In this work we address the robotic visual semantic navigation problem from a bio-inspired learning perspective, motivated by the ability of natural agents to rapidly adapt their behavior by exploiting prior experience. We explore meta-imitation learning as an artificial counterpart of biological learning-to-learn mechanisms, enabling fast adaptation to new tasks from limited demonstrations. Technically, we propose the model Meta Visual Semantic Navigation (MetaNav), a meta-imitation learning framework that combines imitation learning from human demonstrations with meta-learning to acquire adaptable navigation policies. Rather than optimizing for a single task, the model learns an adaptive prior that can be efficiently specialized to new object-goal navigation tasks using only a few examples. The approach is trained on a small set of environments from the HM3D dataset following a task-based formulation inspired by natural adaptation across related scenarios. Experimental results show that the learned policies exhibit fast adaptation to previously unseen tasks, highlighting the potential of bio-inspired meta-learning for robotic navigation, while also revealing current limitations and directions for future research.