผ๋ฐ˜ํ™”๋ฅผ ๋‹ฌ์„ฑํ•œ ์ค‘์š”ํ•œ ๊ธฐ์—ฌ๋ฅผ ์ œ์‹œํ•˜๋ฉฐ, ๊ด‘๋ฒ”์œ„ํ•œ ์‹คํ—˜์„ ํ†ตํ•ด ์ œ์•ˆ ๋ฐฉ๋ฒ•์˜ ํšจ๊ณผ๋ฅผ ์ž…์ฆํ–ˆ์œผ๋‚˜, ์นด๋ฉ”๋ผ ์˜์กด์„ฑ๊ณผ ์ด์‚ฐํ™” ํ•ด์ƒ๋„ ์ œ์•ฝ ๋“ฑ์˜ ํ•œ๊ณ„๊ฐ€ ์กด์žฌํ•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˜์ •