The area of machine perception especially computer vision has attracted a lot of attention over years, and great progress has been made with the booming of deep learning techniques. However, the robot is faced with unstructured environment that is very different from current well-defined machine perception problems. Existing machine perception solutions cannot be directly applied in robotics. In the active perception, the robot can actively move to adapt to the dynamic environment collecting useful perception information, which provides an efficient approach to bridge the gap between robotics and machine perception. Besides, intelligent robots today are accessible to multimodal sensory information rather than just single modality. Therefore, it is necessary to investigate the specific problem of multimodal robotic intelligence leveraging active perception approaches.