Proposed Architecture & Comprehensive study of employment of Deep Reinforcement Learning in Military Applications

This paper focuses on comprehensive study of employment of Deep Reinforcement Learning in Military Application using efficient methods as per the environment and application for making application more efficient, faster, accurate in real time scenarios. DRL adds different dimension to military application implementing various methods as per the condition and limitation of the environment such as Model based learning where prior situation is known and model free learning where agent need no confined scenario. Other prominent methods like actor critic methods is implemented in robots, drones which gives stability to the model and performs dynamically in real time situations and can be used in policy based or value based methods Deep Q-Network (DQN), double DQN & Dueling DQN, A2C, A3C, DDPG, TD3 & SAC works in specific applications. In static environment basic algorithms can be applied but in real time dynamic situation AC3 and other specific algorithms can be implemented. Paper discusses the limitations, future dimensions of DRL in military application.