메뉴 건너뛰기
.. 내서재 .. 알림
소속 기관/학교 인증
인증하면 논문, 학술자료 등을  무료로 열람할 수 있어요.
한국대학교, 누리자동차, 시립도서관 등 나의 기관을 확인해보세요
(국내 대학 90% 이상 구독 중)
로그인 회원가입 고객센터 ENG
주제분류

추천
검색
질문

논문 기본 정보

자료유형
학술대회자료
저자정보
최진우 (서울대학교) 서승우 (서울대학교)
저널정보
대한전자공학회 대한전자공학회 학술대회 2022년도 대한전자공학회 하계종합학술대회 논문집
발행연도
2022.6
수록면
1,322 - 1,325 (4page)

이용수

표지
📌
연구주제
📖
연구배경
🔬
연구방법
🏆
연구결과
AI에게 요청하기
추천
검색
질문

초록· 키워드

오류제보하기
An intelligent agent can start by learning easy skills and gradually learn more difficult skills to solve complex problems. On the other hand, reinforcement learning has been shown to be able to solve various problems, but most of the existing RL agents have limitations because they search the environment only with primitive tasks. To solve this problem, a method was proposed to learn multiple policies to solve different simple tasks and to find a combination of learned policies to solve complex tasks. However, in most existing studies, unnecessary sub-policies are learned or the difficulty of sub-policies is learned without adjustment, which hinders solving complex tasks. In this paper, we propose a framework for learning subtasks in order from easy to difficult and solving downstream tasks through combinations of these subtasks. The proposed model consists of three components. First, the sub-task difficulty selector adjusts the difficulty of the sub-task to be learned, and the sub-goal generator generates a sub-goal for reaching the target state according to the set difficulty. A sub-goal planner has a hierarchical structure, which a high-level agent selects the optimal sub-task to solve a problem, and a low-level agent outputs a low-level action to reach each sub-goal. Our method showed high performance compared to other baselines in the maze environment, which is a long-horizon task.

목차

Abstract
Ⅰ. 서론
Ⅱ. 본론
Ⅲ. 구현
Ⅳ. 결론 및 향후 연구 방향
참고문헌

참고문헌 (0)

참고문헌 신청

함께 읽어보면 좋을 논문

논문 유사도에 따라 DBpia 가 추천하는 논문입니다. 함께 보면 좋을 연관 논문을 확인해보세요!

이 논문의 저자 정보

최근 본 자료

전체보기

댓글(0)

0

UCI(KEPA) : I410-ECN-0101-2022-569-001549591