Sign In

asRoBallet: Closing the Sim2Real Gap via Friction-Aware Reinforcement Learning for Underactuated Spherical Dynamics

Created by
  • Haebom
Category
Empty

์ €์ž

Fang Wan, Guangyi Huang, Tianyu Wu, Zishang Zhang, Bangchao Huang, Haoran Sun, Mingdong Chen, Chaoyang Song

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ์—ฐ๊ตฌ๋Š” ์‚ฌ๋žŒ ํ˜•ํƒœ์˜ ๋ณผ๋ด‡ ํ•˜๋“œ์›จ์–ด ํ”Œ๋žซํผ์— ์ ์šฉ๋œ ์ตœ์ดˆ์˜ ์ข…๋‹จ๊ฐ„ ๊ฐ•ํ™”ํ•™์Šต(RL) ๋ณดํ–‰ ์ •์ฑ…์ธ asRoBallet์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ๋ณต์žกํ•œ ๋งˆ์ฐฐ ๋ชจ๋ธ์˜ ํ˜„์‹ค ๊ฒฉ์ฐจ๋กœ ์ธํ•ด ๋ณผ๋ด‡ ์ œ์–ด์˜ ๋‚œ์ œ๊ฐ€ ์žˆ์—ˆ์œผ๋‚˜, ๋ณธ ๋…ผ๋ฌธ์€ ํœ -๋ณผ-๋ฐ”๋‹ฅ ์ƒํ˜ธ์ž‘์šฉ์˜ ๋งˆ์ฐฐ์„ ๊ณ ๋ คํ•œ ๊ณ ์ถฉ์‹ค๋„ ์‹œ๋ฎฌ๋ ˆ์ด์…˜๊ณผ ๋งˆ์ฐฐ ์ธ์ง€ ๊ฐ•ํ™”ํ•™์Šต ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ํ†ตํ•ด ์ œ๋กœ์ƒท Sim2Real ์ „์†ก์„ ๋‹ฌ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๋ณผ๋ด‡์˜ ์•ˆ์ •์ ์ธ ์ด๋™ ์ œ์–ด๋ฅผ ์„ฑ๊ณต์ ์œผ๋กœ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
Sim2Real ๊ฒฉ์ฐจ ํ•ด์†Œ: ์‹ค์ œ ํ•˜๋“œ์›จ์–ด์˜ ๋ณต์žกํ•œ ๋งˆ์ฐฐ ๋™์—ญํ•™์„ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์— ์ •ํ™•ํžˆ ๋ชจ๋ธ๋งํ•˜๊ณ , ์ด๋ฅผ ๊ฐ•ํ™”ํ•™์Šต์— ํ†ตํ•ฉํ•จ์œผ๋กœ์จ ์‹œ๋ฎฌ๋ ˆ์ด์…˜๊ณผ ์‹ค์ œ ํ™˜๊ฒฝ ๊ฐ„์˜ ๊ฒฉ์ฐจ๋ฅผ ํšจ๊ณผ์ ์œผ๋กœ ์ค„์˜€์Šต๋‹ˆ๋‹ค.
โ€ข
๋งˆ์ฐฐ ์ธ์ง€ ๊ฐ•ํ™”ํ•™์Šต์˜ ์ค‘์š”์„ฑ: ํœ -๋ณผ ๋ฐ ๋ณผ-๋ฐ”๋‹ฅ ์ ‘์ด‰๋ฉด์—์„œ์˜ ๋ณตํ•ฉ์ ์ธ ๋งˆ์ฐฐ ์ฑ„๋„(๊ตฌ๋ฆ„, ์ธก๋ฉด, ๋น„ํ‹€๋ฆผ)์„ ํ•™์Šตํ•˜๋Š” ๊ฒƒ์ด underactuated ์‹œ์Šคํ…œ์˜ ์„ฑ๊ณต์ ์ธ ํ•˜๋“œ์›จ์–ด ์ ์šฉ์— ํ•„์ˆ˜์ ์ž„์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
โ€ข
ํ•˜๋“œ์›จ์–ด ๊ฐœ๋ฐœ์˜ ์‹ค์šฉ์„ฑ: ๊ธฐ์กด ๋กœ๋ด‡ ๋ถ€ํ’ˆ์„ ์žฌํ™œ์šฉํ•˜๊ณ  ์ €๋น„์šฉ์œผ๋กœ ์—ฐ๊ตฌ ํ”Œ๋žซํผ์„ ๊ตฌ์ถ•ํ•˜๋Š” ์ ‘๊ทผ ๋ฐฉ์‹์€ ํ–ฅํ›„ ๋กœ๋ด‡ ์—ฐ๊ตฌ ๋ฐ ๊ฐœ๋ฐœ์— ์œ ์šฉํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
โ€ข
ํ•œ๊ณ„์ : ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์•ˆ๋œ asRoBallet์€ ํŠน์ • ์œ ํ˜•์˜ ๋ณผ๋ด‡๊ณผ ๋งˆ์ฐฐ ๋ชจ๋ธ์— ์ตœ์ ํ™”๋˜์–ด ์žˆ์„ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ๋‹ค์–‘ํ•œ ํ™˜๊ฒฝ ์กฐ๊ฑด ๋ฐ ๋Œ๋ฐœ ์ƒํ™ฉ์— ๋Œ€ํ•œ ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ ๊ฒ€์ฆ์ด ์ถ”๊ฐ€์ ์œผ๋กœ ํ•„์š”ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage