Reinforcement learning plays an important role in fields such as game AI, nuclear fusion control, chip design, industrial scheduling, recommendation systems, quantitative trading, matrix multiplication acceleration, and robot control. Since the process of reinforcement learning does not require pre-collected labeled data and can train agents that exceed human performance through continuous interaction with the environment, reinforcement learning is also considered the learning paradigm closest to artificial general intelligence (AGI). The design of reinforcement learning algorithms requires mastering the theoretical knowledge of reinforcement learning and combining it with extensive practical experience to formulate efficient solutions for different decision-making tasks. Next, we will introduce the design ideas for solving reinforcement learning tasks from three aspects: agent design, model design, and algorithm design.