ABSTRACT. An efficient truck arrival management is crucial in minimizing congestion and adverse environmental effects in container terminal operations. A Truck Appointment System (TAS), which involves scheduling truck arrivals based on time slots, plays a role in balancing demand in ports. Nevertheless, the traditional TAS strategy may fail to be efficient enough. In this study, an optimization problem associated with the truck arrival time slot assignment model for container terminals involving single and dual transaction trucks, along with different truck capacities, is investigated. In the process, dual transaction trucks undertake pickup and delivery during the same trip, enabling the formulation to account for truck turnaround time (TTT) and capacity effectively. The objective is to assign appointment slots that smooth truck arrivals, avoid peak-time demand surges, and minimize total deviation. This entails minimizing deviation from demand as well as from the target proportion of dual transaction trucks. The problem is formulated as a Mixed-Integer Linear Programming (MILP) model and solved for a small instance. Results show that the model can serve as a practical decision-support tool for terminal operators by identifying the optimal number of trucks. The study also provides a basis for future extensions to larger, multi-terminal, and intermodal transport networks

ABSTRACT. Container yard pre-marshalling and retrieval present compounding operational challenges, where a shared RTG crane must serve time-sensitive truck appointments across multiple bays simultaneously. Inefficient sequencing leads to excessive relocations, delayed retrievals, and reduced terminal throughput. This paper introduces the Multi-Bay Container Relocation and Retrieval Problem (MBCRRP) and formulates it as a scalarized Multi-Objective Markov Decision Process (MOMDP), optimising jointly for relocation efficiency and schedule adherence under a shared crane time budget. The problem is solved using a Q-learning agent enhanced by two heuristic components: heuristic-guided Q-value initialisation that accelerates convergence, and a rule-constrained action space that enforces smart-relocation principles. A Unified Priority Labelling scheme assigns each container a calendar-relative urgency rank, enabling consistent decision-making across diverse yard configurations. A Rolling-Horizon Training protocol structures learning across consecutive planning days, with each day inheriting the yard state produced by the previous one, allowing pre-marshalling decisions to accumulate and benefit future retrievals. The framework is validated on operational data from a container terminal in Sydney, Australia. Results demonstrate high retrieval coverage and reliable service within narrow operational windows, outperforming benchmark heuristics in both relocation efficiency and schedule adherence.

ABSTRACT. Container terminals serve as key interfaces between maritime and inland logistics, and prolonged truck turnaround time (TAT) can reduce terminal efficiency and cause surrounding congestion. However, terminal-wide average TAT cannot fully explain delays that occur during peak periods or in specific yard blocks. While previous studies have mainly examined gate appointment systems, gate queues, or aggregate congestion, the internal yard-side mechanism after trucks enter the terminal has received less attention. This study develops a truck-level multi-agent discrete-event simulation model using terminal operating system (TOS) records from a container terminal in Tokyo. Truck operations are modeled as a sequence of gate-in, movement to the assigned yard block, yard-side handling, and gate-out. Each yard block is represented as an effective handling service station calibrated from observed records, with block-time workload incorporated as a delay factor. The calibrated baseline simulation was evaluated at both terminal-wide and yard-block levels. Figure 1 compares observed and simulated block-level mean TAT. The model achieves an R² of 0.8632, indicating that it captures the main spatial variation of truck delay across yard blocks. Most blocks are close to the 1:1 reference line and within the ±3 min band, while larger deviations appear in special-purpose or low-volume areas. These results suggest that the proposed simulation can reproduce not only aggregate terminal behavior but also block-level congestion heterogeneity. Figure 2 further validates the calibrated baseline from a temporal perspective by comparing observed and simulated terminal truck occupancy over the operating day. The simulation reproduces the main morning and afternoon peaks, as well as the mid-day decline in terminal occupancy. Although deviations remain during peak periods, the overall temporal pattern is reasonably captured. This supports using the calibrated baseline to evaluate how yard-side workload concentration affects truck delay. After baseline validation, the simulation was used to evaluate yard-side workload interference scenarios. Compared with the 26.02-minute baseline mean TAT, mean TAT decreases to 21.08 minutes in the no-interference scenario, 23.49 minutes in the reduced-interference scenario, and 24.88 minutes in the increased-capacity scenario. These results show that yard-side workload interference contributes to truck delay, especially when trucks concentrate in the same yard block and time period. The no-interference scenario provides an ideal lower bound, while reduced interference represents a practical improvement. The increased-capacity scenario also reduces TAT, but its effect is smaller, suggesting that truck delay is related to both handling capacity and the temporal-spatial concentration of yard operations. The proposed TOS-based multi-agent simulation framework can reproduce temporal truck occupancy patterns and block-level TAT variation. These results show that even a simplified block-level model can provide useful insights into how yard workload affects truck turnaround performance. Future work will incorporate AIS-based vessel arrival and berth-operation information to better examine the relationship between vessel-side workload and yard-side truck congestion.

ABSTRACT. Estuarine ports face persistent siltation challenges that compromise navigational safety and impose substantial dredging costs. This study presents a comparative spatio-temporal analysis of suspended sediment dynamics in the Hugli and Karnaphuli estuaries — home to Haldia Port (India) and Chittagong Port (Bangladesh) — using Sentinel-2 MSI imagery (10 m resolution) over 2020–2025. Normalised Difference Water Index (NDWI), Normalised Difference Turbidity Index (NDTI), Suspended Sediment Concentration (SSC), Turbidity Maximum Zone (TMZ) probability, and spectral signatures were computed across upstream, port, and estuary mouth zones using Google Colab. Results reveal that the Hugli estuary exhibits consistently higher turbidity than the Karnaphuli estuary across all zones. Haldia Port exhibits higher concentration of sediment suspension than Chittagong port. In the Hugli system, SSC increases downstream from upstream to estuary mouth, while the Karnaphuli records elevated upstream SSC driven by steeper catchment gradients. TMZ mapping identifies persistent high-turbidity conditions at both estuary mouths (probability >0.6). Seasonal analysis confirms pre-monsoon and monsoon as the peak turbidity period in Hugli and Karnaphuli systems respectively. Based on these patterns, maintenance dredging during winter (December–February) is recommended for both ports to maximise efficiency. This comparative framework offers a transferable approach for sediment monitoring in tropical estuarine ports.

ABSTRACT. The impact of increasing communication information on truck platoon operation remains unclear. To address this question, this paper adopts a leader–follower framework, establishes a longitudinal dynamic model, and improves a linear controller to adapt to different communication modes. Based on this framework, simulation experiments are conducted to quantitatively investigate the effects of Vehicle-to-Vehicle (V2V) communication topology and information content on the energy consumption of connected freight vehicle platoons. The results show that communication topology and communication information should not be considered independently. When acceleration is not included in the communication content, introducing leader vehicle information does not significantly reduce energy consumption, but it effectively suppresses error propagation and reduces acceleration fluctuations. In contrast, the bidirectional topology shows the opposite trend, with the mean acceleration standard deviation increasing by 6.3%. When acceleration information is included, directly introducing acceleration signals in topologies with two communication targets can trigger high-frequency oscillations, leading to a 20% to 30% increase in energy consumption compared with the scenario without acceleration information.

ABSTRACT. This study develops a methodology for estimating the operational structure of container trailers by integrating ETC2.0 probe data with Terminal Operation System (TOS) data, explicitly accounting for specialized cargo handling behaviors. While conventional probe data analysis faces difficulties in identifying the sequence of cargo handling activities, by combining a state transition model—based on the attributes of stay points and vehicle movement—with dynamic programming (Viterbi algorithm), it becomes possible to estimate trailer states with high consistency. Validation results demonstrate the model’s robust generalization capability, successfully reproducing container Origin-Destination (OD) trip structures even for unknown datasets. Furthermore, a scenario analysis focused on the utilization of Inland Container Depots (ICD) was conducted using the estimation results. The analysis quantitatively illustrates that integrating ICDs could potentially reduce empty container transport distances by approximately 42%. Departing from traditional reliance on manual questionnaire surveys, this methodology provides a digital foundation for objectively and continuously capturing the actual operational status of container trailers. The findings of this study offer significant value for the development and evaluation of port logistics policies.

ABSTRACT. Although primarily designed for passenger transport, the taxi industry possesses spare capacity, as one taxi often carries only one person. This underutilized resource can improve the efficiency of the freight market. This paper constructs a taxi market with passenger and freight transport services. In this market, some taxi passengers inclined to travel alone finish their trips as traditional taxi passengers, while the others amenable to travel and receive some economic compensation carry some cargo items in their trips. The passengers carrying cargo items may redesign their trajectories to receive the cargo items and convey them, thus they have additional time cost and compensation in their generalized cost. In the process, the passengers determine their modes (traditional passengers or passengers carrying cargo items) and routes based on their generalized costs. Meanwhile, the vacant taxis are taken into consideration in the market, involving their route choices. A variational equality (VI) is proposed to describe the equilibrium of this taxi market. At the equilibrium, passengers have their minimal generalized cost and no passenger can reduce his/her generalized cost by changing the mode or the route. So it is the same with the vacant taxis. The constraints include (1) all taxis meet the demand, (2) the capacity of total freight transport taxis between an OD pair is larger than the cargo item delivery demand to guarantee that all cargo items can be delivered. A successive averages and diagonalization algorithm is designated to solve the VI to capture the demand and the flow patterns effectively. The VI model and algorithm are applied to a simple network. The results fall in a reasonable range, demonstrating the feasibility and effectiveness of the model and algorithm. Then we proceed on the sensitivity analysis. The results show that the total cost exhibits non-linear changes with increasing freight handing cost. If the cost is excessively small, there is no impact on total cost. Then the cost increases sharply from approximately 2.5 to 3, and fluctuates. This is related to the travellers choices based on the generalized cost. In terms of increasing ratio of operating cost for passenger transport to that for freight transport per unit distance, the total cost display descending tendency, but increases at the last higher values. This demonstrates that there exists optimal ratio for the whole system, because passenger and freight transport have their respective advantages for the efficiency. However, a larger taxi fleet brings higher total cost. This is because excessive taxis lead to more serious congestion on the network.

ABSTRACT. Extended Abstract: Fuel surcharges (FSCs) are a central mechanism through which U.S. Less‑Than‑Truckload (LTL) carriers adjust pricing in response to diesel market volatility. Although FSCs are intended to provide a transparent and formula‑based link between fuel costs and freight rates, carriers differ substantially in how they structure and update their FSC programs. These differences raise important questions about the degree of fuel price pass‑through, the responsiveness of FSCs to energy shocks, and the extent to which carriers’ pricing adjustments influence one another. This study provides a systematic analysis of fuel price spillovers and dynamic interconnectedness across major LTL carriers’ FSC programs.

Using weekly data from 2010–2024, we compile a panel of publicly posted FSC schedules from leading U.S. LTL carriers and match them with U.S. Energy Information Administration (EIA) diesel price data. To capture evolving competitive behavior and time‑varying responsiveness to fuel shocks, we employ a Time‑Varying Parameter Vector Autoregression (TVP‑VAR) with stochastic volatility. We then apply the Diebold–Yilmaz connectedness framework to quantify total, directional, and net spillovers among carriers’ FSC adjustments.

Our results reveal three key findings. First, fuel price pass‑through is incomplete and highly time‑varying. FSC responsiveness increases sharply during periods of elevated diesel volatility—most notably during the 2011 fuel spike, the COVID‑19 recovery, and the 2021–2022 diesel surge. Second, we document significant cross‑carrier spillovers in FSC adjustments, indicating that carriers do not update FSCs in isolation. Instead, pricing behavior becomes more interconnected when fuel markets are unstable. Third, the network structure of FSC spillovers evolves over time, shifting from a relatively sparse configuration in stable fuel markets to a dense, highly interconnected network during fuel price crises. Several large national carriers consistently emerge as net transmitters of FSC shocks, while smaller or regional carriers tend to be net receivers.

These findings have important implications for freight pricing strategy, shipper contracting, and regulatory oversight. For carriers, understanding spillover dynamics can improve margin management and competitive positioning. For shippers, the results highlight that FSCs—though formula‑based—are influenced by broader market behavior. For policymakers, periods of heightened interconnectedness may warrant closer monitoring of pricing transparency and competitive conduct.

Overall, this study contributes to transportation economics by linking energy price volatility, competitive freight pricing, and dynamic network spillovers within a unified empirical framework. It also demonstrates the value of TVP‑VAR models for analyzing evolving market behavior in freight transportation.

(Full paper is in process. Presentation slides are uploaded)

ABSTRACT. Ports in many countries strive to embrace sustainable development for long-term wellbeing. In view of artificial intelligence (AI)’s rapid advancement, it has an immense potential to contribute to port sustainable development. This paper, by adopting a strategic technology scanning framework, synthesises the applications of AI technology in environmental sustainability of ports. The study illustrates how AI enables the achievement of Sustainable Development Goals through examples from various parts of the world. Findings show that AI technology is very versatile in meeting different requirements of ports in diverse applications and at a wide spectrum of scales. We highlight the potential and opportunities offered by these AI solutions in the port sector. Relevant future research directions are also recommended.

ABSTRACT. The main purpose of this paper is to assess the impact of new international services at Kobe Airport on its’ centrality indicators and those at the five major airports in Japan and South Korea from social network approach. The results reveal that, in the case of an undirected graph including the five major airports in Japan and South Korea and all airports connected to them by routes, the four centrality indicators (closeness centrality, degree centrality, eigenvector centrality and betweenness centrality) consistently show the highest values at Incheon International Airport throughout the 15-year period analyzed. In the case of a directed graph representing passengers departing from Japan and traveling to overseas destinations via one of the five major airports in Japan and South Korea, a trend broadly similar to that observed in the undirected graph was also identified. Furthermore, closeness and eigenvector centralities at Kobe Airport have increased to a considerable degree and Betweenness centrality at Incheon International Airport has slightly increased by adding Kobe Airport as a new spoke.

ABSTRACT. Abstract Several strategies can be adopted to improve operational freight efficiencies. Three review strategies are: examining actual operating network design, fleet restructuring, and subjecting a new network and/or fleet to a level of optimization. The paper’s focus is what ‘resource’ savings are potentially available if a operational review has not been undertaken for several years. The analysis examined 23 case studies spanning a 20 year examination period in which the time scales are examined preceding the implementation of an efficiency strategy is implemented. The analysis examines what ‘resources’ are saved when those, efficiency strategies are implemented. Network efficiencies gained through network restructures are not often reviewed. In many cases as this is not possible for small operators. Fleet restructures which also generates efficiencies usually happen as new vehicles become available and older ones retired. Aircraft, rail locomotives, and container ships have much longer replacement times than road freight vehicles, which is this paper’s focus. The third efficiency tool, optimization”, is often available for large companies working complex networks but optimization will deliver efficiencies. Target optimization problems examined are often for established networks with existing fleets and limited vehicle types. Using even one of the above strategies can achieve resource savings from a few percent to potentially forty percent if reviews are not undertaken across a 20 year time frame.

1 Introduction – three efficiency strategies. The paper re-examines a three efficiency strategy approach to deliver significant freight efficiencies. Generally the focus is on road and to a lesser extent road to rail operations, although in theory both air and maritime operations could consider one or more of these strategies. The application and impacts are examined through 23 case studies that spanned a 20 year period.

1.1 Rethinking networks

There has been very little academic analysis as to what makes a good or a bad freight network. There are some well known network structures such as ‘hub and spoke’, corridors, cross dock Hubbing as examples. Some twelve networks shapes are examined., Many specially constructed networks lead to efficiencies and some less so. Several network options focus on urban operations supplemented with regional network options and some long distance transformations. In some cases bi-modal rod/rail terminals can generate efficiencies not available to either a road only or rail only freight option.

1.2 Fleet Restructuring

The upgrading of a fleet is often a slow process based on what new vehicle types become available across operational time frames. For large vehicles such as aircraft and container and bulk ships timeframes are often long often 25 years or longer. However, even large urban and regional truck fleets often upgrade in shorter time periods ranging from ten to fifteen years. However, this can be done more expeditiously if higher capacity vehicles become available within this usual replacement period.

Because road freight vehicles have shorter economic lives than rail locomotives, aircraft, or large seagoing vessels there is greater opportunity to upgrade road vehicles to higher greater capacity vehicles that may often be more fuel efficient. In Australia the introduction of High Productivity Vehicles through the Performance Based Standards (PBS) has allowed the adoption of higher capacity road vehicles. (Austroads 2014,) Pre the approval of this scheme restructuring conventional vehicle fleets was also quite possible.(Hassall,2003) . 1.3 Operational Optimization

Optimization tools can deliver significant network and vehicle operational benefits. However, optimization tools are used more often by larger companies. Generally the focus examines a single vehicle type operating in a fixed network. This paper’s discussion will examine greater benefits becoming available by incorporating network and fleet restructuring.

1.4 Finding a relationship between time between efficiency strategy adoption and resource benefits The longer it has been between an efficiency review the greater the inefficiency buildup. The presented case studies demonstrate significant resource benefits when one or more efficiencies are adopted. The finding of the analysis shows a linear relationship between resource benefits and the time between the adopting one or more efficiency strategies.

ABSTRACT. The challenges of small-value, individual shipments in the dynamic cross-border e-commerce business, including international transport and simplified customs clearance. Consumers are satisfied with convenient purchasing systems, and much spending is being done through EC. Japan customs has just revised its laws for NACCS system and started simplified customs clearance for ocean freight last year. Digital forwarders are coming on the scene, providing information like dashboards and visualizing logistics through connectivity with customs, airlines, shipping companies, and trucking companies. The wave of DX is now reaching the global logistics industry. This report details the key features of EC player and digital forwarders and their approach to the market.

ABSTRACT. The increasing global maritime traffic has significantly heightened the complexity of managing maritime environments, posing unprecedented challenges for effective supervision and safety management. Current methods, such as rule-based analyses and statistical models, fall short in capturing the multidimensional and dynamic interactions that characterize maritime traffic, leaving a critical gap in accurately assessing traffic complexity. Addressing this gap is essential for improving navigational safety and optimizing resource allocation in increasingly congested maritime zones. To overcome these limitations, we propose a novel approach that integrates Conditional Generative Adversarial Networks (CGAN) with Minimum Redundancy Maximum Relevance-Extreme Gradient Boosting (mRMR-XGBoost). This methodology effectively generates synthetic data that mirrors real-world maritime traffic scenarios, mitigating the issue of insufficient labeled data. The mRMR-XGBoost model subsequently identifies key features from the enriched dataset, significantly enhancing the accuracy of complexity evaluations. When applied to the congested waters of the Zhenjiang Branch Yangtze River, our method demonstrated a substantial improvement in predictive performance over traditional models. This advancement not only fills a critical gap in maritime traffic management but also paves the way for more reliable and efficient safety and supervision systems, thereby contributing to the broader field of intelligent maritime management.

ABSTRACT. As the primary mode of transportation for global trade, maritime trade is forecasted to grow by over 2% annually through 2028. However, the sector faces multiple challenges, from operational risks posed by complex sea conditions to disruptions caused by geopolitical tensions like the Russia-Ukraine war. The industry also faces a pressing sustainability challenge: it currently accounts for 3% of global emissions, a share that could be rocketed by 150-200% by 2050 without intervention. To ensure sustainable development and operational safety, addressing these issues requires digital transformation through technologies like autonomous shipping, virtual reality, and artificial intelligence. While vessel identification is a key tool for improving sustainable marine monitoring, port management, and safe navigation, current advancements are narrowly focused on YOLO variants and training data is primarily designed for Unmanned Surface Vehicles (USVs). In recent years, computer vision has progressed significantly with powerful foundation models like SAM, DINO, and CLIP. However, maritime object recognition has not incorporated these models, causing ship detection to fall behind other applications due to inconsistent methods, a lack of evaluation standards, and the absence of benchmarks. To bridge the gap, this study proposes an effective vessel semantic segmentation model that adapts the state-of-the-art SAM for maritime application. The model is trained with the recently released and expertly labelled Ocean AI Segmentation Initiatives (OASIs) dataset to specifically investigate the performance of foundation models on maritime tasks. The resulting model demonstrates excellent potential for real-world applications, showing a mean IoU of 98.30% for accuracy and an inference time of 38.75 FPS on a single NVIDIA P100 GPU.

ABSTRACT. This study proposes an AI-driven decision support framework for capacity allocation in airline logistics systems under shared payload constraints. In airline operations, passenger, baggage, and cargo demands compete for limited aircraft capacity, creating a complex resource allocation problem. However, existing approaches typically model these demand components independently, limiting their applicability for operational decision-making. To address this issue, this study develops a multi-stage forecasting framework that captures interdependencies among demand components through a cascading structure. Passenger demand is first predicted and subsequently used to estimate baggage and cargo demand. The forecasting outputs are then integrated into a payload feasibility model to evaluate whether total demand can be accommodated within aircraft capacity constraints. Empirical results indicate that incorporating demand interdependencies significantly improves baggage prediction, while cargo demand remains more complex due to external influencing factors. The payload analysis reveals substantial variability in capacity utilization across routes, ranging from 6.0% to 83.4%. This indicates the presence of both underutilized and near-capacity operations, highlighting opportunities for improving capacity allocation strategies. The proposed framework provides a practical approach for integrating predictive analytics with decision support, enabling more effective capacity allocation and resource planning in airline logistics systems.

ABSTRACT. This study improves a global maritime container shipping network simulation (GMCS-NS) model by incorporating observed flow data and machine-learning-based error analysis. First, service-level loading and unloading link flows at major US ports are constructed from PIERS data and compared with model outputs. Subsequently, the error structure is analyzed using gradient-boosting methods and permutation importance. The results indicate that the estimation error is closely related to the unrepresented cost structure and service schedule slack. Based on these findings, the model is enhanced by introducing generalized cost assignment, a stochastic distribution of value-of-time, and schedule-aware vessel speed settings. Sensitivity analysis is then conducted on the key parameters, and the final model reduces bias in transshipment estimation and improved the regression slope for service-level cargo handling shares, while keeping the corresponding correlation coefficients within a practically comparable range. The study demonstrates a framework for diagnosing assignment errors using observed flow data and translating the findings into model refinement.

ABSTRACT. Under the background of the rural revitalization strategy, promoting rural development and increasing farmers’ income through local food industries has become a focus of academic attention. This paper takes cherries as an example and uses the discrete choice experiment approach, combined with the mixed Logit model and latent class model, to analyze consumers' purchase preferences for local food and their heterogeneity. The study finds that in the overall preference structure of consumers, green certification, traceability, local origin, and the “help farmers” label all significantly enhance purchase intention, while the e-commerce channel exerts a significant negative effect on purchase intention. There are interaction effects among attributes, with the "help farmers" label and full-chain traceability information having a synergistic effect; and local origin and basic traceability information also showing a complementary interaction, but at the highest traceability level the marginal utility of the local attribute is crowded out. Based on the heterogeneity of preferences, consumers can be classified into three types: “safety-oriented”, “comprehensive-oriented”, and “local-oriented”. The research provides empirical evidence and policy implications for deepening the market segmentation of local food, optimizing the transmission of quality signals, and promoting the effective connection between rural characteristic industries and the consumer market.

ABSTRACT. This paper addresses the coordinated scheduling of monitoring pod deployment and retrieval in offshore aquaculture. We formulate a mixed-integer programming (MIP) model integrating vessel capacity, crew working hour limits, equipment reuse, minimum monitoring duration precedence, and time windows. To solve it exactly, we develop a branch-and-cut algorithm with three families of problem-specific valid inequalities, namely a minimum fleet size cut, precedence incompatibility cuts, and task--vehicle infeasibility cuts, which tighten the linear relaxation. On instances with up to 16 task nodes and 9 vehicle-cycles, the cuts reduce the average large-scale optimality gap from 87.39\% to 6.32\% relative to a pure branch-and-bound baseline. To complement the exact method, we design an enhanced adaptive large neighborhood search (E-ALNS) heuristic whose destroy-and-repair operators and feasibility-guided acceptance are tailored to the paired deployment-retrieval structure. Within a 60-second budget, the E-ALNS attains the proven optimum on every instance the exact method solves to optimality and returns strictly better solutions on the hardest instances, where the exact method cannot close the gap within 1200 seconds. The two methods are complementary: the exact algorithm certifies optimality and provides dual bounds, while the E-ALNS delivers high-quality solutions about an order of magnitude faster on large instances, together forming a practical toolkit for aquaculture monitoring mission planning.

ABSTRACT. This study employs a global logistics intermodal network simulation model to simulate cold-chain cargo flows in the Philippines and the impact of policy implementation. By categorizing cargo into dry, fresh, and frozen types, introducing distinct cost structures and time values for each type, and transport and low-temperature infrastructure conditions based on field surveys, the model explicitly represents the country's unique cold-chain transport constraints. After confirming the model's reproducibility on current port throughput and domestic flow patterns if applied to the national network, policy scenarios are analyzed. Results indicate that road and bridge improvements yield only segment-specific enhancements, with port usage shifts varying by cargo type. Cold-chain infrastructure development would promote the decentralization of dry cargo to regional ports and the concentration of fresh and frozen cargo at major ports, while strengthening domestic reefer transport could help stabilize inter-island shipments.

ABSTRACT. Cambodia is experiencing remarkable economic growth, with active port development driven by both the public sector and private capital. However, despite the Mekong River system traversing the country, the utilization of Inland Waterway Transport (IWT) remains limited. The government intended to promote the use of the Mekong River system for waterborne transport by identifying future nationwide bulk cargo demand and necessary investment levels, while facilitating appropriate modal choice adjustments. To this end, the authors collaborated with the Cambodian Ministry of Public Works and Transport (MPWT) to forecast nationwide demand for bulk cargo and estimate the volume of cargo generation and concentration across six designated zones in Cambodia. Furthermore, the study estimates the required number of bulk berths and terminals for each zone. It also assesses the shifts in bulk cargo flows resulting from the development of the Funan Techo Canal. This paper introduces the methodologies used for these estimations and the issues that arose during the estimation process (specifically, the limitations of current demand forecasting methods for bulk cargo). The authors hope that the study findings will serve as a resource for improving future demand estimation techniques.

ABSTRACT. Against the backdrop of heightened concerns over food safety and consumption upgrading, the merchandise selection in the food retailing industry is transitioning from a conventional mode to a stringent selection mode. This paper focuses on a food supply chain consisting of a manufacturer and a retailer, constructing Stackelberg game models under the two selection modes. The study shows that an increase in cleanliness-related costs inhibits the quality of stringently selected products and market demand. Furthermore, the stringent selection mode is not always superior; its advantage primarily depends on the relative magnitude of the retailer's inspection costs and the manufacturer's cleanliness investment costs. Moreover, this mode shifts quality decision-making authority to the retailer, thereby squeezing the manufacturer's profits. Retailers can only benefit from the transformation when their inspection costs are relatively low, while the manufacturer's cleanliness investment costs are relatively high.

ABSTRACT. Drone delivery is an important implementation mode of human-machine interaction services in the field of urban last-mile logistics.Current research lacks coordination with the existing urban instant e-commerce pre-position fulfillment warehouse (PFW) network, and is difficult to balance the multi-objective trade-offs among economic cost, service coverage capacity and airspace resource utilization efficiency simultaneously.This study aims to explore an optimization method for constructing an air-ground collaborative drone delivery service network in a specific urban area, using the reuse of existing PFWs and new adapted sites as takeoff and landing nodes.We construct an optimization model with three core objectives (minimizing total cost, maximizing order demand coverage, and minimizing service area overlap ratio), and adopt an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II) with stratified initialization, adaptive mutation and constraint penalty mechanism to solve it and generate a Pareto optimal solution set, revealing the quantitative trade-off relationship among the objectives. Taking Futian District of Shenzhen City as the case study. The results show that the Pareto optimal solutions cover a reasonable siting interval; cost, coverage rate and overlap ratio present significant constrained trade-off characteristics; and representative schemes with different orientations can adapt to diverse demands such as pilot deployment, service provision in core areas, and dynamic adjustment of resources. The research findings can provide a practical example and technical methodological support for the construction of human-machine friendliness from the perspective of community delivery services.

ABSTRACT. Urban terminal transportation connects port-based freight flows with urban customers and is critical to city logistics systems. However, most studies on ground-air collaborative delivery focus on simplified last-mile settings and do not fully capture the integrated transshipment and service constraints of urban terminal transportation. This paper studies a truck-van-drone collaborative transportation problem with transshipment (TVDCTP-T), in which shipments move between ports, warehouses, and customers through coordinated operations of tractors, trailers, vans, and drones. The problem represents a two-stage terminal logistics process: containers are first transported to selected transshipment warehouses, and customer requests are then served by vans and drones. Requests include delivery, pickup, and combined pickup-and-delivery tasks. High-floor requests must be served by drones, whereas low-floor requests can be served by vans or drones. The model also allows multi-customer drone sorties and flexible docking, enabling drones to land on vans different from their launch platforms. Time windows, vehicle capacities, drone payload and endurance limits, open routes, warehouse unloading, and inter-stage synchronization are incorporated. A mixed-integer linear programming model is first developed to formulate the problem. Given its computational complexity, this paper proposes a two-stage heuristic framework. The first stage determines container transshipment, van routes, drone sub-routes, and export-request pickup operations, while the second stage optimizes return drayage from warehouses to ports. For larger instances, a reinforcement learning-enhanced cascade-based adaptive large neighbourhood search algorithm (RL-C-ALNS) is developed. The cascade mechanism jointly updates dependent van routes, drone sub-routes, and transshipment decisions when customer assignments change, while a proximal policy optimization module guides destroy-repair operator selection without altering the objective function, neighbourhood design, or simulated annealing acceptance rule. Computational experiments are conducted on randomly generated instances and realistic Shenzhen-based cases. Results show that exact optimization is only effective for very small instances, whereas RL-C-ALNS consistently outperforms greedy construction and standard C-ALNS in solution quality for medium- and large-scale instances and often reduces computational time. Sensitivity analyses reveal that request composition, high-floor demand ratio, drone fleet configuration, drone capacity, and time-window tightness affect system performance. Overall, this study contributes an integrated modelling framework, a cascade-based search mechanism for coupled ground-air decisions, and an RL-guided operator selection strategy, while providing guidance for transshipment coordination, drone resource sizing, high-floor demand management, and flexible service-time design.

ABSTRACT. Road freight decarbonization is increasingly constrained by the lack of suitable energy replenishment infrastructure for electric trucks. This challenge is critical because the efficiency of road freight electrification depends not only on the charging strategy, but also on the spatial context in which electrification is deployed. This study proposes a framework for comparing battery swapping stations and fast charging stations across two distinct settings, intercity corridors and metropolitan areas. The framework recognizes that corridor and urban networks differ in route structure, accessibility, station spacing, and demand distribution, which can lead to different infrastructure outcomes. A two-stage framework is proposed. In the first stage, feasible station locations for each technology are identified through spatial optimization using technology-specific candidate site sets and network-based feasibility constraints. In the second stage, the economic implications of the selected layouts are evaluated by translating spatial outcomes into operational and cost inputs, including recharge distance, utilization, and ton-kilometer cost. The framework enables a consistent comparison between battery swapping and fast charging while accounting for their different infrastructure requirements and deployment logics. Hence, this study supports decision-makers in identifying which charging strategy is better suited to different freight network settings.

ABSTRACT. This paper presents an overview of the 6th Tokyo Metropolitan Freight Survey (TMFS) conducted in 2023–2024 and examines recent trends in urban logistics and their implications for urban and transport policy in the Tokyo Metropolitan Area (TMA). The analysis identifies two key trends. First, logistics facilities are increasingly concentrated in areas with high transport accessibility, such as coastal zones and locations near expressway interchanges. Second, the expansion of e-commerce has significantly increased parcel deliveries in residential areas, intensifying logistics activities and delivery vehicle traffic. Based on the results, it is important to promote measures to improve logistics efficiency in urban planning and community development, while ensuring compatibility with the urban environment.

ABSTRACT. The rapid emergence of low-altitude micromobility systems, such as unmanned aerial vehicles and flying e-scooters, introduces significant safety challenges in complex urban airspace. High trajectory uncertainty, dynamic interactions, and limited spatial separation increase the risk of conflicts and collisions. Existing studies often address trajectory prediction, risk assessment, and control separately, lacking an integrated framework for urban micromobility applications. To address this gap, this study proposes an AI-driven framework that integrates trajectory prediction, probabilistic risk assessment, and adaptive control within a unified closed-loop system. A hybrid Extended Kalman Filter–Long Short-Term Memory (EKF–LSTM) model is used to capture nonlinear motion under noisy conditions. A Trajectory Dispersion Cone (TDC)-based approach quantifies trajectory uncertainty and estimates collision probability using Monte Carlo simulation. In addition, a Velocity Obstacle-based Model Predictive Control (VO-MPC) strategy enables real-time trajectory adjustment and collision avoidance. The proposed framework enhances both safety and adaptability compared to conventional decoupled approaches. Simulation results demonstrate improved prediction accuracy and reduced collision risk under varying conditions. This study extends existing safety models to urban low-altitude micromobility systems and provides a scalable framework for future airspace-integrated mobility networks.

ABSTRACT. The shift to automated transportation, with a combination of human-driven vehicles (HDV) and cautious autonomous vehicle (AV) traffic, presents considerable safety concerns. This research employs VISSIM/SSAM to assess three dedicated lane (DL) scenarios on the Jakarta Harbour Toll Road. Scenario 1 (Cautious AV DL) offers the greatest long-term mobility, but it presents significant lateral safety hazards (TET) when the Market Penetration Rate (MPR) falls below 60%. Scenario 2 (Cautious AV with Bus DL) represents the ideal balanced strategy for early adoption (10-30% MPR). Scenario 3 (Bus DL) is demonstrated to be unfeasible in the initial and intermediate phases, as the integration of cautious CAVs into normal traffic significantly compromises safety. A sequential implementation approach is effective: start with Case 2, then transition to Case 1 at 40% MPR for increased efficiency. At 80% to 90% MPR, Case 3 emerges as the ideal selection for greatest safety, effectively mitigating the significant lateral dangers associated with alternative dedicated lane options.

ABSTRACT. Autonomous vehicles (AVs) promise to significantly outperform human drivers and reshape future urban mobility systems. Their advanced capabilities rely on real-time processing of large volumes of data. However, equipping AV with powerful computing devices could substantially increase vehicle cost and hinder AV adoption. Deploying roadside units (RSUs) to enable vehicle edge computing for task processing has been widely regarded as an effective alternative. Furthermore, integrating remote cloud computing with RSUs can better handle complex traffic conditions and provide more reliable computing services. However, few research has investigated RSU deployment in vehicle–road–cloud (VRC) context. Considering the uncertainty of vehicle demand, this study develops a two-stage stochastic programming model to minimize the sum of RSU investment cost and expectation of task processing latency cost. The first stage determines RSU deployment decisions, while the second stage optimizes task offloading between RSUs and cloud servers. The sample average approximation (SAA) method and piecewise linearization is employed to reformulate the problem as a two-stage stochastic mixed-integer linear programming model (SMILP). Then an L-shaped algorithm with three acceleration techniques is developed to find the global optimal solution efficiently. Numerical experiments demonstrate the effectiveness of the proposed model and solution approach.

ABSTRACT. The decline in mobility among aging populations has become a significant challenge in transportation systems, particularly in regions with limited public transit. While various personal mobility solutions have been proposed, many fail to incorporate users’ lived experiences into engineering design processes. This study proposes a narrative-based design framework that integrates qualitative user narratives with systematic engineering methods for developing personal mobility systems tailored to transportation disadvantaged individuals. The proposed framework consists of three key components: narrative analysis, Needs–User–Problem–Specs (NUPS) modeling, and Quality Function Deployment (QFD). Narrative data collected from user interviews and open-ended questionnaires are decomposed into meaning units, coded, and categorized to identify latent needs embedded in daily mobility experiences. These needs are then structured into NUPS models and translated into engineering specifications through QFD, enabling a traceable linkage from qualitative insights to design parameters. To validate the framework, a license-free personal mobility vehicle, “TinyCabin,” was designed and developed as a case study. A pilot user evaluation was conducted involving individuals with mobility constraints. The results indicate improvements in perceived safety, ease of operation, and willingness to travel, demonstrating the effectiveness of the proposed approach in capturing user-centered requirements and supporting practical mobility design. The significance of this research lies in enabling independent mobility for elderly and physically constrained individuals through a systematically designed personal mobility system.

ABSTRACT. With the growth of global air transport demand, accurate four-dimensional trajectory prediction is crucial for modern air traffic management. Existing methods mainly predict non-linear flight trajectory data through monolithic physical modelling methods or deep learning algorithms, which makes it difficult to achieve accurate prediction of long time-series data. To address this, we propose a hybrid prediction framework KF-RWKV that fuses Kalman Filtering (KF) and Receptance Weighted Key Value (RWKV), which combines physical interpretability with time-series inference capability of deep learning through a four-layer system. First, the KF algorithm is used to generate an initial baseline trajectory. Second, a residual mapping sequence of real trajectory and KF baseline trajectory is constructed. Third, the residual sequence is processed using the Time-Mixing Layer and Channel-Mixing Layer of RWKV. Last, the neural gating dynamically adjusts the weights between KF and RWKV, and solves the probability interval of the predicted trajectory. The experimental validation results based on the trajectory data of Nanjing Lukou International Airport (NKG) show that the KF-RWKV algorithm has excellent prediction performance. Taking altitude as an example, compared with other algorithms, KF-RWKV reduces Mean Absolute Error (MAE) by 9.57%-48.26%. Additionally, KF-RWKV algorithm shows the best prediction results in all three descent phases.

ABSTRACT. This study proposes a Remote Operation Centre (ROC) role framework and a hierarchical alert escalation framework for the remote supervision of Maritime Autonomous Surface Ships (MASS). The study addresses the question of how an internal decision-making process should evaluate operational inputs to determine whether, when, and how to alert the ROC operator. To support this, a structured five-step scenario development framework is combined with state-chart modelling to represent alert behaviour under different operational conditions. The proposed framework integrates stakeholder requirements, alert related functionality, Operational Design Domain (ODD) constraints, system level goals, and literature informed alert design patterns. The resulting state-chart models six alert states, ranging from normal autonomous operation to fallback, and captures both gradual and abrupt escalation paths. Scenario-based interpretation suggests that the framework can support navigational safety, structured operator involvement, transparency, operational efficiency, and regulatory compliance, while also enabling deescalation when conditions improve. At the same time, the analysis highlights that ROC workload, realistic ODD specification, and effective operator training remain important conditions for the practical use of the framework.

ABSTRACT. Unmanned Aerial Vehicle (UAV) shore-to-ship delivery faces challenges including limited flight ranges, dynamic interception complexities, and multi-task queuing congestion. To quantify the benefits of offshore relay hubs, we propose a cooperative delivery model integrating dynamic interception and a multi-level queuing state machine, explicitly considering node capacity constraints. An adaptive heuristic algorithm is employed for optimization. Simulations reveal two core mechanisms of maritime relay networks: (1) A "Spatial Springboard" effect. The architecture overcomes range limitations, boosting order fulfilment from 63.3% (direct-flight) to 93.3%. Demonstrating diminishing marginal returns, deploying 75% of core hubs captures 90.0% of the fulfilment benefits. (2) A "spatiotemporal buffer" effect. Despite a 47.4% surge in effectively handled orders, the network's traffic diversion capacity reduces the global makespan (from 168.1s to 160.8s). It restricts average queuing delays to a sub-linear increase (from 47.2s to 57.6s), successfully preventing exponential congestion deadlocks triggered by throughput spikes. This study bridges the theoretical gap in highly dynamic, resource-constrained logistics modelling, providing a rigorous quantitative basis for cost-effective maritime infrastructure planning.

ABSTRACT. The rapid growth of the electric vehicle (EV) industry has precipitated a critical challenge regarding the sustainable management of end-of-life (EOL) power batteries. Existing reverse supply chain (RSC) networks often struggle with multifaceted uncertainties in supply, demand, and material recovery processes, leading to inefficient operations and suboptimal economic and environmental outcomes. This study addresses this gap by proposing a strategic and operational planning model for a multi-period, multi-echelon RSC network dedicated to EV battery recycling. The network encompasses collection centers, testing and disassembly centers, utilization centers, and secondary markets. To effectively capture and mitigate the risks posed by uncertainty, the problem is formulated as a Hybrid Distributionally Robust Stochastic Optimization (HDRSO) model. This approach combines scenario-based stochastic programming for uncertainties in battery state classification and market demand with a distributionally robust optimization (DRO) framework, based on the Wasserstein metric, to handle the ambiguity in the probability distribution of used battery supply. The primary objective is to minimize the total expected system cost, which includes fixed facility activation costs, variable collection/processing/holding costs, transportation costs, environmental taxes for waste disposal, and benefits from government subsidies. The model is also constrained by a carbon cap-and-trade policy, ensuring that the total carbon emissions from transportation and processing activities do not exceed the predefined caps. The solution determines the optimal operational decisions, including material flows and inventory levels across periods, under the worst-case distribution within the Wasserstein ambiguity set. Our findings demonstrate that the proposed HDRSO-based RSC network design provides a robust and resilient strategy against supply volatility and other uncertainties, significantly improving cost-effectiveness and environmental compliance compared to deterministic or traditional stochastic models. This research offers a comprehensive decision-support tool for policymakers and industry stakeholders aiming to establish a sustainable and circular economy for EV batteries.

ABSTRACT. Maritime fuel distribution in archipelagic regions presents a complex routing challenge that classical vehicle routing models fail to address adequately, as they assume constant vessel speed regardless of environmental conditions. This study formulates the problem as a Weather-Aware Multi-Depot Split-Delivery Capacitated Vehicle Routing Problem (WA-MDCVRP-W), incorporating real-time weather data into an ETA-minimization framework. Vessel effective speed is modelled as a function of significant wave height, wind speed, and ocean current projection, retrieved via the Open-Meteo Marine API and Forecast API. Routing is optimized using three independent metaheuristic algorithms Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Tabu Search (TS) applied to 670 delivery cases across 20 depots, 51 destination ports, and 7 vessel types over a 10-month period. Results demonstrate that all three algorithms converge to identical ETA solutions in 99.9% of cases (mean ETA: 10.71 days), confirming solution robustness at the tested problem scale. The primary differentiator is computational efficiency: GA requires an average of 3.7 seconds per case (7–14× slower than PSO and TS), making TS the most recommended algorithm for real-time operational deployment. Weather analysis shows that 98.8% of observations fall under safe conditions, with peak ETA occurring in July (9.09 days) coinciding with the eastern monsoon season in Indonesian waters. The proposed framework demonstrates that integrating dynamic weather data into maritime routing produces more realistic and operationally reliable distribution plans compared to static models.

ABSTRACT. Large concerts, sporting events, and other planned special events generate intense short-term outbound demand immediately after the event ends, creating a need for co-ordinated post-event transport operations. This paper studies a post-event multimodal shuttle dispatch problem in which spectators leave a venue through multiple exits and are transferred to multiple external transport hubs by a heterogeneous shuttle fleet. A discrete-time optimisation model is developed to capture time-varying exit demand, passenger loading, route accessibility by vehicle type, hub absorption limits, segment-level congestion approximation, and reusable vehicles over multiple trips. To balance validation and scalability, the study adopts a two-layer solution framework consisting of an exact baseline for small instances and an adaptive large neighbourhood search heuristic for larger cases. Computational experiments on synthetic instances show that the exact baseline provides reliable benchmark solutions on small cases and most medium-scale cases, whereas the heuristic continues to generate feasible solutions when exact optimisation reaches its practical tractability boundary. Sensitivity analysis reveals an asymmetric pattern: demand growth and constrained hub capacity are the primary drivers of system deterioration, whereas fleet augmentation yields diminishing returns once downstream receiving throughput becomes the binding constraint.

ABSTRACT. Freight transport in Australia is highly dependent on roads, largely driven by the adoption of Performance-Based Standards (PBS) vehicles (high-productivity vehicles), while railway infrastructure faces significant constraints. This study constructs a global logistics intermodal network simulation model that endogenously integrates international and interstate cargo to quantitatively analyze time-series changes from 2013 to 2023 and future scenarios across Australia. The model features detailed origin-destination matrices by transport mode and commodity, estimated using various statistics and night-time light data. Additionally, distance-proportional freight rates were redesigned based on actual cost structures, and a path-search algorithm was introduced to prevent interstate freight from passing through international maritime transport links. The analysis reveals that the popularization of PBS vehicles over the past decade has drastically reduced unit transport costs, establishing a highly road-dominant structure. Regulating PBS vehicles would result in an estimated social loss of 1.85 billion USD. Meanwhile, over 99% of the utilization of the Inland Rail, a dedicated freight railway currently under construction, is attributed to domestic freight, indicating limited contributions to international logistics. This study quantitatively demonstrates the impacts of infrastructure investments and vehicle regulations on the entire logistics network, providing valuable insights for policymaking.

ABSTRACT. Most parcel-locker research has focused on location optimisation, routing efficiency, user acceptance, and environmental performance, with comparatively limited attention to the policy and governance conditions under which locker systems are deployed and scaled. This study addresses that gap by examining which policy initiatives should be prioritised to expand parcel locker utilisation in Korea and how those initiatives interact as an interdependent portfolio. Nine initiatives spanning infrastructure, incentives, and sustainability/governance were identified through a structured literature review, cross-national comparison, and a two-round Delphi survey with domain experts, then evaluated using the Best–Worst Method (BWM) and fuzzy DEMATEL. The results suggest that infrastructure-oriented interventions receive the highest priority from the expert panel. Mandatory installation in residential and high-density areas, accessibility- and equity-oriented location optimisation, and integration with public transport stations consistently form the top tier across both additive and multiplicative integration schemes. Subsidies and tax incentives, safety and operational standards, and public–private partnerships function as upstream enabling drivers within the policy system. The findings suggest that parcel lockers may be better understood as a supply-side infrastructure provisioning challenge rather than merely a demand-side convenience option, and that wider utilisation may be best advanced through a coordinated policy package rather than through isolated demand-side measures.

ABSTRACT. Urban driving environments are frequently affected by varying illumination conditions, including solar glare, shadows, and rapid light transitions, which can significantly degrade drivers’ visual perception and increase safety risks. To address this challenge, this paper presents an integrated framework for vehicle route navigation under dynamic illumination conditions by combining environmental perception, illumination risk assessment, and path planning. The proposed framework, termed Glare Defender (GD), incorporates both static and dynamic illumination information to identify high-risk road segments and generate safer navigation routes. A static illumination risk map is constructed using building geometry, terrain elevation, and solar position information to estimate long-term illumination patterns in urban road networks. Meanwhile, a dynamic illumination perception module utilizes onboard sensing data and lightweight machine learning models to detect real-time illumination hazards. To improve detection robustness under complex urban scenarios, a shadow-aware enhancement strategy is introduced to distinguish hazardous glare regions from non-critical lighting variations.Experimental results demonstrate that the proposed illumination perception approach achieves an accuracy of 93.2% while maintaining a response latency below 100 ms, enabling real-time deployment. The navigation module effectively reduces exposure to illumination-risk zones while preserving acceptable travel efficiency.

ABSTRACT. Electrification and truck platooning can decarbonize long-haul freight, yet their operational and economic interactions in multi carrier settings remain underexplored. We examine electrified cross-carrier platooning on long-haul corridors where electric trucks coordinate departures, charging, and leader and follower roles under varying electricity prices and battery constraints. We propose an integrated optimization framework that jointly decides charging quantities, departure synchronization, and segment level platoon composition, and couples these decisions with a cooperative cost allocation scheme to ensure coalition stability. The model captures state of charge dynamics, station specific charging power, leader energy asymmetry, and delivery deadlines. We quantify the efficiency stability trade-off and compute the minimum transfers or subsidy required when the grand coalition is unstable. Experiments show savings of up to 22.00% versus decentralized charging and about 19.77% lower charging demand at high price stations via temporal load shifting, supporting grid related emission benefits when incentives align.

ABSTRACT. This study takes freight forwarding enterprises as the core decision-makers, focusing on capacity fluctuations in the multimodal transport of Chinese new energy vehicle exports and carbon emission cost accounting under low-carbon policies. An optimization model is constructed to minimize customers' total logistics costs. First, a container multimodal transport route optimization model under deterministic capacity is established as a benchmark. Second, capacity uncertainty is introduced to develop a two-stage robust optimization model. Through an export transportation case study from the southwestern region of China to Qingdao Port, the column-and-constraint generation algorithm is employed to solve the model. The impacts of the uncertainty budget level Γ and shipment volume on total cost, carbon emissions, and transit time are analyzed. The results indicate that by systematically mapping key supply chain links centered on freight forwarders and integrating the three-dimensional metrics of "cost–time–carbon emissions," logistics processes can be effectively optimized. The use of polyhedral uncertainty sets to characterize capacity fluctuations and the application of a two-stage robust optimization model enhance the scientific rigor of decision-making and improve the robustness of solutions.

ABSTRACT. This paper studies the coordinated optimisation of vessel scheduling and speed decisions for multi-type coastal liquid cargo transportation under dynamic vessel states, time-window constraints, port-compatibility requirements, and multiple operating-cost components. A unified framework is developed in which a constructive heuristic first generates an initial feasible schedule, adaptive large neighbourhood search then improves vessel–task assignments and service sequences, and two speed-decision mechanisms, namely post-scheduling optimisation and embedded joint optimisation, are introduced to capture different levels of schedule–speed interaction. Computational experiments on six real-world industrial instances show that the proposed ALNS procedure reduces total cost by 3.11% on average relative to the constructive solution, while maintaining high cross-seed stability. Post-scheduling speed optimisation provides a further average reduction of 0.29%, mainly through ballast-leg slow steaming, whereas joint optimisation achieves the lowest or tied-lowest final cost on most instances by feeding true post-adjustment costs back into the search. Additional experiments show that alternative search objectives can occasionally outperform direct cost minimisation, highlighting both the path dependence of heuristic search and the practical value of business-informed surrogate objectives.

ABSTRACT. The decarbonization of freight transport has accelerated the adoption of electric heavy-duty trucks (e-trucks), which offer significant potential for reducing greenhouse gas emissions and advancing sustainable logistics systems. However, their deployment remains constrained by limited driving ranges, high battery costs, and operational inefficiencies associated with prolonged charging times. In addition, carriers face considerable uncertainty regarding battery depreciation and degradation, which affects asset valuation and lifecycle costs. Battery swapping technology has emerged as a promising alternative, enabling rapid energy replenishment, decoupling battery ownership from vehicle ownership, and thereby alleviating carriers’ concerns over battery performance and residual value. Despite its potential, the large-scale implementation of battery swapping infrastructure raises complex planning challenges, particularly in coordinating facility location decisions with vehicle routing under uncertain market adoption. These challenges are especially pronounced in heavy-duty intercity line-haul transport, which is characterized by relatively fixed routes and schedules, higher reliance on battery swapping, and limited availability of swapping options.

This paper investigates an integrated facility location and energy-constrained shortest path problem for a freight system composed of e-trucks equipped with swappable batteries, operated by multiple carriers. The study briefly discusses the roles and interactions of key stakeholders, including battery swapping station managers, freight carriers, energy providers, and government agencies, whose decisions jointly shape infrastructure deployment and system performance. In particular, it focuses on the strategic and operational decisions of swapping station managers.

We formulate the problem as a two-stage stochastic mixed-integer linear program (MILP), where the first stage determines the optimal placement of swapping stations and battery investments, and the second stage captures operational energy-constrained path planning decisions under stochastic scenarios of e-truck adoption rates. The operational planning involves scheduling the battery swapping activities of e-trucks – their visiting sequences of swapping stations, which necessitates a degree of coordination and resource sharing across carriers, reflecting an open and interconnected freight ecosystem. The two-stage structure enables the application of decomposition-based solution approaches, enhancing computational tractability for large-scale networks. Moreover, the proposed model can be naturally extended to incorporate alternative energy replenishment systems, such as fast charging stations integrated with battery storage.

Numerical experiments are conducted on two representative networks: an inter-city corridor between Melbourne and Sydney in Australia, and a regional network in Hebei, China. Preliminary results highlight the sensitivity of infrastructure deployment to market penetration levels and demonstrate the potential of battery swapping systems to enhance the efficiency of electric freight operations. The findings provide managerial insights for infrastructure providers, freight operators, and policymakers in supporting the transition toward sustainable freight systems.

ABSTRACT. The growth of e-commerce and the shortage of freight drivers have intensified pressure on urban last-mile logistics. This study examines a fixed-station two-echelon delivery system in which trucks transport parcels to intermediate stations and delivery robots complete the final delivery stage. The aim is to clarify how station number and location affect delivery efficiency and the required fleet size of delivery robots. An integrated framework is developed by combining a two-echelon vehicle routing problem model, a discrete-event simulation, and a station deployment optimisation procedure based on variable neighbourhood search. The case study focuses on the east exit district of Omiya railway Station in Saitama, Japan, using district freight survey data and a real urban road network. The results show a clear trade-off. Increasing the number of stations reduces robot delivery time, especially when moving from one station to multiple stations, but it also increases the total number of robots that must be deployed. In the study area, the marginal time-saving benefit becomes limited beyond two stations, whereas robot fleet requirements continue to grow. The findings indicate that excessive station deployment is not necessarily efficient and that station planning should balance time savings against fleet requirements.

ABSTRACT. Japan’s logistics distributors are facing a severe shortage of drivers due to an ageing population, declining birth rates and recent labour reform policies. While some distributors have introduced routing software, many still rely on fixed service territories. In Japan, parcel carriers and fuel distributors often allocate fixed territories to drivers to improve customer satisfaction and driver productivity, and these territories are periodically reviewed by managers. However, territory design is still largely based on managerial experience. Many previous studies have relied on continuous approximation models or have assumed that multi-period customer demand is known in advance. In practice, while distributors know the location of each customer, their delivery volume and average delivery frequency, they cannot accurately predict which customers will require service several days ahead. To address this, we have developed a territory design algorithm inspired by the location-based heuristic for the capacitated vehicle routing problem. We reformulated the problem as a facility location model by aggregating customer address blocks and minimising the expected total travel distance. Using data from a real service area with over 3,000 customers served by six drivers, we demonstrate that the proposed territories significantly reduce the total distance travelled for deliveries compared with existing territories.

ABSTRACT. Artificial intelligence (AI) is increasingly embedded in last-mile online food delivery (OFD) platforms, yet how AI acceptance translates into customer loyalty, particularly through trust remains underexplored in emerging markets such as Thailand. Drawing on the Technology Acceptance Model and Trust Transfer Theory, this study tests a structural model linking AI acceptance, multi-dimensional trust and customer loyalty using survey data from 300 OFD consumers, analysed via PLS-SEM with bootstrapping. AI acceptance was captured across five service-interface dimensions: real-time tracking, contactless delivery, last-mile efficiency, sustainable delivery practices and perceived transparency. While trust was captured across five dimensions: platform reputation, security and privacy assurance, trust transfer, crowd logistics experience and trust in AI systems. Loyalty was measured via continued use intention and advocacy. Three of four hypothesised paths were supported: AI acceptance strongly predicted trust (β = 0.861, t = 23.735, p < .001) and trust strongly predicted loyalty (β = 0.783, t = 11.392, p < .001), whereas the direct AI acceptance, loyalty path was non-significant (β = 0.117, t = 1.629, p = .103). The indirect effect through trust was significant and substantial (β = 0.675, t = 12.572, p < .001), confirming trust as the dominant mechanism linking acceptance to loyalty (Trust R² = 0.742; Loyalty R² = 0.785; SRMR = 0.031). Customer loyalty toward AI-enabled OFD platforms thus appears to be earned through perceived trustworthiness rather than technological sophistication itself, repositioning platform competitiveness around transparency and reliability rather than AI complexity. The study extends technology acceptance theory by positioning multi-dimensional trust as the boundary mechanism for AI-enabled logistics services in an emerging economy.

ABSTRACT. The world of warehousing is rapidly evolving towards increasingly automated and efficient solutions capable of storing a huge number of items in small quantities. The parts-to-picker system discussed in this article is an innovative solution that combines AS/RS (miniloads) for retrieving/storing totes, and autonomous mobile robots for handling them. We studied the performance of such a system through a discrete event simulation, varying the layout of the warehouse and vehicle fleet and analysing the relationships between them. The results show that productivity is mainly determined by the number of aisles, equivalent to the number of miniloads, while autonomous mobile robots show decreasing returns beyond a saturation point. A regression model is proposed to approximate the performance of the system. The study also qualitatively compares the hybrid architecture with traditional miniload systems.

ABSTRACT. The growing demand for efficient parking management in urban areas has driven interest in smart parking reservation systems. While such systems can reduce congestion and searching time, their success largely depends on user acceptance. This study applies ensemble learning approaches—majority voting, bagging, boosting, and stacking—to predict acceptance of parking reservation systems and identify influencing factors. Data were collected from a mixed preference survey of 750 respondents in Klang Valley, Malaysia. Model performance was assessed using accuracy, precision, recall, F1 score, AUC, log-loss, and Matthews Correlation Coefficient (MCC). Results show that stacking with Elastic Net as the meta-learner achieved the strongest overall performance, combining high accuracy, recall, and F1 score. Majority voting and bagging offered balanced performance, while boosting (LightGBM) provided the highest precision. The findings highlight stacking as the most reliable ensemble framework, offering valuable insights for the design and implementation of future smart parking solutions. Besides, it is found that the key influencing factors that contribute to users’ acceptance of parking reservation applications are reservation charges, highest education, satisfactory level of parking availability, travel distance, age, frequency of parking challenges, number of cars, success rate, parking needs, and parking searching time.

ABSTRACT. Estimating traffic flow on unobserved roads is a critical yet challenging task for achieving comprehensive network-wide traffic state awareness, which serves as the fundamental backbone for Intelligent Transportation Systems (ITS). In urban settings, however, prevailing methods are often impractical due to their reliance on costly auxiliary data, or fundamentally inadequate because they model roads merely as spatial nodes—a practice that ignores the heterogeneous context shaping urban traffic. To bridge the gap, we propose a Knowledge Graph-based Pairwise Difference Learning Framework (KG-PDLF). First, we recast roads as semantic entities within a custom-designed Urban Traffic Context Knowledge Graph (UTCKG). It explicitly models the diverse factors and distinct relationships of the urban context that shape traffic dynamics, enabling the generation of context-aware road representations via knowledge graph embedding. Second, we develop a novel estimation model integrating Pairwise Difference Learning (PDL) with a network-state-aware architecture. This model (1) exploits the inherent benefits of PDL to improve generalization by learning from relatively stable pairwise flow differences, and (2) extends conventional PDL to enable dynamic estimation by conditioning on the observable network state. Extensive experiments on two real-world urban datasets (Paris and Torino) comprehensively validate the effectiveness and practicality of the proposed framework. The results explicitly show that KG-PDLF consistently outperforms state-of-the-art baseline methods, including current Large Language Model (LLM)-based approaches, across diverse data scarcity scenarios, demonstrating both superior accuracy and stability. Furthermore, in-depth error analyses and ablation studies confirm its strong adaptability to varied roads and time slots, as well as the critical contribution of each core component. Crucially, KG-PDLF also exhibits excellent computational efficiency and scalability, offering a highly practical solution for real-world, large-scale urban traffic deployments where detector coverage is strictly limited. Ultimately, this study provides a cost-effective framework for achieving network-wide traffic state awareness, paving the way for future integrations of LLMs and multi-source data to further enhance semantic understanding and overall performance.

ABSTRACT. Safety on the roads remains a very significant issue, particularly when consider-ing the increased number of accidents that occur due to distractions on both a physical and psychological level. In this study, a framework for a system to mon-itor the drivers' condition based on emotion detection and drowsiness detection is proposed. Two benchmarks, FL3D, and AffectNet, are used for training purpos-es. Although the drowsiness detection model learns to recognize whether the eyes are open, closed, and yawning, emotion detection learns four different emotions: happiness, neutral, anger, and sadness. Preprocessing techniques like scaling, greyscale conversion, and data augmentation are used to increase model accuracy. The system recognizes emotions with a moderate accuracy of 71.4% and detects tiredness with a high accuracy of 97.6%. A new safety score matrix that deter-mines an alert level and initiates the appropriate warnings is created by combining the two detection results. The integrated approach aims to improve driver safety by detecting subtle cognitive and physical signs of impairment in real time. Future developments include implementation on edge devices for greater scalability, bet-ter emotion classification, and personalization.

ABSTRACT. The subject research proposed the risk control concept of ventilation system for ship engine control room when adopting ammonia as the new future fuel. The research considered the assumption scenario of ammonia leakage in engine room and built the CFD simulation model and numerical analysis method for the prevention of ammonia gas entering engine control room based on analysis the flow distribution and differential pressure. The efficiency of setting chamber room to improve the gas safety is shown. The operation scenario for chamber room is proposed based on the CFD analysis result of the differential pressure and the ammonia gas concentration during actual ventilation operation of chamber room.