Abstract: Luminal diseases are the most common human ailments. Flexible surgical robots, owing to their compliance, can intervene in tortuous and narrow luminal pathways, yet they are prone to be obstructed due to contact interactions. To enhance the safety and intelligence level of intervention, an embodied perception method based on fiber Bragg grating (FBG) is proposed, enabling the robot to transit from ontology perception to cognition through the acquisition, processing, interpretation, and behavioral decision-making of morphological information. Firstly, shape reconstruction of multi-core FBG optical fibers is achieved based on the discrete arc assumption. Then, an improved moving average filter (IMAF) method is proposed by progressively extracting the influence of neighboring arc states, to smooth signal noise. Additionally, a blockage detection method based on spatiotemporal variation is proposed, which interprets shapes through dynamic interaction with the environment. Using the interaction states derived from this understanding, an intermittent intervention strategy based on human-like operation and the instantaneous global inconsistency index (IGII), is adopted to ensure compliant intervention. Finally, simulations and physical experiments are conducted, demonstrating the superiority of IMAF by comparing with different filtering methods. Furthermore, the feasibility and effectiveness of the blockage detection and intervention methods are validated in various scenarios.