Recently, great success has been achieved in the visual object detection, which is one of the most important tasks in the area of computer vision. However, most methods are developed to deal with well-captured images in a static environment. In real robotic applications, there are many practical problems such as object scale, occlusion, and viewing direction, which severely hinder the direct application of existing object detectors. Therefore, it is necessary for the robot to have the ability to actively change its viewpoints until the target object is discovered. In this chapter, the active visual object discovery problem is introduced, and a deep Q-learning network (DQN) with dueling architecture is proposed to generate actions for the robot to actively explore the environment. A case study is also conducted to illustrate the active visual object discovery problem and the effectiveness of the proposed method.