Overview of Reinforcement Learning Based on Value and Policy

Reinforcement learning methods are mainly divided into two categories based on value functions and policies. This article systematically introduces and summarizes reinforcement learning methods from these two categories. First, it summarizes the reinforcement learning methods based on value functions, including classic Q-learning, DQN, and effective improvement methods based on DQN. Then it introduces policy-based reinforcement learning methods, including policy gradient, policy optimization, actor critic, and their improvements. Finally, the frontier research and applications of reinforcement learning is summarized.