SMASH: Mastering Scalable Whole-Body Skills for Humanoid Ping-Pong with Egocentric Vision

์ €์ž: | ๋‚ ์งœ: 2026-04-01 | URL: https://arxiv.org/abs/2604.01158 📄 PDF


Essence

Figure 2

Fig. 2: Overview of SMASH. Our system connects scalable motion generation, task-aligned policy learning, and egocentric

ํœด๋จธ๋…ธ์ด๋“œ ๋กœ๋ด‡์˜ ํƒ๊ตฌ ๊ฒŒ์ž„์„ ์œ„ํ•ด ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ์ „์‹  ๋™์ž‘ ํ•™์Šต๊ณผ ์ž์ฒด ์—๊ณ ์„ผํŠธ๋ฆญ ๋น„์ „์„ ํ†ตํ•ฉํ•œ SMASH ์‹œ์Šคํ…œ์„ ์ œ์‹œํ•˜๋ฉฐ, ์™ธ๋ถ€ ์นด๋ฉ”๋ผ๋‚˜ ๋ชจ์…˜ ์บก์ฒ˜ ์—†์ด ์‹ค์™ธ์—์„œ ์—ฐ์†์ ์ธ ํƒ๊ตฌ ์ŠคํŠธ๋ผ์ดํ‚น์„ ์ฒ˜์Œ์œผ๋กœ ๋‹ฌ์„ฑํ–ˆ๋‹ค.

Motivation

Achievement

Figure 1

Fig. 1: SMASH: Our system enables the first outdoor humanoid ping-pong player and the first whole-body smash on a humano

How

Figure 2

Fig. 2: Overview of SMASH. Our system connects scalable motion generation, task-aligned policy learning, and egocentric

Originality

Limitation & Further Study

Evaluation

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

์ดํ‰: ์ด ๋…ผ๋ฌธ์€ ํœด๋จธ๋…ธ์ด๋“œ ํƒ๊ตฌ์—์„œ ์—๊ณ ์„ผํŠธ๋ฆญ ์˜จ๋ณด๋“œ ์ง€๊ฐ๊ณผ ์ „์‹  ํ˜‘์‘ ์ œ์–ด๋ฅผ ํ†ตํ•ฉํ•œ ์ตœ์ดˆ์˜ ์ž์œจ ์‹œ์Šคํ…œ์„ ๊ตฌํ˜„ํ•จ์œผ๋กœ์จ ๋กœ๋ด‡ ๋™์  ์ƒํ˜ธ์ž‘์šฉ ์—ฐ๊ตฌ์— ์ค‘์š”ํ•œ ๊ธฐ์—ฌ๋ฅผ ํ•˜์˜€๋‹ค. Motion VAE ๊ธฐ๋ฐ˜ ๋™์ž‘ ํ™•์žฅ๊ณผ task-aligned motion matching์ด๋ผ๋Š” ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ๋ฐฉ๋ฒ•๋ก ์€ ๋‹ค๋ฅธ ๋™์  ๋กœ๋ด‡ ๊ณผ์ œ์—๋„ ์ ์šฉ ๊ฐ€๋Šฅํ•œ ์ž ์žฌ๋ ฅ์ด ์žˆ๋‹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˜์ •