Physics Informed Intrinsic Rewards in Reinforcement Learning

Title: Physics Informed Intrinsic Rewards in Reinforcement Learning
Creator: Jiang, Jiazhou; Fu, Minyue; Chen, Zhiyong
Relation: 2022 Australian & New Zealand Control Conference (ANZCC). Proceedings of the 2022 Australian & New Zealand Control Conference (ANZCC) (Gold Coast, Queensland 24-25 November, 2022) p. 69-74
Publisher Link: http://dx.doi.org/10.1109/ANZCC56036.2022.9966956
Publisher: Institute of Electronics and Electrical Engineers (IEEE)
Resource Type: conference paper
Date: 2022
Description: Model-free algorithms in Reinforcement Learning (RL) are known to be a powerful learning tool and have performed well in solving complex issues. However, RL training results are often poor when the reward function is sparse or misleading in short term. In this paper, we propose a physics informed intrinsic reward function to assist the agent to overcome this difficulty. We evaluate the proposed intrinsic reward method on different types of actor-critic (AC) algorithms. The experimental results show noticeable improvement.
Subject: space vehicles; training; moon; reinforcement learning; physics; tuning
Identifier: http://hdl.handle.net/1959.13/1489514
Identifier: uon:52718
Identifier: ISBN:9781665498876
Language: eng
Reviewed

		Thumbnail	File	Description	Size	Format