RESEARCH PROGRESS ABOUT DEEP REINFORCEMENT LEARNING

DRL is a kind of decision theory, which is based on probability theory, combines machine learning with Turing machine and other powerful tools, and establishes a technology that uses strong attraction algorithm, deep neural network, statistical machine learning, etc. to optimise machine control and super solutions. This technology is a machine learning technology used to learn and optimise the use of policy technology through interaction with the environment, to guide human decision-making behaviour, and to solve many difficult and complex problems. It has a deep connection with robot research, and its analysis technology can help machine learning to carry out effective knowledge representation and give full play to the intelligence of the machine. Our manuscript mainly discusses the following issues: In the first place, the development of deep learning, reinforcement learning, and deep reinforcement learning (DRL) are reviewed; secondly, the main algorithms of DRL are introduced; thirdly, applications of DRL algorithm are summarised, such as application of DRL in game theory, robot control, computer vision, and task scheduling; and finally, the main research directions of DRL are prospected.