Deep Reinforcement Learning: Key Algorithms, Applications, and Future Challenges

Abstract

Deep Reinforcement Learning (DRL) is a field of study that is growing very fast and brings together reinforcement learning and deep learning for agents who learn complex behaviors from observations from the environment. This integration enables DRL to solve tasks that before could not be solved with ordinary RL techniques. DRL has proven itself across multiple categories including robotics, gaming including AlphaGo and OpenAI Five, finance including sophisticated trading strategies and portfolio management, healthcare including treatment planning and interpretation of medical images and autonomous vehicles including decision making and path planning. The primary goals of this survey are introducing DRL and its fundamental ideas, reviewing major algorithms and methods in DRL, describing advanced research topics and methods, and exhibiting versatile applications. Furthermore, the survey intends to establish current contradictions and recommend further work and research agendas to enhance the study of DRL. In this way, this survey would be useful for researchers and practitioners who are interested in studying and advancing the field of DRL.