<p>Artificial intelligence plays a crucial role in enhancing the accuracy of positioning in the global navigation satellite system (GNSS). Although existing reinforcement learning-based positioning correction methods can effectively reduce positioning errors caused by challenging urban environments, they require a substantial amount of training data as a foundation and suffer significant performance losses when dealing with scenarios involving distribution shifts of training data. These issues severely restrict the widespread application of reinforcement learning-based positioning correction methods in the field of GNSS positioning. In this paper, we propose a positioning correction framework based on position-sequence-enhanced meta-reinforcement learning with adaptive fine-tuning. In the pretraining phase, the model is enabled to learn stable correction strategy from few-shot samples through observation representations enhanced by position sequences. During the fine-tuning phase, adaptive fine-tuning for specific target urban environments is performed by calculating the similarity among target urban environments. This is the first work using an effective GNSS positioning correction method to address the real-world few-shot and distribution shift using meta-reinforcement learning with adaptive fine-tuning. Our proposed method is evaluated on real-world data sets. The results demonstrated that our approach outperformed model-based methods and the state-of-the-art learning-based methods under the real-world few-shot and distribution shift urban environments, with an average improvement in positioning accuracy of 8%, as well as significantly enhanced model training stability.</p>