Transportation Research Part C-Emerging Technologies最新文献

{"title":"Multi-element probabilistic collocation solution for dynamic continuum pedestrian models with random inputs","authors":"Zepeng Liu ,&nbsp;S.C. Wong ,&nbsp;Liangze Yang ,&nbsp;Chi-Wang Shu ,&nbsp;Mengping Zhang","doi":"10.1016/j.trc.2025.105104","DOIUrl":"10.1016/j.trc.2025.105104","url":null,"abstract":"<div><div>This study focuses on dynamic continuum pedestrian flow models with random inputs, which can be represented by sets of partial differential equations with some modeling parameters being randomized. Under random conditions, the model outputs are no longer fixed and may differ appreciably from their respective average levels. Simulating the resulting distribution is important as it helps quantify the effects of uncertainties on traffic behaviors when evaluating walking facilities. Through two examples based on continuum models, the effect of random inputs on pedestrian flow propagation is qualitatively analyzed. Crowd evacuation is found to be effective in reducing the variation and risk produced by randomness, while congestion is observed to significantly increase the uncertainty within the system. For a general system without an explicitly known exact solution, an existing efficient solver — the multi-element probabilistic collocation method (ME-PCM) — is introduced to derive the solution distribution numerically. The ME-PCM is non-intrusive and flexible and has no limitations in terms of governing partial differential equations and the numerical schemes for solving them. The ME-PCM’s use of element-wise local orthogonal polynomials to represent the solution enables it to converge efficiently even if shocks occur during the modeling period. As a demonstration case, the well-known Hughes model is applied in a numerical example with a corridor and an obstacle. The demand at the inflow boundary is randomized to a lognormal distribution that represents day-to-day demand stochasticity. The results indicate that the ME-PCM’s solution converges more rapidly than those of the Monte Carlo and generalized polynomial chaos methods. Statistical information on pedestrian density is derived from the ME-PCM solution and can be used to identify the locations in walking facilities where the average pedestrian density is moderate but where exceptional congestion with a large variance can occur. This successful application shows the possibility of quantifying the uncertainty in pedestrian flow models using the ME-PCM. The proposed approach can also be applied to models with other similar random inputs, given that a well-established algorithm for deterministic cases is available.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105104"},"PeriodicalIF":7.6,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143643006","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sparse Gaussian process-based strategies for two-layer model predictive control in autonomous vehicle drifting","authors":"Cheng Hu ,&nbsp;Yangyang Xie ,&nbsp;Lei Xie ,&nbsp;Michele Magno","doi":"10.1016/j.trc.2025.105065","DOIUrl":"10.1016/j.trc.2025.105065","url":null,"abstract":"<div><div>Vehicle safety is paramount in autonomous driving, particularly when managing vehicles at extreme side-slip angles—a challenge often overlooked by conventional controllers. Recent studies have focused on vehicle drift control under such extreme conditions. However, tracking complex trajectories while drifting is challenging, especially in the presence of a model mismatch. This paper proposes a two-layer model predictive controller based on sparse variational Gaussian processes. The first layer is responsible for computing the optimal drift equilibrium points, while the second layer is tasked with tracking these points. A variational free energy-based Gaussian process is utilized to compensate for errors in the upper-layer drift equilibrium point calculations and mismatches in the lower-layer controller model. Moreover, the vehicle’s state is determined to be either in transit drift or deep drift based on whether the slip angle and steering angle have reached critical values. Gaussian models are established for each state to enhance prediction accuracy. The effectiveness of the controller is demonstrated through joint simulations on MATLAB and CarSim platforms. First, the proposed two-layer model predictive controller was compared with three state-of-the-art drift controllers, demonstrating at least a 48.64% reduction in average lateral error when tracking trajectories with varying curvature. Second, when combined with sparse Gaussian processes, the controller’s learning ability was validated in scenarios with a 5% to 20% friction coefficient mismatch. Specifically, in the scenario with a 20% friction coefficient mismatch, its average lateral error was reduced by 95.09% after model error learning. Additionally, the controller was compared with both Fully Independent Training Conditional (FITC) GP-based MPC and Full GP-based MPC controllers, demonstrating better trajectory tracking capability and model error learning ability.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105065"},"PeriodicalIF":7.6,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143680625","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 multi-layer multiclass cell transmission model for modeling heterogeneous traffic flow","authors":"Jialin Liu ,&nbsp;Rui Jiang ,&nbsp;Yang Liu ,&nbsp;Shiteng Zheng ,&nbsp;Bin Jia ,&nbsp;Hao Ji","doi":"10.1016/j.trc.2025.105099","DOIUrl":"10.1016/j.trc.2025.105099","url":null,"abstract":"<div><div>Modeling heterogeneous traffic with different free-flow speeds poses theoretical challenges for the existing multiclass cell transmission models (MCTMs) with identical cells. Specifically, the existing MCTMs face two challenges due to the numerical diffusion: delaying flow and rushing flow. These challenges arise because slow vehicles cannot travel through a cell within a time interval under free-flow conditions, leading to inaccurate estimates of cell occupancy and travel time. To address these challenges, we propose a Multi-layer Multiclass Cell Transmission Model (MMCTM) with multi-size cells. Firstly, a road is divided into a multi-layer multi-size cell network based on different free-flow speeds of multiclass vehicles. Class-specific vehicles move within a class-specific cell network, while other class vehicles are equivalently projected into class-specific cells using density mutual projection formulas. Secondly, we formulate the flow propagation rules for multiclass vehicles based on the original rules of CTM and conversion coefficients of different classes. We prove the capability of the MMCTM by showing that it can avoid unrealistic situations where the densities in the cells are negative or exceed the maximum density. Finally, numerical experiments demonstrate that our proposed model can effectively address the two challenges and reproduce essential phenomena of mixed traffic flow, such as the moving bottleneck effect of slow vehicles, shockwave propagation, overtaking, FIFO, and oscillatory waves. In particular, the MMCTM can reproduce the drop and resurge of the discharge rate of fast vehicles. Furthermore, we calibrate and validate the MMCTM using NGSIM I-80 dataset and the I-24 MOTION dataset. The results indicate that (1) our proposed model improves the estimation accuracy of travel time and cell occupancy for multiclass vehicles; (2) the MMCTM outperforms the general MCTM with identical cells (GMCTM) under traffic congestion conditions.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105099"},"PeriodicalIF":7.6,"publicationDate":"2025-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143643005","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":"Integrating battery-related decisions into truck-drone tandem delivery problem with limited battery resources","authors":"Zhongshan Liu ,&nbsp;Bin Yu ,&nbsp;Tingting Chen ,&nbsp;Li Zhang","doi":"10.1016/j.trc.2025.105082","DOIUrl":"10.1016/j.trc.2025.105082","url":null,"abstract":"<div><div>The truck-drone tandem delivery mode provides a promising application in last-mile package delivery but is limited by the duration of drones. The battery swap strategy is a widely adopted approach to extend the cruising ranges of drones, ensuring that depleted batteries can be swapped for fully charged ones in a matter of minutes. However, most existing studies assume that there are sufficient batteries available at the depot, which is impractical as storing a large number of batteries is expensive. To bridge this gap, this paper considers the truck-drone tandem delivery problem with a battery swap strategy, under the condition of a limited number of batteries. To address the challenges posed by the practical limitation of the number of batteries, we propose a joint optimization problem integrating two types of interdependent decisions, i.e., battery-related decisions and route-related decisions. The battery-related decisions identify which batteries to be installed on drones and establish optimal battery charging schedules at the depot. And the route-related decisions determine the truck-drone tandem delivery routes. The studied joint optimization problem is formulated as a mixed integer linear programming model, and this model is integrated into a well-designed adaptive large neighborhood search algorithm to determine the two types of decisions. Specifically, on the basis of traditional operators, we design a series of depot-related operators tailored to the feature of route-related decisions. Furthermore, regarding the features of battery swapping and charging schedules, we introduce a novel battery operator to determine optimal battery-related decisions. The numerical experiments show that introducing the battery-related decisions can bring flexible battery schedules when the total number of batteries is limited. The effects of battery capacity, charging rate, charging cost, drone speed, and the number of batteries and drones are analyzed to provide practical suggestions for companies.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105082"},"PeriodicalIF":7.6,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143631945","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":"Scaling laws of dynamic high-capacity ride-sharing","authors":"Wang Chen ,&nbsp;Linchuan Yang ,&nbsp;Xiqun Chen ,&nbsp;Jintao Ke","doi":"10.1016/j.trc.2025.105064","DOIUrl":"10.1016/j.trc.2025.105064","url":null,"abstract":"<div><div>This study discovers a few scaling laws that can effectively capture the key performance of dynamic high-capacity ride-sharing through extensive experiments based on real-world mobility data from ten cities. These scaling laws are concise and contain only one dimensionless variable named system load that reflects the relative magnitude of demand versus supply. The scaling laws can accurately measure how key performance metrics such as passenger service rate and vehicle occupancy rate change with the system load. The scaling laws strongly agree with experimental results, with the values of <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> exceeding 0.95 across all scenarios. In addition, the scaling laws can accurately reproduce experimental results of dynamic high-capacity ride-sharing involving different road networks, supply–demand patterns, vehicle capacities, and matching algorithms, indicating these scaling laws could be general and applied to other cities. These scaling laws provide a reference for transportation network companies and governments to efficiently manage dynamic ride-sharing services. For example, according to these scaling laws, when the demand is relatively high, e.g., system load equals 3, ride-sharing services with a capacity of 2 passengers can only accommodate 50% of demand. In comparison, high-capacity ride-sharing services with a capacity of 4 passengers can satisfy 72% of demand. The findings provide valuable insights into the expected performance of ride-sharing, informing decisions about how to operate a fleet to improve transportation efficiency.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105064"},"PeriodicalIF":7.6,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143631946","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":"Optimal pricing and vehicle allocation in local ride-sharing markets with user heterogeneity","authors":"Wen-Xiang Wu ,&nbsp;Rui Sun ,&nbsp;Xiao-Ming Liu ,&nbsp;Hai-Jun Huang ,&nbsp;Li-Jun Tian ,&nbsp;Hua-Yan Shang","doi":"10.1016/j.trc.2025.105084","DOIUrl":"10.1016/j.trc.2025.105084","url":null,"abstract":"<div><div>A ride-sharing platform (RSP) typically provides both solo and pooled ride services to passengers. Passengers opting for pooled rides pay a lower fare but generally experience longer travel times. Pooled ride services gain from improving occupancy per car, thereby serving more passengers, but this comes at the cost of a lower profit margin per passenger compared to solo ride services. We develop a stochastic queueing model to characterize the user equilibrium in a local on-demand market for solo and pooled ride services. In this model, passengers have heterogeneous values of time (VOTs), and drivers operate as independent agents. We find that in equilibrium, the VOT threshold value regulated by the set trip fares for solo and pooled ride services determines passengers’ travel mode choices. Specifically, passengers with lower VOTs than the threshold value choose to pool, while the others choose to ride alone. Built upon the user equilibrium, we then design customized optimal pricing and vehicle allocation strategies to maximize the total expected revenue of the RSP. This approach adapts the revenue-maximizing pricing and vehicle allocation strategies to a specific local ride-sharing market. It achieves this customization by considering factors such as users’ VOTs, supply and demand levels, spatial distances, and prevailing traffic conditions. Numerical results demonstrate that, in situations of high demand but limited supply, our proposed optimal pricing and vehicle allocation strategy effectively maximizes the total expected revenue of the RSP in the presence of spatial–temporal variations in ride-sharing demand. In such scenarios, solo ride prices are set at higher levels, and a majority of idle vehicles are allocated to solo passengers. Conversely, when demand is low but supply is sufficient, combining the optimal pricing strategy with a proportional vehicle allocation strategy also nearly maximizes the total expected revenue. In this case, the optimal vehicle allocation strategy is deemed non-critical due to the surplus supply. Solo ride prices are adjusted differently than those in high-demand situations to incentivize solo ride selection while discouraging pooled rides, ultimately resulting in the highest total expected revenue.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105084"},"PeriodicalIF":7.6,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143637682","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 data-driven MPC approach for virtually coupled train set with non-analytic safety distance","authors":"Xiaolin Luo ,&nbsp;Dongming Wang ,&nbsp;Tao Tang ,&nbsp;Hongjie Liu","doi":"10.1016/j.trc.2025.105087","DOIUrl":"10.1016/j.trc.2025.105087","url":null,"abstract":"<div><div>Resorting to the emerging virtual coupling technology, multiple train units can operate as a virtually coupled train set (VCTS) to improve the flexibility and efficiency of train operations. To strictly guarantee collision avoidance, the space–time separation principle should be employed, where a non-analytic safety distance is used to safely separate units among VCTS. Thus, the formation control of VCTS is challenging, since it lacks analytic models to tune controllers with tracking accuracy and computational efficiency. To solve this problem, this paper proposes a data-driven model predictive control (DDMPC) approach. Based on a database with previously measured VCTS trajectories, we present a linear data-driven model to describe the non-analytic VCTS formation, such that the controller of DDMPC is yielded by solving a quadratic programming problem in a computationally efficient way. Next, to improve tracking accuracy, we optimize the modeling accuracy in the cost function of DDMPC, and bound the uncertainties from data-driven modeling and coupled states of VCTS. Furthermore, sufficient conditions are derived to guarantee constraint satisfaction and stability for VCTS. Finally, the advantages of the proposed DDMPC approach are demonstrated by comparing with several approaches in tracking accuracy and computational efficiency.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105087"},"PeriodicalIF":7.6,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143629205","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":"Advancing lane formation and high-density simulations in bidirectional flow: A humanoid pedestrian model incorporating gait dynamics and body rotation","authors":"Xiaoyun Shang ,&nbsp;Rui Jiang ,&nbsp;S.C. Wong ,&nbsp;Ziyou Gao ,&nbsp;Wenguo Weng","doi":"10.1016/j.trc.2025.105086","DOIUrl":"10.1016/j.trc.2025.105086","url":null,"abstract":"<div><div>Current bidirectional pedestrian flow models face challenges in accurately simulating lane formation and high-density conditions. This study addresses these issues by developing an improved humanoid pedestrian model (HPM), which extends the applicability of the original HPM from one-dimensional to two-dimensional scenarios and offers a more realistic simulation of pedestrian behavior. The improved HPM incorporates two distinct gaits—walking while rotating and walking while turning, which capture the complex dynamics of human walking—and an innovative gait-planning process. Additionally, a novel energy-based heuristic rule that considers factors such as deviation from the target direction, body rotation to navigate gaps, and reduced walking velocity is introduced. The energy expression is designed according to the form of mechanical energy, with no parameters requiring calibration. This design enables our model to demonstrate, to some extent, that pedestrians determine their walking direction by minimizing mechanical energy consumption. Simulations are conducted under conditions replicating previous experiments to validate the improved HPM against both experimental results and two classic models, namely the heuristic-based model and the social force model. The improved HPM shows minimal trajectory deviation; effectively replicates body rotation that facilitates efficient lane formation; and transitions swiftly from a randomized flow to stable, well-ordered flow patterns. Moreover, the improved HPM achieves a maximum density of 7 ped/m², representing a significant advancement in modeling high-density scenarios. Overall, the improved HPM offers deep insights into the crowd dynamics of bidirectional flow and thereby improves the accuracy of simulations in high-density situations.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105086"},"PeriodicalIF":7.6,"publicationDate":"2025-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143619226","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Two-step optimization of train timetables rescheduling and response vehicles on a disrupted metro line","authors":"Hui Wang ,&nbsp;Feng Li ,&nbsp;Jialin Liu ,&nbsp;Hao Ji ,&nbsp;Bin Jia ,&nbsp;Ziyou Gao","doi":"10.1016/j.trc.2025.105078","DOIUrl":"10.1016/j.trc.2025.105078","url":null,"abstract":"<div><div>Metro disruption management is currently one of the hot issues in metro research. Existing research has primarily focused on rescheduling normal train timetables or the design of bus bridging services, with limited consideration of the traffic dynamics. In this paper, we introduce a two-step optimization framework to derive a comprehensive evacuation plan encompassing the rescheduled train timetable and the response vehicle scheduling scheme. In the first step, an integer programming model is proposed to reschedule the normal train timetable. The objective function of this model is to minimize total passenger waiting time while considering various constraints such as the timetable rescheduling strategies (i.e., cancellation and short-turning), train headway, and train capacity. In the second step, the response vehicle scheduling model is established based on the Cell Transmission Model (CTM). This model aims to minimize the total travel time of the response vehicles and is capable of capturing traffic dynamics on the evacuation network. To bridge the gap between the mathematical models of the first and second steps, we establish a demand transformation process, which provides a formula for transforming the stranded passenger demand into the demand for response vehicles. Numerical cases of Beijing Metro Line 9 verify the efficiency and effectiveness of our proposed model, and results show that: (1) the direction with fewer train services experiences a greater impact from the disruption. The disruptions occurring within the central region of the metro line tend to affect a greater number of normal train services during peak hours, whereas disruptions occurring within the terminal areas of the metro line tend to affect a greater number of normal train services during off-peak hours; (2) compared with the static shortest route scheme, the dynamic shortest routes of response vehicles contribute a 7% reduction in total travel time.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105078"},"PeriodicalIF":7.6,"publicationDate":"2025-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143619225","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":"Coordinating ride-pooling with public transit using Reward-Guided Conservative Q-Learning: An offline training and online fine-tuning reinforcement learning framework","authors":"Yulong Hu ,&nbsp;Tingting Dong ,&nbsp;Sen Li","doi":"10.1016/j.trc.2025.105051","DOIUrl":"10.1016/j.trc.2025.105051","url":null,"abstract":"<div><div>This paper introduces a novel reinforcement learning (RL) framework, termed Reward-Guided Conservative Q-learning (RG-CQL), to enhance coordination between ride-pooling and public transit within a multimodal transportation network. We model each ride-pooling vehicle as an agent governed by a Markov Decision Process (MDP), which includes a state for each agent encompassing the vehicle’s location, the number of vacant seats, and all pertinent information regarding the passengers on board. We propose an offline training and online fine-tuning RL framework to learn the optimal operational decisions of the multimodal transportation systems, including rider-vehicle matching, selection of drop-off locations for passengers, and vehicle routing decisions, with improved data efficiency. During the offline training phase, we develop a Conservative Double Deep Q Network (CDDQN) as the action executor and a supervised learning-based reward estimator, termed the Guider Network, to extract valuable insights into action-reward relationships from data batches. In the online fine-tuning phase, the Guider Network serves as an exploration guide, aiding CDDQN in effectively and conservatively exploring unknown state–action pairs to bridge the gap between the conservative offline training and optimistic online fine-tuning. The efficacy of our algorithm is demonstrated through a realistic case study using real-world data from Manhattan. We show that integrating ride-pooling with public transit outperforms two benchmark cases—solo rides coordinated with transit and ride-pooling without transit coordination—by 17% and 22% in the achieved system rewards, respectively. Furthermore, our innovative offline training and online fine-tuning framework offers a remarkable 81.3% improvement in data efficiency compared to traditional online RL methods with adequate exploration budgets, with a 4.3% increase in total rewards and a 5.6% reduction in overestimation errors. Experimental results further demonstrate that RG-CQL effectively addresses the challenges of transitioning from offline to online RL in large-scale ride-pooling systems integrated with transit.</div></div>","PeriodicalId":54417,"journal":{"name":"Transportation Research Part C-Emerging Technologies","volume":"174 ","pages":"Article 105051"},"PeriodicalIF":7.6,"publicationDate":"2025-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610246","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}