Dynamic Shielding for Reinforcement Learning in Black-Box Environments
Dynamic Shielding for Reinforcement Learning in Black-Box Environments
复制标题
黑盒环境中强化学习的动态屏蔽
DOI:
10.1007/978-3-031-19992-9_2
复制
发表时间:
2022
期刊:
影响因子:
--
通讯作者:
Hasuo Ichiro
中科院分区:
文献类型:
--
作者:
Waga Masaki;Castellano Ezequiel;Pruekprasert Sasinee;Klikovits Stefan;Takisaka Toru;Hasuo Ichiro
It is challenging to use reinforcement learning (RL) in cyber-physical systems due to the lack of safety guarantees during learning. Although there have been various proposals to reduce undesired behaviors during learning, most of these techniques require prior system knowledge, and their applicability is limited. This paper aims to reduce undesired behaviors during learning without requiringanyprior system knowledge. We proposedynamic shielding: an extension of a model-based safe RL technique calledshieldingusingautomata learning. The dynamic shielding technique constructs an approximate system model in parallel with RL using a variant of the RPNI algorithm and suppresses undesired explorations due to the shield constructed from the learned model. Through this combination, potentially unsafe actions can be foreseen before the agent experiences them. Experiments show that our dynamic shield significantly decreases the number of undesired events during training.
登录
查看更多内容
DOI:
10.1007/11901914_11
发表时间:
2006
期刊:
--
影响因子:
--
作者:
O. Kupferman;Robby Lampert
通讯作者:
Robby Lampert
DOI:
--
发表时间:
2019-04
期刊:
ArXiv
影响因子:
--
作者:
Maxime Bouton;J. Karlsson;A. Nakhaei;K. Fujimura;Mykel J. Kochenderfer;Jana Tumova
通讯作者:
Maxime Bouton;J. Karlsson;A. Nakhaei;K. Fujimura;Mykel J. Kochenderfer;Jana Tumova
DOI:
--
发表时间:
2019
期刊:
American Control Conference
影响因子:
--
作者:
Suda Bharadwaj;R. Bloem;Rayna Dimitrova;Bettina Könighofer;U. Topcu
通讯作者:
U. Topcu
DOI:
10.1007/978-3-662-48395-4_4
发表时间:
2016
期刊:
--
影响因子:
--
作者:
Damián López;P. García
通讯作者:
P. García
DOI:
--
发表时间:
2015
期刊:
International Conference on Tools and Algorithms for Construction and Analysis of Systems
影响因子:
--
作者:
R. Bloem;Bettina Könighofer;Robert Könighofer;Chao Wang
通讯作者:
Chao Wang