RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation

์ €์ž: Songhao Han, Boxiang Qiu, Yue Liao, Siyuan Huang, Chen Gao, Shuicheng Yan, Si Liu | ๋‚ ์งœ: 2025-06-07 | URL: https://arxiv.org/abs/2506.06677 📄 PDF


Essence

Figure 1

Figure 1: We shift the focus of robotic imitation learning from fast, reactive System 1 behavior to

RoboCerebra๋Š” ์žฅ๊ธฐ๊ฐ„ ๋กœ๋ด‡ ์กฐ์ž‘ ์ž‘์—… ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ ๋Œ€๊ทœ๋ชจ ๋ฒค์น˜๋งˆํฌ๋กœ, VLM์˜ System 2 (deliberative reasoning) ๋Šฅ๋ ฅ์„ ํ™œ์šฉํ•œ ๊ณ„์ธต์  ๊ณ„ํš-์‹คํ–‰ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ œ์•ˆํ•œ๋‹ค.

Motivation

Achievement

Figure 1

Figure 1: We shift the focus of robotic imitation learning from fast, reactive System 1 behavior to

How

Figure 2

Figure 2: Task generation pipeline in RoboCerebra. (a) Objects are randomly sampled from

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: RoboCerebra๋Š” VLM์˜ System 2 ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•œ ์ฒซ ๋Œ€๊ทœ๋ชจ ๋ฒค์น˜๋งˆํฌ๋กœ์„œ, ๊ธฐ์กด ์žฅ๊ธฐ ๋กœ๋ด‡ ์กฐ์ž‘ ๋ฒค์น˜๋งˆํฌ์˜ ํ•œ๊ณ„๋ฅผ ๋ช…ํ™•ํžˆ ์ง€์ ํ•˜๊ณ  ์ฒด๊ณ„์ ์ธ ๋ฐ์ดํ„ฐ ๋ฐ ํ‰๊ฐ€ ํ”„๋กœํ† ์ฝœ์„ ์ œ์‹œํ•œ๋‹ค. ๋‹ค๋งŒ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜๊ฒฝ ์ œํ•œ๊ณผ ์‹ค์ œ ๋กœ๋ด‡ ์ ์šฉ ๊ฒ€์ฆ ๋ถ€์žฌ๊ฐ€ ์‹ค์šฉ์„ฑ ์ธก๋ฉด์˜ ๊ณผ์ œ์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˜์ •