Material perception is an extremely important but challenging problem for robotic intelligence. The primary challenges arise from the difficulty in comprehensively evaluating material properties with single perception modality such as visual, auditory, or tactile. The fusion of multimodal perception information can better reveal the intrinsic properties of the material. Conventional multimodal fusion methods usually require collecting all of the multimodal information before the recognition, which is a passive process that is expensive, redundant, and may incur large latency. To tackle this problem, a new active multimodal perception fusion framework for the material recognition is proposed in this chapter. To bridge the gap between different modalities, an adversarial learning method is firstly adopted to obtain the modal-invariant representations, and then a reinforcement learning method is developed for active perception modality selection. In the case study, the developed framework is evaluated on a publicly available dataset and promising material recognition results are observed.