Recently, substantial progress has been made in the machine perception, particularly computer vision, largely due to the advancements in deep learning techniques. However, robots often operate in unstructured environments, which differ greatly from the well-defined problems typically addressed in computer vision. Consequently, many existing computer vision solutions are not directly applicable to robotics. Additionally, modern intelligent robots have access to multi-modal sensory information, rather than relying on a single modality. Therefore, it is essential to explore the specific challenges of multi-modal robotic intelligence.

A key requirement for all intelligent robots is the capability of active perception. Active perception is an effective approach to bridge the gap between robotics and machine perception. Intelligence emerges when the robot actively interacts with the environment, embodying a close coupling of perception and action within a continuous feedback loop. Perception guided action, and each movement generate information that informs subsequent movements. With a variety of available multi-modal sensory information, it is crucial for robots to leverage active perception techniques to achieve multi-modal robotic intelligence.