์ ์: Jipeng Kong, Xinzhe Liu, Yuhang Lin, Jinrui Han, Sรถren Schwertfeger, Chenjia Bai, Xuelong Li | ๋ ์ง: 2026-02-05 | DOI: 10.48550/arXiv.2602.05310 📄 PDF
Fig. 2: Overview of the Perception-Action integrated Decision-making (PAiD) framework. Our pipeline progressively acquir
๋ณธ ๋ ผ๋ฌธ์ humanoid robot์ด human-like kicking๊ณผ whole-body balance๋ฅผ ๋์์ ์ํํ๋ soccer skill์ ์ต๋ํ๊ธฐ ์ํด, ์ธ ๋จ๊ณ๋ก ๊ตฌ์ฑ๋ Perception-Action integrated Decision-making (PAiD) ํ๋ ์์ํฌ๋ฅผ ์ ์ํ๋ค.
Fig. 4: Quantitative analysis of soccer shooting proficiency across the workspace. The heatmaps visualize the spatial di
Fig. 2: Overview of the Perception-Action integrated Decision-making (PAiD) framework. Our pipeline progressively acquir
์ดํ: ๋ณธ ๋ ผ๋ฌธ์ humanoid robot์ ๋ณต์กํ embodied skill ์ต๋์ ์ํ ์ฒด๊ณ์ ์ธ progressive framework๋ฅผ ์ ์ํ๋ฉฐ, motion tracking-perception integration-sim-to-real transfer์ ์ธ ๋จ๊ณ ๋ถํด๋ฅผ ํตํด ๊ธฐ์กด ๋ฐฉ์์ training instability์ reward conflict๋ฅผ ํจ๊ณผ์ ์ผ๋ก ํด๊ฒฐํ๋ค. 91.3% ์ฑ๊ณต๋ฅ ์ robust real-world kicking ์ฑ๋ฅ๊ณผ diverse condition์์์ ์ผ๊ด์ฑ์ ์ ์ ๋ฐฉ๋ฒ์ ํจ๊ณผ๋ฅผ ์ ์ฆํ๋ฉฐ, divide-and-conquer ์ ๋ต์ ํฅํ complex embodied skill ์ต๋์ scalable framework๋ก ํ์ฉ ๊ฐ๋ฅํ๋ค.