Sign In

Learning Sim-Grounded Policies for Bimanual Rope Manipulation from Human Teleoperation Data

์ž‘์„ฑ์ž
  • Haebom
์นดํ…Œ๊ณ ๋ฆฌ
Empty

์ €์ž

Gina Wigginghaus, Tim Missal, Berk Guler, Simon Manschitz, Jan Peters

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ์—ฐ๊ตฌ๋Š” ๊ฐ€์ • ๋ฐ ์‚ฐ์—… ํ˜„์žฅ์—์„œ ํ”ํžˆ ์ ‘ํ•˜์ง€๋งŒ ๋‹ค๋ฃจ๊ธฐ ์–ด๋ ค์šด ๋Š˜์–ด๋‚˜๋Š” ์„ ํ˜• ๊ฐ์ฒด(DLO)์˜ ์ด์ข… ์กฐ์ž‘์„ ์œ„ํ•œ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๊ธฐ๋ฐ˜ ์ •์ฑ… ํ•™์Šต ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ธ๊ฐ„์˜ ์›๊ฒฉ ์กฐ์ž‘ ๋ฐ์ดํ„ฐ๋กœ๋ถ€ํ„ฐ ํ•™์Šตํ•˜๋Š” ๋ชจ๋ฐฉ ํ•™์Šต ๋ฐฉ์‹์˜ ํ™•์žฅ์„ฑ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•˜๊ธฐ ์œ„ํ•ด, ์‹œ๊ฐ์  ๊ด€์ฐฐ ๊ณต๊ฐ„ ์ž์ฒด์˜ ๋ฌธ์ œ์ ์„ ๋ถ„์„ํ•˜๊ณ  DLO์˜ 3D ์ž…์ž ์ƒํƒœ๋ฅผ ์ด์šฉํ•œ ์ •์ฑ…์ด RGB ์˜์ƒ ๊ธฐ๋ฐ˜ ์ •์ฑ…๋ณด๋‹ค ์šฐ์ˆ˜ํ•œ ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ์„ ๋ณด์ž„์„ ์ž…์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ œํ•œ๋œ ๋ฐ์ดํ„ฐ๋กœ DLO ์กฐ์ž‘์„ ํ•™์Šตํ•˜๋Š” ๋ฐ ์žˆ์–ด ๋ฐ์ดํ„ฐ ํšจ์œจ์„ฑ์„ ๋†’์ผ ์ˆ˜ ์žˆ์Œ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
DLO์™€ ๊ฐ™์ด ๋ณต์žกํ•œ ๋ฌผ์ฒด๋ฅผ ๋‹ค๋ฃจ๋Š” ๋ฐ ์žˆ์–ด, ํ”ฝ์…€ ์ˆ˜์ค€์˜ ์‹œ๊ฐ ์ •๋ณด๋ณด๋‹ค ๋ฌผ๋ฆฌ์ ์œผ๋กœ ์ผ๊ด€๋œ 3D ์ƒํƒœ ์ •๋ณด๊ฐ€ ๋” ๋‚˜์€ ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
โ€ข
์ œํ•œ๋œ ์ธ๊ฐ„ ์‹œ์—ฐ ๋ฐ์ดํ„ฐ๋กœ๋„ ํšจ๊ณผ์ ์ธ ๋กœ๋ด‡ ํ•™์Šต์ด ๊ฐ€๋Šฅํ•˜๋ฉฐ, ํŠนํžˆ DLO ์กฐ์ž‘๊ณผ ๊ฐ™์€ ๊ณผ์ œ์—์„œ๋Š” ๊ด€์ฐฐ ๊ณต๊ฐ„์˜ ์„ค๊ณ„๊ฐ€ ๋งค์šฐ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค.
โ€ข
๋ณธ ์—ฐ๊ตฌ๋Š” ํŠน์ • ์ž‘์—…(๋งค๋“ญ ํ’€๊ธฐ)์— ๋Œ€ํ•œ ๊ฒฐ๊ณผ์ด๋ฉฐ, ๋‹ค์–‘ํ•œ DLO ์กฐ์ž‘ ์ž‘์—…์— ๋Œ€ํ•œ ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ ๊ฒ€์ฆ ๋ฐ ๋” ๋ณต์žกํ•œ ๋ฌผ๋ฆฌ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜๊ฒฝ์—์„œ์˜ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์— ๋Œ€ํ•œ ์ถ”๊ฐ€ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๐Ÿ‘