Navigational guidance – A deep learning approach

Abstract

This paper addresses the navigation problems facing many companies, including logistics companies, couriers, and Uber, helping users find the best route to multiple destinations in the shortest amount of time. Navigation problems involving multiple destinations are formulated in this paper as Directed Steiner Tree (DST) problems on directed graphs. We propose an end-to-end deep learning approach to tackle the DST problems in a supervised and non-autoregressive manner. The core of our approach is Graph Neural Networks (GNNs) in estimating whether a node belongs to the optimal solution. Experiments are conducted to evaluate the proposed approach, and the results suggest that using our approach can effectively solve the DST problems with at least 95.04% accuracy. Compared to solving DST problems using traditional methods, our approach significantly improves the solvability of DST problems with acceptable execution time. We further explore how our approach can be applied to different scenarios, such as large-scale graphs. Moreover, we show that our approach can be smoothly applied to solve several variants of the Steiner Tree problem, including Steiner Forest problems. In summary, the proposed approach shows promising results and can be implemented in practice. Research limitations and future directions are also discussed.