Autonomous nano–unmanned aerial vehicles (UAVs) are increasingly deployed in cluttered, GPS-denied environments where compactness, maneuverability, and limited onboard sensing impose stringent constraints on navigation and control. Although deep learning–based policies have demonstrated promising performance in mapless navigation and target-driven flight, most existing approaches rely on offline training under the assumption that deployment conditions remain within the training distribution. In operational scenarios characterized by structural damage, perceptual aliasing, and environmental uncertainty, this assumption is frequently violated, leading to misclassification events, collisions, and mission failure. Continual learning (CL) offers a principled mechanism for online adaptation, yet its integration into resource-constrained nano-UAV swarms remains insufficiently explored, particularly in the presence of collision-driven updates and inter-agent knowledge transfer.