NeRC: Neural Ranging Correction through Differentiable Moving Horizon Location Estimation

The introduction of the Neural Ranging Correction (NeRC) framework marks a significant advancement in GNSS localization, particularly in urban environments where satellite signal propagation is complex and often leads to inaccuracies. Traditional methods have struggled with these challenges, primarily due to the difficulties in annotating ranging errors. NeRC addresses this by employing a data-driven approach that utilizes easily obtainable ground-truth locations for training, thus alleviating the burden of requiring extensive labeled data. This framework leverages differentiable moving horizon location estimation to optimize positioning and backpropagate gradients effectively. The end-to-end learning paradigm, supported by Euclidean Distance Field cost maps, further enhances the training process. Evaluations on public benchmarks and collected datasets demonstrate NeRC's distinguished improvement in positioning accuracy, making it a promising solution for real-time deployment on mobile…