Reinforcement Learning (RL) algorithms have been successfully applied to real world situations like illegal smuggling, poaching, deforestation, climate change, airport security, etc. These scenarios can be framed as Stackelberg security games (SSGs) where defenders and attackers compete to control target resources. The algorithm's competency is assessed by which agent is controlling the targets. This review investigates modeling of SSGs in RL with a focus on possible improvements of target representations in RL algorithms.

翻译：强化学习算法已成功地应用于诸如非法走私、偷猎、砍伐森林、气候变化、机场安全等现实世界局势。这些假设情景可被描述为维权者和袭击者竞相控制目标资源的斯达克尔伯格安全游戏(SSG ) 。算法的能力由控制目标的代理商评估。本审查调查了在RL建立SSG模型的情况,重点是可能改进RL算法的目标表述。