Purpose of this paper:

The transition toward Industry 5.0 is redefining the logistics sector by shifting focus from automation toward human-centric collaboration between operators and intelligent systems. The widespread use of Autonomous Mobile Robots (AMRs) in intralogistics highlights the need for engineers to understand autonomous behaviour and the underlying perceptual processes that enable it. While Simultaneous Localization and Mapping (SLAM)-based perception systems are well established in robotics research, their internal functioning often remains opaque in logistics educational contexts. This paper aims to address this gap by developing a transparent-box educational demonstrator that makes SLAM-based perception in AMRs observable and intuitively understandable for learners without prior expertise in robotics or artificial intelligence.

Design/methodology/approach:

The research follows a three-phase methodology: (1) demonstrator development based on user stories and a transparent-box principle, (2) a pilot test to assess usability and visual clarity, and (3) iterative refinement for educational deployment. The system utilises a layered architecture in which an OAK-D-S2 camera provides visual input for vSLAM-based perception, the Robot Operating System (ROS) handles data integration and processing, and Unity is employed for real-time visualisation of perceptual outputs. Central to this approach is a two-stage perceptual architecture designed to incrementally expose the principles of robotic perception: the first stage generates a real-time geometric SLAM map to visualise localization and structural mapping, while the second stage utilizes a YOLOv8 neural network to augment this map with object-level semantic information, illustrating context-aware perception.

Findings:

The demonstrator enables learners to directly observe how geometric mapping and semantic enhancement are constructed and fused during operation. By visualising these outputs in real time and embedding them in a logistics task, abstract algorithmic concepts become accessible and relatable to operational contexts. The transparent-box approach supports intuitive understanding of robot perception without requiring system construction or advanced programming skills.

Originality/value:

This study shifts the educational focus from robotic system construction to perceptual transparency, providing a scalable, modular platform for transdisciplinary science, technology, engineering and mathematics (STEM) education. It addresses the specific call for innovative tools that equip future engineers to manage the complex, interconnected systems of the Industry 5.0 era. The modular design further allows other institutions to replicate the setup or extend it to include additional autonomy layers such as path planning and decision-making.

Purpose of the paper

Maritime logistics is the backbone of global trade, but it is continuously exposed to disruptions that propagate across ports, shipping networks, and downstream supply chains. These disruptions include congestion, schedule unreliability, adverse weather-related operational effects, and capacity shocks, which often materialize as measurable outcomes such as arrival delay beyond a service threshold, excessive anchorage time, missed berthing windows, or elevated queue risk. From a decision-support perspective, the most relevant signal is not a retrospective explanation but a prognostic forecast of disruption risk with actionable lead time. This paper, therefore, aims to develop a unified forecasting framework for sea freight risk prediction that estimates calibrated probabilities of future disruption outcomes by integrating heterogeneous quantitative data sources, namely AIS trajectories and kinematics, port and PMIS operational signals, and macro and trade indicators.

Design/methodology/approach

The paper proposes SeaRisk Flash, a unified, streaming-capable, early-fusion forecasting framework for multi-task, multi-horizon prognostic risk prediction in sea freight. The design begins with a canonical time grid to align heterogeneous inputs with different update regimes. AIS data are treated as irregular observations, PMIS and port signals as event-driven operational streams, and macro and trade indicators as mixed frequency data with publication lags. To prevent leakage, the framework applies strict as-of semantics and vintage-aware handling of macro and trade variables, including release latency and age features. Missingness masks and quality flags are explicitly retained to enable the model to learn under partial observability and noisy inputs.

After temporal alignment, the framework uses a hybrid quantitative tokenization strategy to build unified token packets that combine encoded AIS, port, and PMIS, and macro trade context at each decision time. Modality-specific encoders transform each stream into a shared latent token space using causal and lightweight components tailored to the source characteristics. These fused representations are processed by a transformer-based Mixture of Experts backbone with router gating and sparse expert activation, enabling conditional computation and specialization across high-frequency vessel kinematics, port state changes, and slower macroeconomic context. The model outputs calibrated probabilities for multiple endpoints, including delay, congestion, and queue risk, across multiple future horizons. The framework also includes chunk-wise synchronized interleaving and sparse-dense sampling to support efficient streaming updates and computational scalability. In addition, it introduces a curriculum-inspired, progressive training schedule and a modality-decoupled training strategy to manage heterogeneity and stabilize training throughput.

Findings

This paper is a conceptual and methodological contribution and does not yet report a full empirical benchmark on a production deployment. Its main finding is the specification of a coherent and operationally plausible design for unified prognostic risk forecasting in sea freight under realistic data constraints. The framework shows how heterogeneous maritime inputs can be integrated into a single forecasting pipeline while preserving temporal validity, handling missingness and noise, and remaining compatible with streaming inference across many vessels and ports. It also clarifies why early fusion can be beneficial in maritime risk forecasting, namely because disruption outcomes often arise from interactions across sources, such as vessel deceleration becoming meaningful only in combination with high berth utilization and elevated demand regimes. The proposed architecture is therefore positioned to capture cross-source interactions directly, rather than combining independently modeled signals only at the final decision layer.

Value/Originality

The originality of the paper lies in transferring and adapting design principles from LongCat Flash Omni to a maritime quantitative forecasting setting, where the modalities are heterogeneous numerical streams rather than audiovisual inputs. The contribution is not a simple architectural reuse, but a domain-specific redesign for sea freight risk forecasting under as-of constraints, mixed update frequencies, and operational deployment requirements. The paper combines early fusion, Mixture of Experts, curriculum-inspired progressive training, chunk-wise interleaving, sparse-dense sampling, and modality decoupling in a single framework for sea freight. This addresses an important methodological gap in the current literature, where AIS, port operations, and macro trade context are often modeled separately or only weakly integrated, despite their joint relevance for disruption risk.

Research limitations/implications

Several limitations are explicitly acknowledged. First, the framework assumes access to proprietary AIS and PMIS data, which may restrict immediate reproducibility or deployment in some settings. Second, real-world risk endpoints require domain-specific labeling choices, for example, when differentiating weather-driven delays from labor or infrastructure-related disruptions. Third, the framework is predictive rather than causal, meaning it forecasts risk probabilities but does not by itself provide counterfactual recommendations such as reroute versus wait. These limitations define a clear research agenda for future work, including real multi-port evaluation, richer external covariates such as weather and ocean exposure, and decision-aware learning objectives that better reflect operational utility.

Practical implications

If implemented and empirically validated, SeaRisk Flash could provide earlier, more reliable risk signals for voyage delays, port congestion, and queue conditions, enabling more proactive planning in maritime and supply chain operations. Potential application areas include voyage monitoring, port operations planning, congestion surveillance, scheduling support, and resilience-oriented decision support across fleets and terminals. The streaming design, sparse computation logic, and decoupled encoder backbone workflow are particularly relevant for operational environments that require frequent updates across many entities while controlling latency and compute cost.

Purpose of this paper: To ensure the smooth trading and regional integration, the logistics are important. However, the logistics development levels of different ASEAN countries are diverse, leading to difference in logistics effectiveness, costs, and reliability. It can therefore possibly result in logistics inequality and hinder the successful regional integration. To support the identification and measurement of potential regional logistics inequality, the first and foremost step is to accurately evaluate and rank the logistics performance among ASEAN countries. To achieve this goal, this paper attempts to develop a method to comprehensively evaluate the ASEAN countries’ logistics performance and support decision-making for relevant stakeholders, including governments, policy makers, and companies in ASEAN area.

Design/methodology/approach: To achieve the research goal, a comprehensive evaluation framework combining hybrid weighting methods and multi-criteria decision-making (MCDM) models is proposed and applied to the logistics performance index (LPI) dataset released by the World Bank. Specifically, equal weighting, entropy weighting, and MEREC (i.e., method based on the removal effect of criteria) weighting methods are simultaneously adopted to derive criteria weights of six dimensions of LPI, and a novel dynamic CoCoSo with non-consecutive series is developed to evaluate the logistics performance of ASEAN countries. The contribution of this method to the existing literature is to effectively handle the issue of the non-consecutive feature in LPI series which are not released in a regular basis. By applying this method, a fair logistics performance comparison among countries and among different periods can be expectedly achieved.

Findings: The hybrid weighting CoCoSo results suggest that Singapore is the country having the highest logistics performance, followed by Malaysia and Thailand. The results also indicate that Lao PDR seems to have the lowest logistics performance by holistically considering its LPI across multiple years based on the CoCoSo results.

Value: To the best of our knowledge, this is the first paper applying a hybrid weighting non-consecutive series dynamic CoCoSo approach to evaluate the LPI of ASEAN countries. The newly proposed method can expectedly capture more information in the LPI dataset to derive a more appropriate performance evaluation. More importantly, it is promising to address the issue of systematic data-missing problem and contribute to the time series analysis literature. Meanwhile, it can bring practical values, and the value of the results can potentially support policy makers to design and implement proper logistics-related import/export policies, as well as inform multi-national companies to design their international trade strategies.

Research limitations/implications: The research implications of this study are twofold. First, this study adopts an ASEAN perspective to analyse the LPI, enriching the theoretic basis of international logistics and advancing the academic literature of international trade. Also, this study provided a framework to evaluate LPI, offering new methodological tool for prospective studies for logistics performance evaluation. However, the current research is limited by fixed indicators used in LPI static dataset, which might not be able to capture more information from other perspectives.

Practical implications: This study can inform policy makers and practitioners for their decision-making in logistics and international transport. Based on our results, the policy makers can design proper policies to enhance multi-lateral logistics cooperations among ASEAN countries, while practitioners in international companies can use our results and methods to evaluate their business partners in different countries from logistics perspective.

References:

Jiang Y, Zhang J, Asante D, et al. Dynamic evaluation of low-carbon competitiveness (LCC) based on improved Technique for Order Preference by similarity to an Ideal Solution (TOPSIS) method: A case study of Chinese steelworks. Journal of cleaner production, 2019, 217: 484-492.

Lin Q, Zhang K, Huang D, et al. Evaluating the impact of Trans-Asian railway on logistics mode selection between Thailand and China: An AHP-TOPSIS approach. Alexandria Engineering Journal, 2024, 98: 147-158.

Yazdani M, Zarate P, Kazimieras Zavadskas E, et al. A combined compromise solution (CoCoSo) method for multi-criteria decision-making problems. Management decision, 2019, 57(9): 2501-2519.