Embodied agents can now navigate to target objects without prior training using a novel self-evolving framework called EvolveNav. This approach addresses the limitations of existing methods, which rely on static priors and lack adaptation, leading to repeated errors. EvolveNav enables continuous test-time adaptation, allowing agents to learn from their environment and improve their navigation skills. By integrating proactive preflection and self-evolving memory, EvolveNav enhances the agent's ability to explore and locate target objects efficiently. This framework has significant implications for zero-shot object-goal navigation, as it can reduce the need for costly trial and error1. The development of EvolveNav matters to practitioners because it can be applied to various real-world scenarios, such as robotics and autonomous systems, where adaptability and efficiency are crucial.