Learning to select goals in Automated Planning with Deep-Q Learning

作者：

Highlights：

• Deep Q-Learning is used by a planning agent to learn to select subgoals.

• Our approach reduces planning time in online execution.

• Our approach generalizes better than standard Deep Q-Learning.

• Our approach is more sample-efficient than standard Deep Q-Learning.

摘要

•Deep Q-Learning is used by a planning agent to learn to select subgoals.•Our approach reduces planning time in online execution.•Our approach generalizes better than standard Deep Q-Learning.•Our approach is more sample-efficient than standard Deep Q-Learning.

论文关键词：Automated Planning,Goal selection,Deep Q-Learning

论文评审过程：Received 3 July 2020, Revised 13 April 2022, Accepted 14 April 2022, Available online 28 April 2022, Version of Record 7 May 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117265