The core of active perception lies in the integration of perception and action. Therefore, employing machine learning to generate action policies is a crucial approach for realizing active perception. In this chapter, we will first review some fundamental concepts of the Markov decision process (MDP), which forms the foundation of reinforcement learning (RL). And then, a detailed overview of several relevant algorithms is presented, including value function-based methods, policy-based methods, the actor-critic method, proximal policy optimization, and imitation learning. The value function-based and policy-based methods are considered classical algorithms, and they can be integrated to create the actor-critic method, which is currently a dominant framework in reinforcement learning. Proximal policy optimization is recognized as a widely utilized and effective algorithm. Imitation learning is a powerful tool proposed to tackle the inefficiencies and other shortcomings of reinforcement learning.