상세정보
학술대회
Virtual-to-Real Transfer via Dynamics Models
- 발행일
- 201906
- 출처
- International Conference on Mobile Systems, Applications, and Services (MobiSys) 2019, pp.516-517
- 협약과제
- 19ZH1100, 사물-사람-공간의 유기적 연결을 위한 초연결 공간의 분산 지능 핵심원천 기술, 박준희
- 초록
- The virtual world is essential for deep reinforcement learning. Since the deep reinforcement learning agent learns the optimal policy by interacting with the environment in a trial and error manner, training the agent in the real world is not only cost expensive and time-consuming but also unsafe. Several researches are ongoing in the field of but not limited to drone, vehicle, and robot arm control as the deep reinforcement learning is proven to be an effective solution to sequential decision-making problems such as Atari games [2] and several board games including Go [4]. Due to the above issue, most of these researches are done in the virtual world that mimics the real world.
- KSP 제안 키워드
- Deep reinforcement learning, Dynamics model, Optimal policy, Real-world, Reinforcement Learning(RL), Robot arm control, Sequential decision-making, Virtual world, board games, learning agent, trial and error