Transportation Research Part C-Emerging Technologies最新文献

{"title":"From voice to safety: Language AI powered pilot-ATC communication understanding for airport surface movement collision risk assessment","authors":"Yutian Pang, Andrew Paul Kendall, Alex Porcayo, Mariah Barsotti, Anahita Jain, John-Paul Clarke","doi":"10.1016/j.trc.2026.105540","DOIUrl":"10.1016/j.trc.2026.105540","url":null,"abstract":"<div><div>Surface movement collision risk is critical for airport safety. These models play a vital role in identifying and mitigating potential hazards during airport ground operations by providing warnings of near-miss incidents, thereby reducing the risk of accidents that could jeopardize human lives and financial assets. However, existing models, developed decades ago, have not fully integrated recent advancements in machine intelligence, where incorporating additional functionalities presents promising opportunities for improved risk assessment. This work provides a feasible solution to the existing airport surface safety monitoring capabilities (i.e., Airport Surface Surveillance Capability (ASSC)), namely language AI-based voice communication understanding for collision risk assessment. The proposed framework consists of two major parts, (a) rule-enhanced Named Entity Recognition (NER); (b) surface collision risk modeling. NER module generates information tables by processing voice communication transcripts, which serve as references for producing potential taxi plans and calculating the surface movement collision risk. We first collect and annotate our dataset based on open-sourced video recordings and safety investigation reports. Additionally, we refer to FAA Order JO 7110.65W and FAA Order JO 7340.2N to get the list of heuristic rules and phase contractions of communication between the pilot and the Air Traffic Controller (ATCo). Then, we propose the novel ATC Rule-Enhanced NER method, which integrates the heuristic rules into the model training and inference stages, resulting in a hybrid rule-based NER model. We show the effectiveness of this hybrid approach by comparing different setups with different token-level embedding models. For the risk modeling, we adopt the node-link airport layout graph from NASA FACET and model the aircraft taxi speed at each link as a log-normal distribution and derive the total taxi time distribution. Then, we propose a spatiotemporal formulation of the risk probability of two aircraft moving across potential collision nodes during ground movement. Furthermore, we propose the real-time implementation of such a method to obtain the lead time, with a comparison with a Petri-Net based method. We show the effectiveness of our approach through case studies, (a) the Haneda airport runway collision accident happened in January 2024; (b) the KATL taxiway collision happened in September 2024; (c) the Tenerife airport disaster in March 1977. We show that, by understanding the pilot-ATC communication transcripts and analyzing surface movement patterns, the proposed model estimates the surface movement collision probability within machine processing time, thus enabling proactive measures to possible collisions at a certain node, which improves airport safety. A study on validating the log-normal assumption of aircraft taxi speed distributions is also given. We provide the link to code and data repository <span><span","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105540"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dynamic charging optimization for electric buses under photovoltaic-storage-grid energy supply mode","authors":"Yuting Ji ,&nbsp;Yiming Bie ,&nbsp;Dongfang Ma","doi":"10.1016/j.trc.2026.105539","DOIUrl":"10.1016/j.trc.2026.105539","url":null,"abstract":"<div><div>With the growing emphasis on sustainable transportation, the photovoltaic-storage-grid energy supply (PSG-ES) mode has been adopted in electric bus (EB) systems, showing promising performance and strong potential. However, environmental uncertainties cause random fluctuations in both solar output and energy demand, posing challenges to stable system operation. Most existing studies use static or low-frequency dynamic models based on day-ahead forecasts. These methods struggle to adapt to real-time changes, limiting the economic and environmental benefits of the PSG-ES mode. To address this issue, we propose a minute-level dynamic charging scheduling method for multi-route EB systems under the PSG-ES mode. The problem is formulated as a Markov Decision Process, with a penalty function and an action correction mechanism introduced to handle complex operational constraints. To improve learning and adaptability, we develop a deep reinforcement learning algorithm, featuring multi-head networks and composite experience replay to address distribution shifts and policy conflicts. Experiments using real-world EB data show that the proposed method effectively manages supply–demand uncertainties and generates more cost-efficient and environmentally sustainable charging plans. Compared to static scheduling with day-ahead forecasts, it reduces charging costs by 7.48% and carbon emissions by 2.99%.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105539"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dynamic on-demand delivery with spatial divisions of labor","authors":"Yue Yang ,&nbsp;André van Renssen ,&nbsp;Mohsen Ramezani","doi":"10.1016/j.trc.2025.105509","DOIUrl":"10.1016/j.trc.2025.105509","url":null,"abstract":"<div><div>The rapid evolution of on-demand delivery services has significantly influenced traditional logistics, particularly in urban areas where there is a surge in customer demand for timely and efficient delivery of small to medium-sized parcels. This paper investigates the concept of <em>spatial divisions of labor</em> in on-demand deliveries where the delivery network is partitioned into several operational regions, each managed by designated couriers. To facilitate parcel transshipment between regions, a set of accessible lockers is strategically placed at the shared borders of adjacent regions. We introduce and address a multi-hop delivery with spatial divisions of labor (MDSDL) problem, which involves dynamic parcel-courier dispatching and routing to minimize total operational costs. Within a rolling-horizon decision framework, the MDSDL problem is decomposed into two interdependent subproblems: (i) Region-Level Path Optimization (RLPO) that determines the coarse-grained, multi-hop path each parcel should take through the network, from its pickup region to its destination region, via the lockers, with the objective of minimizing delivery lateness. This path specifies the sequence of service regions a parcel must traverse to reach its final destination. (ii) Courier Route Optimization (CRO) that manages fine-grained, intra-region dispatching and routing by assigning incoming pickup and drop-off tasks to local couriers, who each has a continuously updated schedule. Subsequently, we develop a novel heuristic approach to dynamically solve RLPO and CRO in real time considering a rolling-horizon formulation. Extensive comparative experiments are conducted to demonstrate the advantages of the proposed approach. Implementing <em>spatial divisions of labor</em> not only enhances system efficiency and reduces operational costs but also improves the customer experience by reducing lateness and shortening ready-to-pickup times.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105509"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927801","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dynamic route redundancy-oriented strategic planning towards resilient transportation networks","authors":"Kai Qu ,&nbsp;Xiangdong Xu ,&nbsp;Anthony Chen","doi":"10.1016/j.trc.2025.105503","DOIUrl":"10.1016/j.trc.2025.105503","url":null,"abstract":"<div><div>Sufficient route redundancy ensures the availability of alternative routes during disruptions for critical trips (e.g., evacuations, relief transportation), which is essential for sustaining transportation network resilience. Existing studies on route redundancy focus mainly on assessment rather than optimization, and most adopt a static perspective that ignores the inherently dynamic nature of resilience. This study introduces the definition, evaluation, and optimization methods for dynamic network redundancy. We propose period-based metrics that account for travelers’ adaptive behavior and network congestion and develop a link-based day-to-day model under uncertainty coupled with Dial counting to consistently measure redundancy. A two-stage stochastic bi-level programming model is then formulated to identify investment strategies that maximize expected dynamic redundancy under uncertain disruptions. In the first stage, planners allocate budgets between link retrofitting and new link construction, while in the second stage, travelers dynamically reroute on a daily basis following disruptions. To solve the resulting non-convex optimization problem, we implement a Bayesian optimization framework with parallel scenario evaluation. Experiments on both test and real-world networks demonstrate the properties, features, and applicability of the proposed methods. Results indicate that accounting for dynamic traffic evolution and congestion can reduce redundancy estimates by up to 40% compared to static assessments in the 16-node test network, particularly under severe disruptions and high congestion levels. The spatial–temporal evolution of congestion patterns, which influences travelers’ perception of alternative routes, is naturally captured by dynamic redundancy but overlooked in static assessments. In the Anaheim network, increasing the budget from $300 million to $1200 million raises dynamic redundancy from 4.5 % to 10.1 %, illustrating diminishing marginal returns. The framework developed in this study provides a decision-support tool for more informed, resilience-oriented network planning.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105503"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145824128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deconstructing driving behaviors in interactions with pedestrians at uncontrolled crosswalks: an imitation learning method","authors":"Tao Wang ,&nbsp;Minh Kieu ,&nbsp;Chengmin Li ,&nbsp;Wenqiang Chen ,&nbsp;Ying-En Ge","doi":"10.1016/j.trc.2025.105508","DOIUrl":"10.1016/j.trc.2025.105508","url":null,"abstract":"<div><div>The objective of this paper is to deconstruct driving behaviors in interactions with pedestrians at uncontrolled crosswalks. Trajectory data are used to extract variables describing driver–pedestrian interactions, including position, acceleration, velocity, yaw rate, and interaction risk. Driving behavior is modeled as utility-driven, intelligent, and rational decision-making within the framework of a finite-state Markov decision process (MDP). The vanilla generative adversarial imitation learning (GAIL) framework is improved to reconstruct a human-like driving behavior model where the utility function is defined as the deviation between the agent’s behavior distribution and that of human drivers. Maximizing this utility through a deep reinforcement learning (RL) approach drives agents to progressively clone the behavioral policies of human drivers in the real world. The behavioral policy is formulated as a pre-trained driving behavior model and validated on a simulation platform for its ability in reproducing human driving behavior. Experimental results show that the model successfully reproduces the rationality of human drivers and generates human-like interaction trajectories in the simulation environment. Transfer experiments further demonstrate the generalizability of the pre-tained behavioral model. The interaction policy map and the state-value map are visualized to elucidate the generative mechanisms underlying human-like trajectories by revealing risk- and context-dependent layered patterns and latent behavioral preferences. This work contributes to the advancement of human-like behavioral models, thereby enhancing the fidelity of traffic microsimulation and improving behavior modeling in complex driver–pedestrian interactions.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105508"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146014568","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"On the applicability of time series anomaly detection methods to real-world traffic volume data","authors":"Iman Taheri Sarteshnizi,&nbsp;Majid Sarvi,&nbsp;Saeed Asadi Bagloee,&nbsp;Neema Nassir","doi":"10.1016/j.trc.2026.105536","DOIUrl":"10.1016/j.trc.2026.105536","url":null,"abstract":"<div><div>Time Series Anomaly Detection (TSAD or TAD) refers to the automatic and data-driven identification of abnormal segments in time series data, a task that has been studied extensively for decades. Despite recent transformative and novel findings revealed by efforts in this field, the literature on traffic anomaly detection has not yet fully reflected on these emerging trends to draw practical conclusions. In this paper, we focus on the applicability of state-of-the-art and well-established TAD methods to road traffic volume data, making contributions in two main ways. First, given the proven and major contribution of evaluation data to TAD outcomes, we argue that existing anomaly-labeled datasets from transportation and traffic systems require substantial enhancements in terms of both data size and label quality. To address this, we propose a new platform to inspect and label large-scale volume data of urban areas based on its unique characteristics and the latest taxonomy of time series anomalies. Second, based on the established framework, we also formulate the TAD problem in traffic volume data and introduce a discord-based, context-embedded, and light-weight traffic anomaly detection method, named Step-isolated Traffic Discords Discovery (Si-TDD), to address this problem. Benefiting from our labeling platform, AnoLT (Anomaly Labeled Traffic) is presented in this paper for the first time as a comprehensive, open-source, and anomaly-labeled spatiotemporal dataset collected from 147 locations across Melbourne, Australia. Comparative results with more than 20 baselines also indicate that Si-TDD considerably outperforms recent TAD solutions when it comes to traffic volume data, achieving a 67% F1 score with the AnoLT dataset. This paper highlights the key role of incorporating context-related information into existing TAD solutions to boost their effectiveness in traffic anomaly detection, a factor that is often overlooked in the current literature.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105536"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072518","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A joint, context-aware neural network-based travel demand and scheduling model","authors":"Joel Fredriksson,&nbsp;Anders Karlström","doi":"10.1016/j.trc.2025.105512","DOIUrl":"10.1016/j.trc.2025.105512","url":null,"abstract":"<div><div>Recent advancements in machine learning, and neural networks in particular, have introduced new opportunities for activity-based travel demand modeling and scheduling, providing data-driven alternatives to traditional theory-driven methods. While previous machine learning-based scheduling models have integrated combinations of activity, destination, and mode choice as separate sub-models, none have yet, to the best of our knowledge, unified these components into a single, jointly learned framework.</div><div>This paper introduces Skyline-NNjoint, a novel fully neural network-based scheduling model that jointly predicts an agent’s activity, destination, and mode choice decisions at each discrete time step throughout the day. To capture substitution effects and interdependencies among alternatives, the model introduces a Global Context Module (GCM) that enables each alternative to adjust its attractiveness based on the context of all others. While similar context-based approaches have been used in other domains, this is, to the best of our knowledge, the first application of such a mechanism in travel demand modeling. This provides a data-driven approach to relax the Independence of Irrelevant Alternatives (IIA) assumption inherent in multinomial logit models. The effectiveness of the GCM is evaluated by comparing Skyline-NNjoint to a baseline version without it, isolating its contribution to model performance.</div><div>The model is trained on travel survey data from Stockholm and evaluated using both cross-entropy loss and simulated daily activity–travel trajectories. Cross-entropy loss results confirm that the GCM improves predictive performance. Simulation results show that Skyline-NNjoint produces patterns of activity participation, trip timing, and mode choice that closely match observed data. Notably, the model accurately reproduces mode distributions across activity purposes, highlighting its capacity to capture interdependencies in joint decision-making.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105512"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978362","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-reservoir traffic dynamics: Outflow network MFD and state estimation with sparse traffic data","authors":"Omid Mousavizadeh,&nbsp;Mehdi Keyvan-Ekbatani","doi":"10.1016/j.trc.2025.105504","DOIUrl":"10.1016/j.trc.2025.105504","url":null,"abstract":"<div><div>This study introduces an innovative framework for traffic state estimation in multi-reservoir networks, tackling the challenges posed by sparse data in urban traffic networks. By integrating Floating Car Data (FCD) and Loop Detector Data (LDD), the framework estimates traffic dynamics, such as inflows, outflows, transfer flows, and accumulation, without requiring detailed trip information or path flow distribution methods. A step forward in this study is a method for the estimation of internal outflow and transfer flows, enabling the direct estimation of outflow Network Macroscopic Fundamental Diagrams (outflow-NMFDs) which overcomes the shortcomings of outflow-NMFD estimation through the full set of sensors or indirect estimation via the production-NMFD. As argued, the outflow-NMFD is deemed the more robust foundation for the accumulation-based modelling in multi-reservoir systems. Accordingly, the estimated outflow-NMFD is used to build the accumulation-based model, in which the model outputs are combined with sparse real-time measurements, improving the model’s alignment with actual traffic conditions. The framework’s adaptability allows it to function effectively under varying levels of probe vehicle penetration, making it suitable for real-world scenarios where data availability can be inconsistent or limited. Simulation results validate the model’s robustness in capturing both steady-state and dynamic traffic behaviours, maintaining high accuracy even during abrupt demand changes. Furthermore, the framework performs reliably under stochastic conditions, demonstrating its resilience to daily traffic fluctuations. By reducing dependence on widespread sensor deployment across the network or its boundaries, this cost-effective approach offers a practical solution for real-time traffic monitoring and management in multi-reservoir systems, even when the boundaries of the system are not fixed in the network.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105504"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Uncertainty quantification for joint demand prediction of multi-modal ride-sourcing services using spatiotemporal Mixture-of-Expert neural network","authors":"Xiaobing Liu ,&nbsp;Yu Duan ,&nbsp;Yangli-ao Geng ,&nbsp;Yun Wang ,&nbsp;Qingyong LI ,&nbsp;Xuedong Yan ,&nbsp;Ziyou Gao","doi":"10.1016/j.trc.2025.105507","DOIUrl":"10.1016/j.trc.2025.105507","url":null,"abstract":"<div><div>Given the sparse uncertainty and highly imbalanced distribution of origin-destination demand in ride-sourcing, probabilistic prediction methods offer rich information to enhance operation decision-making reliability. However, existing uncertainty quantification methods face critical limitations: parametric approaches struggle with irregularities of real-world demand distributions. In contrast, nonparametric methods based on quantile regression are scarcely implemented and suffer from quantile crossing, compromising probabilistic consistency. Furthermore, current approaches seldom consider the crucial inter-service correlations among ride-sourcing services. This paper introduces the Spatiotemporal Mixture-of-Experts Residual-based Multi-quantile Regressive Network (ST-MoE-RMQRN).<span><span>¹</span></span> This novel hybrid framework integrates distribution-free quantile estimation with service correlation modeling. Our three main contributions are: 1) a Residual-based Multi-quantile Regressor that employs monotonicity-constrained residual fitting to ensure non-crossing quantiles while preserving distributional flexibility; 2) an environment-aware MoE architecture that captures service correlations through multiple shared-expert networks, addressing potential demand switches among service modals with sensitivity to temporal heterogeneity and exogenous contextual factors such as weather conditions; 3) comprehensive validation on two real-world ride-sourcing datasets demonstrating that our approach consistently outperforms recent probabilistic predictive models in both deterministic and probabilistic forecast metrics, including the Spatiotemporal Zero-Inflated Negative Binomial Graph Neural Network (STZINB-GNN). Additionally, data-wise and component-wise ablation studies confirm the effectiveness of the entire framework and each part of it, showing that explicitly modeling inter-service dependencies substantially improves both deterministic accuracy and probabilistic calibration. Our model is highly efficient in representing extreme demand fluctuations and data-sparse scenarios, which promise to address existing operational inefficiencies faced by transitioning Transportation Network Companies from reactive to anticipatory decision-making.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105507"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885537","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Resilient multi-agent reinforcement learning for centralised tactical conflict resolution under uncertain perturbations and non-cooperative traffic in urban air mobility","authors":"Rodolphe Fremond ,&nbsp;Yan Xu ,&nbsp;Junjie Zhao ,&nbsp;Antonios Tsourdos ,&nbsp;Gokhan Inalhan","doi":"10.1016/j.trc.2026.105542","DOIUrl":"10.1016/j.trc.2026.105542","url":null,"abstract":"<div><div>This research investigates tactical conflict resolution for Unmanned Aircraft Systems (UAS) and Urban Air Mobility (UAM) operations under degraded conditions and in the presence of non-cooperative UAS/UAM and manned Commercial Air Transportation and General Aviation (CAT/GA) intruders. The study adopts a centralised safety-net approach within UAS Traffic Management (UTM) architectures, envisioning ground-based conflict resolution services. We propose a set of Tactical Conflict Resolution Solvers (TCRS), each built upon a Multi-Agent Reinforcement Learning (MARL) core using a shared-policy transformer architecture and executed in a decentralised manner. To assess resilience of TCRS variants, we introduce domain-specific perturbations, including positioning noise, communication loss, and sensor-related defects. The TCRS operates with partial decision-making ability in non-cooperative traffic environments, while the perturbation model increases realism by simulating varying degrees of information availability. Results show that the perturbation-trained models achieve substantial safety gains compared with the baseline TCRS trained in ideal conditions. The most resilient variant; trained under multi-perturbation exposure and evaluated in non-cooperative environments, achieves a threefold reduction in critical safety violations compared with the baseline and remains robust under mixed cooperative/non-cooperative traffic with static intent. It exhibits a modest vulnerability under fully homogeneous non-cooperative scenarios with dynamic intent. Simulations involving concurrent CAT/GA and UAS operations further indicate that integrating UAS operations within the existing airspace classification remains hazardous for ground-based tactical conflict resolution when constrained by short look-ahead horizons and insufficient time to react.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"184 ","pages":"Article 105542"},"PeriodicalIF":7.6,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078071","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}