route encoding

Existing short-term route prediction methods often neglect the influence of road network topology on human navigation decisions. This paper proposes the first knowledge graph–based framework for road network path prediction, explicitly integrating road topology and directional movement relationships to model spatially structured decision-making mechanisms. Methodologically, it innovatively unifies knowledge graph embedding, graph neural networks, and trajectory sequence modeling. We design a novel n-ary tree–based batch generation algorithm for top-K path enumeration and a ranking fine-tuning module, jointly preserving structural expressiveness while enhancing computational efficiency. Evaluated on real-world vehicle trajectory datasets from Chengdu and Shanghai, our approach achieves 12.7%–19.3% higher prediction accuracy over state-of-the-art baselines and supports real-time, link-level traffic flow simulation.

Existing trajectory generation methods employ limited semantic modeling, failing to jointly capture multi-scale spatiotemporal semantics—including road segments, geographic regions, individual trajectory points, and holistic trajectories—resulting in synthetic trajectories with insufficient realism and naturalness. To address this, we propose HOSER, a Hierarchical Omni-Scale Embedding Representation framework that unifies cross-granularity semantic modeling: from microscopic road nodes to macroscopic regions, and from local trajectory segments to global paths. Furthermore, we design a destination-guided navigator that synergistically integrates graph neural networks with sequential modeling to enable high-fidelity, origin-destination (OD)- and departure-time-conditioned trajectory generation. Evaluated on three real-world datasets, HOSER consistently outperforms state-of-the-art methods. Notably, it demonstrates robust generalization under few-shot and zero-shot settings, maintaining strong performance even with scarce or no task-specific training data.

This paper addresses the problems of representational redundancy and computational inefficiency in temporal path queries over dynamic networks. To tackle these challenges, we propose a compact temporal relationship modeling framework based on interval encoding, specifically designed for spatiotemporal navigation scenarios such as traffic flow and disease propagation. Our approach integrates temporal graph models with Allen’s interval algebra, treating temporal triples—comprising source, target, and their validity time interval—as fundamental units to support both path querying and temporal relation composition. We innovatively design four interval encoding schemes; under the dense-time assumption, the optimal scheme guarantees finite, unique, and succinct representations of query results. Experimental evaluation demonstrates systematic trade-offs among the encodings in terms of storage overhead and query efficiency, achieving significant improvements in both compactness and computability of temporal path queries.

This work addresses the challenge of efficient shortest-path planning in scenarios where real-world trajectory data are scarce, by leveraging simulators that exhibit systematic biases. The authors propose a graph Laplacian-regularized bias estimation method that integrates limited real observations, abundant synthetic data, and edge similarity structures within the road network to model smooth simulator-to-reality discrepancies. They establish theoretical guarantees on path suboptimality and devise an active learning strategy applicable even in the absence of initial real-world data. Through finite-sample error analysis and experiments on road networks across multiple cities, the approach demonstrates its ability to closely approximate optimal paths with only a small amount of real data, while providing computable performance certificates.

Handcrafted heuristic functions in search-based navigation suffer from poor generalization across unseen maps and long-distance paths. Method: This paper proposes a local heuristic learning framework that explicitly defines and end-to-end learns either heuristic bias correction or local cost estimation within a spatial neighborhood—replacing conventional global heuristic modeling. By decomposing complex global prediction into lightweight local regression, the approach significantly reduces learning complexity. Integrated with graph search algorithms (e.g., A*), it operates under supervised learning using local state inputs while preserving bounded suboptimality guarantees. Contribution/Results: Experiments demonstrate 2–20× reduction in node expansions, improved training efficiency, and robust generalization to both unseen maps and long-range trajectories—without compromising solution quality or theoretical guarantees.

This work addresses the limitation of traditional Search Trajectory Networks (STNs) in high-dimensional or continuous settings, where reliance on raw spatial discretization hinders applicability. The authors propose the first extension of STNs into semantic space by constructing semantic representations from model prediction vectors over a fixed set of samples. Semantic vectors are then discretized and aggregated into nodes via agglomerative clustering based on normalized Hamming distance, enabling unified visualization and comparison of training dynamics across diverse learning algorithms. Applying this framework reveals systematic structural differences between networks trained on real versus random labels: STNs derived from real labels exhibit denser, more efficient, and centralized topologies, effectively capturing the relationship between neural network generalization and the underlying semantic structure of data.

This work addresses the challenge of achieving scalable and robust ego-vehicle trajectory prediction without reliance on high-definition maps. To this end, we propose an end-to-end system that integrates front-view images, vehicle kinematics, and navigation paths derived from standard-definition (SD) maps. We introduce SD map paths as a semantic prior for trajectory prediction for the first time, and design a dual-hypothesis fusion architecture with a gated classifier to handle challenges such as route corruption or visual ambiguity. Evaluated on 480,000 real-world driving scenarios spanning ten European countries and the United States, our method reduces the average displacement error (ADE) over an 8-second horizon by 16.9% compared to a baseline using only images and kinematics. We also release an open-source toolkit for SD path generation to support community benchmarking.

This work addresses the challenge of efficiently supporting dynamic or user-defined cost metrics in large-scale road network routing under high concurrency, a task hindered by limitations in path storage and reconstruction efficiency. Building upon the Customizable Transit Labeling (CTL) framework, the study proposes an optimized path storage structure and reconstruction mechanism, along with the first batch-processing strategy tailored for CTL that significantly boosts performance by sharing computations across multiple queries. The authors further design several algorithmic variants to balance memory usage and query performance and adopt a decoupled architecture separating graph preprocessing from customization. Extensive experiments on 13 real-world road networks demonstrate that the proposed method substantially outperforms the state-of-the-art in both query speed and memory efficiency, scaling effectively to scenarios with millions of concurrent queries.

This study addresses the finite-time vehicle routing problem (FTVRP) with the objective of maximizing the number of served customers. To this end, the authors propose a Markov decision process formulation that integrates node features, edge features, and adjacency structure, along with a novel context-aware network embedding module that explicitly incorporates remaining time information into both global graph representations and local node embeddings. A policy gradient-based deep reinforcement learning framework is developed to enable efficient solution of the problem. Experimental results demonstrate that the proposed approach significantly improves customer service rates on both real-world and synthetic road networks while achieving faster solution times compared to existing methods.

This work addresses the reliance of existing public transit routing systems on structured maps and complex routing engines by introducing TransitLM, a large-scale dataset comprising over 13 million transit records from four Chinese cities. It presents the first fully data-driven, end-to-end approach to bus route generation that operates without explicit map dependencies. Through continued pretraining of large language models on this dataset, the model implicitly aligns arbitrary GPS coordinates to transit stops, eliminating the need for explicit geospatial infrastructure. Experimental results demonstrate that the proposed method achieves high accuracy in generating structurally valid routes across three complementary tasks, significantly advancing the state of intelligent transit planning under map-free conditions.