route planning system design

To address inflexible road network interaction and difficulties in multi-alternative comparison within traffic planning platforms, this paper proposes an interactive traffic planning decision support system. The method introduces a historical state tree and a road state matrix to enable direct map-based editing, visualizable modification tracking, and parallel comparative analysis of planning alternatives. Integrating interactive map editing, dynamic traffic simulation, versioned state management, and matrix-based comparative visualization, the system enhances both interpretability and decision-making efficiency. Empirical validation is conducted using the Braess paradox case and the Sioux Falls network; expert evaluation confirms that the system effectively supports collaborative analysis and evidence-based decision-making in complex transportation planning scenarios.

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.

Existing path planning methods for mobile robot fleets in warehouse logistics suffer from low efficiency, deadlock susceptibility, and difficulty balancing geometric accuracy with operational constraints. Method: This paper proposes a demand-driven automatic roadmap generation method in continuous space. It jointly models inter-station transportation demand and minimum safety distance constraints within the continuous domain, followed by free-space discretization, K-shortest-path optimization, and trajectory smoothing to construct a compact, low-redundancy roadmap with high geometric fidelity and compliance with real-world kinematic and safety constraints. Contribution/Results: Experiments across diverse warehouse scenarios demonstrate that the generated roadmap yields near-optimal paths and significantly improves scheduling efficiency compared to 4-/8-connected grid and random sampling approaches. System throughput increases markedly, and robustness—particularly against deadlocks and congestion—is substantially enhanced.

Traditional vehicle routing approaches struggle to model the semantic, multi-step tasks and dynamic contextual constraints inherent in human driving—such as urgent requests or user preferences—while their multi-objective optimization relies on labor-intensive parameter tuning and lacks interpretability. This paper proposes PAVe, the first framework that tightly integrates large language model (LLM) agents with the classical multi-objective Dijkstra algorithm: the LLM handles intent parsing, hierarchical task decomposition, and dynamic constraint reasoning, while the pathfinding algorithm generates candidate routes; integration with a city-scale POI geocached knowledge base enables context-aware decision-making. PAVe thus shifts the paradigm from “shortest-path” to “intent-driven routing.” Evaluated on real-world urban benchmarks, it achieves an 88.3% accuracy in initial route selection, significantly enhancing both personalization and explainability.

This study addresses the problem of selecting operational subnetworks for autonomous mobility-on-demand (AMoD) systems in urban road networks, requiring joint optimization of infrastructure deployment and fleet scheduling to meet service quality requirements. The authors formulate subnetwork selection and passenger routing as a strategic network design problem and propose a path-based mixed-integer programming model. To efficiently solve large-scale instances, they develop a column generation algorithm. The key innovation lies in the first integrated optimization of AMoD infrastructure location and fleet control, supporting path-level constraints—such as turn restrictions—and robust optimization under box uncertainty. Experiments using real-world data from Manhattan demonstrate that the approach yields stable, interpretable operational subnetworks and effectively quantifies the trade-off between infrastructure investment and vehicle utilization time.

This study addresses the challenge of real-time integration and interpretation of heterogeneous, multi-source data in Transportation Systems Management and Operations (TSMO). It presents the first systematic review of large language models (LLMs) and multimodal large language models (MM-LLMs) applied to transportation supply, demand, and decision support, employing the PRISMA framework for literature selection. The work emphasizes the capacity of these models—particularly MM-LLMs—to serve as a decision-support layer for operators by fusing structured and unstructured information. Findings highlight MM-LLMs’ strong potential in integrating textual, visual, and sensor data, while identifying critical bottlenecks in multimodal fusion, real-time inference, and model interpretability. The paper concludes by proposing future research directions, including localized model adaptation, edge deployment, and cross-agency collaboration.

This work addresses the challenge of effectively transforming heterogeneous urban transportation data into actionable management intelligence, hindered by the absence of a reliable pathway from behavioral evidence to decision support. To bridge this gap, the authors propose a behavior-centered closed-loop framework that integrates travel records and passenger-generated text as behavioral evidence. Leveraging AI-driven inference, behavioral modeling, anomaly detection, and risk-aware mining techniques, the framework establishes a unified pipeline from raw data input through to decision support and governance feedback. Designed with deployment prerequisites such as privacy preservation, fairness, interpretability, and human accountability, the approach has been successfully applied to tasks including bus arrival prediction, taxi demand forecasting, anomaly identification, and risk perception, significantly enhancing service reliability, planning accuracy, regulatory effectiveness, and passenger experience.

This study addresses the challenge faced by public transit agencies in simultaneously providing immediate confirmation for pre-booked on-demand ride requests and maintaining continuous route optimization. To this end, the authors propose a novel dynamic vehicle routing approach that, for the first time, unifies real-time service feasibility verification with long-term scheduling optimization within a single framework. The method integrates a reinforcement learning–based non-myopic objective function with fast insertion heuristics and anytime optimization algorithms. Experimental evaluation on a real-world U.S. microtransit dataset demonstrates that the proposed approach significantly increases the number of served requests while meeting stringent requirements for immediate response and high service reliability.

This study addresses the challenge of efficiently identifying and evaluating high-quality improvement options in railway traction unit scheduling, where the vast combinatorial space of crossover operation sequences renders existing methods ineffective. To overcome this limitation, the authors propose an interactive optimization framework that integrates scheduling visualization, multi-objective simulation-based evaluation, and a three-tiered guidance mechanism. The approach aggregates candidate crossover schemes through spatial clustering and KPI-driven ranking, embeds simulation outcomes directly into the planning view, and enables nonlinear exploration of modifications grounded in historical provenance. Empirical results demonstrate that the proposed method substantially reduces both the time and manual effort required to discover and validate high-quality scheduling adjustments in real-world operational scenarios.