It is important for the mobile robot to have the ability to actively explore the environment based on its understanding of the observation. However, existing visual scene understanding tasks mainly focus on analyzing images passively, and the semantic understanding of the scenario is separated from the interaction with the environment, which is difficult to be applied to robotics directly. In this chapter, we propose a novel problem of active visual scene understanding, in which the robot actively explores the environment to find an optimal viewpoint for visual scene understanding. A learning framework with the paradigms of imitation learning and reinforcement learning is established to generate proper movements to guide the robot to actively explore the environment for visual scene understanding. Case studies are conducted on both the AI2THOR dataset and a real-world mobile robotic platform, demonstrating the effectiveness of the proposed framework.