Efficient route planning in dynamic road networks remains a critical challenge for intelligent transportation systems. While traditional algorithms offer optimal solutions for static scenarios, they fail to adapt to rapidly changing traffic conditions. This paper introduces a novel meta-transfer routing framework that enables effective zero-shot generalization across diverse road networks without retraining. Our approach leverages hierarchical graph embeddings to capture transferable topological patterns and employs contrastive learning to develop network-invariant representations. Through a meta-learning framework with episodic training, we develop a model that quickly adapts to new environments and dynamic conditions. Extensive evaluation on synthetic and real-world datasets demonstrates that our method outperforms both classical algorithms and recent learning-based approaches, achieving up to 22% reduction in travel time compared to time-dependent Dijkstra and maintaining 95% of optimal performance when transferred to completely unseen road networks. Moreover, our approach shows remarkable resilience to network perturbations and congestion patterns, establishing a new paradigm for adaptive routing in dynamic environments.