ABSTRACT. Traffic flow modeling is crucial for transportation engineering and urban planning, yet obtaining high-quality trajectory data remains challenging. Traditional methods, such as stationary cameras, have limitations in complex urban environments. Existing datasets often lack coverage and resolution, hindering comprehensive analysis. To address this, we utilize drone technology for data collection, focusing on a 900m freeway section in Milan, Italy. Unlike previous datasets, ours offers coverage across all traffic states. We collected over 100,000 vehicle trajectories, enabling analysis of driving behavior and traffic dynamics. Using Density-Based Spatial Clustering, we identified lanes and analyzed lane-changing behavior. Nearly half of the vehicles executed lane changes, with motorcycles showing the highest proportion. Time-space plots illustrate traffic dynamics across different flight campaigns, from free-flowing to congested conditions. Our approach overcomes the limitations of existing datasets; the data will provide valuable insights for traffic modeling, including lane changes, merges, and diverges, offering comprehensive coverage of freeway dynamics and benefiting transportation research and urban planning efforts.

ABSTRACT. Urban air mobility (UAM) is an emerging mode that uses low-altitude airspace to provide point-to-point air travel services. Recent advances in electric vertical take-off and landing vehicles are increasing attention on UAM for its potential to alleviate roadway traffic congestion. Given the spatial heterogeneity of land use in most cities, large-scale UAM will likely be deployed between specific urban areas, for example, from the suburbs to city centers. However, large-scale UAM travel between a few origin-destination pairs increases the risk of aircraft collisions and air traffic congestion, especially at airline intersections. To address this, this work proposes a hybrid framework of traffic simulation and management for large-scale UAM. The framework achieves an elegant trade-off between air traffic safety and efficiency by combining route guidance and collision avoidance for UAM aircraft. With a centralized strategy, route guidance provides system optimal paths (composed of waypoints) for aircraft, aiming to minimize total travel time. With a distributed strategy, collision avoidance generates trajectories between given waypoints, ensuring aircraft safety separation. To the best of our knowledge, this work is one of the first to introduce both dynamic route guidance and collision avoidance for UAM. The results highlight that the framework can effectively prevent air traffic congestion and provide flexible UAM operations, e.g., dynamic airspace access management. The proposed framework has demonstrated great potential for large-scale UAM simulation and management.

ABSTRACT. Unmanned Aerial Vehicles (UAVs) offer advantages as traffic monitoring sensors compared to traditional sensors, such as enhanced quality of measurements. However, the scalability of such systems for large urban traffic road networks has been questioned. To circumvent this issue, we propose a UAV-based Traffic Estimation System (UAV-TES), which leverages real-time traffic measurements from UAVs to estimate traffic densities across observed and unobserved areas of an urban traffic road network. This is achieved through Gaussian Process modelling to handle data sparsity, as well as moving horizon estimation that incorporates non-linear, a-priori knowledge of traffic dynamics. The proposed solution is validated through macroscopic simulations. The results show that accurate traffic density estimates are achieved in real-time even under challenging conditions such as noisy measurements and sparsity of data due to a limited number of UAVs.

ABSTRACT. In the highly dynamic landscape of Urban Air Mobility (UAM), ensuring safety in dense urban and regional areas is a critical challenge. This study introduces a novel strategy to create a dependable intermodal safety network through dynamic and adaptable ground traffic control. The core inquiry investigates the potential for intersections to be used as emergency landing areas, and whether intersections can be safely used with appropriate ground/city traffic management. This traffic management system could be implemented through centralized coordination of traffic across the city, car-to-car communication, or a network of interconnected traffic junctions with sense and react capabilities.

ABSTRACT. In the attached pdf as per the guidelines.

ABSTRACT. In recent years, there has been a gradual shift from conventional survey-based data-gathering approaches to GPS-based due to more precise data from passengers' trips that can be extracted from nomadic devices such as smartphones. Inferring travel modes is essential to understand passengers' travel behavior and various machine and deep learning techniques have shown their ability to extract valuable details from GPS data. In this paper, we develop a convolutional neural network to extract high-level features from GPS data to recognize transport modes of trips. Our model improved the final accuracy of test data by 2.84 percent. Moreover, Determining true value of passengers waiting time, walking time and distance to and from public transit stops from passengers and public transit side is a key index to assessing potential demand, quality, and effectiveness of public transit services. Current studies heavily rely on household surveys, direct observation techniques, GIS, and telephone-based interviews. However, some investigations use GPS trajectories as a primary data source to detect trip phases, and the main drawback is saving a few trips. This study as a first investigation in this field tries to detect trip phases to and from public transit stops for both passengers and the infrastructure side with an automated trip phase recognition algorithm.

ABSTRACT. This study presents a machine learning (ML) meta-model designed to enhance traffic monitoring by efficiently predicting key performance indicators (KPIs) of intersections using vehicle traffic data. Our model integrates static and sequential ML models to process traffic sequences and intersection features. The meta-model was trained and validated using synthetic data generated through microscopic simulation. Results demonstrated the model's capability to consistently predict intersection performance with relatively high accuracy, showing potential for real-time applications in large-scale urban networks. The use of machine learning methods, enabling GPU acceleration, significantly reduced the model's computation times in comparison to full scale simulations, suggesting our model can complement and enhance microscopic simulations in real-time monitoring applications. This approach promises to improve urban traffic management and sustainability by enabling fast and efficient monitoring of intersections in large-scale urban environments while minimising the computational resources required.

ABSTRACT. Eco-driving is a practical strategy for mitigating greenhouse gas emissions in urban transportation systems. By integrating eco-driving principles, intelligent transportation systems (ITS) can nudge drivers towards greener and more fuel-efficient behavior. This work introduces a novel learning-based incentive design that promotes eco-driving within the framework of reverse Stackelberg games. We propose deep reinforcement learning-based optimal incentives that influence human drivers in a transportation network to adopt eco-driving principles. These incentives are designed to ensure compliance by human drivers while adhering to the predetermined budget constraints of a system operator. It is typically challenging to model complex traffic dynamics, drivers' interaction, and their response to incentives. Therefore, we propose an approach where incentives are applied within policy parameters rather than directly influencing specific actions. We first present a learning-based incentive design under the framework of reverse Stackelberg games that effectively captures and navigates the complexities of urban traffic dynamics. Additionally, we integrate a regret minimization method to accurately model drivers' choices. Our results show the effectiveness of learning-based incentives for adopting eco-driving, significantly reducing overall emissions.

ABSTRACT. In major cities, the rapidly growing urban population and fast-expanding e-commerce activities has presented significant challenges to both passenger transportation and city logistics, especially for last-mile passenger movement and goods delivery. It is particularly evident during morning and evening peak hours when an influx of commuters seeks transportation between their homes and the nearest public transportation station. The surging demand requires a large fleet of vehicles to do a demand-responsive transport service, while surplus transport capacity during off-peak hours is in a waste, leading to inefficiencies in utilizing the available transport resources. To address this challenge, this paper proposes a collaborative passenger and parcel transport service. This innovative approach allows for the transportation of both passengers and small parcels in the same vehicle. Priority is given to urgent passengers or parcels during peak hours, while others are served during off-peak hours by utilizing the spare capacity. Hence, it is crucial for operators to establish an optimal routing plan for the passenger-parcel transport.

However, the dynamic and fluctuating demand brings difficulties in making optimal routing plans. Since the demand is unknown before occurring, decisions are made based on incomplete information. The limited visibility makes it challenging to consider future impacts, impeding the optimization of routing plans. It necessitates a strategy to do routing plan in a dynamic environment. In this study, we deal with the dynamic vehicle routing problem with time windows for passenger-parcel transport of last-mile delivery (DVRPTW-PPL) to maximize profit.

Vehicle routing problems have been well studied in the literature, but few papers focus on dynamic vehicle routing problem due to the difficulties in designing an effective algorithm. Most of them used greedy methods which is short-sighted. To address such a complex optimization problem with long-term vision, deep reinforcement learning (DRL) is powerful, that makes decisions considering future impact. To the best of our knowledge, we are the first to design a DRL framework specifically for DVRPTW-PPL. Besides, we design a novel dynamic-attention model in this DRL framework. Attention model (Kool et al., 2018) is influential in feature learning, greatly aiding in decision-making, and has never been applied to dynamic routing problems. This algorithm designs a dynamic encoder-decoder architecture with attention layers to iteratively generate feasible routing solutions. The attention layer masks irrelevant features and learns by adaptively assigning weights to different parts of the input, enabling efficient and accurate learning of decision policies. Additionally, this algorithm addresses the challenge of making decisions considering future demand, which is hard to achieve by optimization models and greedy algorithms. Notably, after offline training in ten thousand instances, the DRL framework demonstrates remarkable speed in generating solutions and shows good quality.

ABSTRACT. Accurate traffic state prediction is important for effectively managing traffic flow during peak hours or in the event of an accident or road closure. The quest for accurate traffic forecasts faces challenges due to the spatial and temporal dependencies that characterize traffic flow across different network segments. Whereas traditional deep learning methods mainly rely on traffic data from isolated sections of the road network for traffic forecasting, recent advancements leverage information from the entire road network to enhance prediction accuracy. This study aims to predict traffic flow using historical traffic flow data, spanning from 1 August 2023 to 1 February 2024 (six-month period), from traffic detectors located at the city of Ioannina, Greece. Employing a phased deep learning approach, we initially explore traffic predictions using basic deep learning models and 1-minute resolution data from a single detector. Subsequently, we employ a Graph Convolutional Network (GCN) to harness 1-minute resolution data from three detectors. In a final effort to refine our predictions, we aggregate the original data into 5-minute intervals and reassess the GCN model’s performance.

ABSTRACT. included as pdf

ABSTRACT. Recently, the demand for traffic simulation has shifted to large-scale urban areas. Among the various levels of traffic simulation, microscopic simulations provide the most accurate and detailed results. However, their high computational cost prevents them from being applied to the large-scale urban network. In this manner, the need to find a compromise between accuracy and computational cost leads to hybrid or multi-level traffic simulation. Multi-level traffic simulation integrates different simulation levels within a single framework. Typically, micro-macro hybrid simulation is used for highways and micro-meso hybrid simulation is suitable for urban networks. Furthermore, in order to maintain a constant simulation performance in terms of accuracy and computational cost, the importance of dynamic properties that change the simulation levels of road segments in response to the traffic conditions over time has been emphasized. Therefore, this study proposed a dynamic multi-level traffic simulation for urban areas by combining microscopic and mesoscopic traffic simulations. A simulation framework and data structure were proposed to ensure compatibility and consistency between different levels of simulations. Temporal and spatial interfaces were modeled and verified for proper functioning, including the preservation of vehicle information and consistency in traffic dynamics. The developed simulation was evaluated in terms of computational cost and accuracy using two demand scenarios. The dynamic multi-level simulation dramatically reduces the computational time compared to the microscopic simulation while showing higher accuracy than the mesoscopic model.

ABSTRACT. Public transport authorities rely on information about the numbers of passengers on-board their vehicles and the resulting comfort levels in order to ensure the quality of their services. Automatic Passenger Counts (APC) can be deployed for generating this information, however, such datasets are often incomplete. This paper proposes a Kalman filtering-based method for filling gaps in APC datasets in the challenging case of complex public transport networks with multiple lines operating at each station. Historical data are also incorporated in the proposed framework. The case study is the west wing of the Helsinki commuter train network. Results indicate that the proposed method performs well in estimating the comfort level on-board vehicles, with the majority of estimations being associated with zero error when historical days are not considered. Results also include invaluable insights on the role of different levels of APC availability in the estimations’ accuracy.

ABSTRACT. Congestion and increased emissions are prominent in global metropolitan areas, largely due to the lack of restraints on car usage. The spatial constraints owing to the high density of infrastructures and population are distinctive city by city. While the industry is innovating in shifting to less carbon-intensive engines, the congestion problem has not been well solved. We propose a novel policy instrument called MobilityCoin System. Based on tradable mobility credits, every user receives a credit budget at the start of each term that can be used to pay for trips. The trip costs are dependent on the externalities caused by the travelling activities and can be either positive or negative values, where a negative price refers to an incentive for active mobility. The limited supply of mobility credits creates a market price that acts as an economic incentive to promote the use of sustainable transportation modes. The managing agency can adjust the volume of credits allocated to the system and the credit prices per trip over time to achieve a long-term objective, e.g., to reduce carbon emissions. The aim of using the MobilityCoin System to reduce greenhouse gas emissions is only attainable with a viable market. As such, the market functionality will be crucial: if the transaction value of a MobilityCoin falls to zero, its regulatory effectiveness will be suspended. As the need for mobility grows, demand for mobility credits is expected to rise. The supply of these credits, however, is guided by considerations of equity, employing methods like uniform distribution, individual allocation, or grandfathering approaches. This strategic allocation decision is pivotal in maintaining a stable market price. However, the effects of the various allocation strategies have not been applied to large-scale networks. In this paper, we investigated the impacts of this innovative approach. In particular, different credit allocations across various user groups and the impacts on mode-shift, MobilityCoin market equilibrium, and emissions in the transportation network of Munich. It represents a follow-up of the conceptional work of. As a methodological novelty, we demonstrated the market clearing condition via an optimization approach using the Brent algorithm, which in this framework is more efficient than variational inequality (VI) or mixed complementarity (MCP) formulations.

ABSTRACT. Urban traffic congestion during morning rush hour is a source of severe discomfort for commuters. On the one hand, a portion of commuters need to make an intermediate stop by the school area to drop off their children before heading to their destination. On the other hand, a different portion of commuters utilize the same network infrastructure to arrive at their destination directly. This mobility pattern arises when the schools start at the same time. Both types of commuters depart from their homes simultaneously to reach the corresponding school and work, respectively, on time. This traffic behaviour causes the formation of a massive demand that the network cannot fully accommodate, resulting in the emergence of congestion. To address this issue, we propose a novel approach that staggers the school start times such that the peak demand is redistributed. Our proposed method incorporates regional and class-specific traffic dynamics, respectively, utilizing the Macroscopic Fundamental Diagram to model the movement of vehicles. The related problem is formulated as a bi-objective mixed integer nonlinear program that jointly minimizes i) the total time spent by all vehicles inside the network and ii) the overall delay observed between the initial and the shifted start time of each school. The nonlinear nature of the resulting formulation can reduce the effectiveness of derivative-based optimization solvers. To this end, we develop a solution approach that combines the proposed staggered school schedule’s paradigm with a well-established derivative-free optimization solver.

ABSTRACT. Pooled Mobility-on-Demand (MoD) services have emerged with numerous benefits, including reducing carbon emissions and providing affordable transportation options. However, developing request-vehicle matching algorithms for pooling services poses more challenges compared to ride-hailing services, due to the additional complexities of shared rides. One significant challenge is to decide the optimal maximum detour time to allocate to a ride---a larger buffer enhances the possibility of accommodating extra passengers en route, but creates a larger inconvenience to the riders on board due to the increased travel time. In response to this challenge, our research presents a matching algorithm designed to determine request-specific slack times as well as the optimal matching pairs. Our framework is built upon the state-of-the-art centralized online matching algorithm with a sequential decision making process. A key contribution of our approach lies in its ability to measure and integrate the future value and acceptance probability of detour values/assignments. To this end, we leverage both discrete choice modeling and reinforcement learning framework, effectively balancing between the immediate acceptance rate of passengers and the future potential to serve more passengers. To validate our approach, extensive experiments were conducted within the Manhattan network using real-world data. Our matching algorithm consistently outperforms baseline models, and the trained model facilitates a revenue increase of up to 15.1%. Furthermore, we show the generalizability of our model through zero-shot transfer performance and scalability up to 200 vehicles and 2,000 customers. This research offers a promising avenue for addressing some of the challenges faced by pooled MoD, ultimately contributing to more efficient and preferable transportation solutions in the future.

ABSTRACT. This paper investigates spatial pricing and fleet management strategies for an integrated platform that provides both ride-sourcing and intracity parcel delivery services, leveraging the idle time of ride-sourcing drivers across a transportation network. Specifically, the integrated platform simultaneously offers on-demand ride-sourcing services for passengers and multiple modes of parcel delivery services for customers, including: (1) on-demand delivery, where drivers immediately pick up and deliver parcels upon receiving a delivery request; and (2) flexible delivery, where drivers pick up (or drop off) parcels only when they are idle and waiting for the next ride-sourcing order. A semi-Markov process (SMP) model is proposed to characterize the status change of drivers under the joint movement of passengers and parcels over the transportation network with limited vehicle capacity. Building on the SMP model, incentives for ride-sourcing passengers, delivery customers, drivers, and the platform are captured through an economic equilibrium model. Subsequently, we derive the platform's optimal spatial pricing by solving a profit-maximization problem. We also validate the proposed model through a case study of San Francisco. Numerical results indicate that ride-sourcing and parcel delivery services exert both complementary and competitive effects on each other, with the integrated business model's overall impact hinging on the complex interaction between ride-sourcing orders, on-demand parcel delivery orders, and flexible parcel delivery orders.

ABSTRACT. The emerging Mobility-as-a-Service (MaaS) transport model is characterized by its ability to offer integrated mobility solutions through subscription plans, commonly known as MaaS bundles. Although MaaS promises efficient and sustainable transport, its success remains uncertain, with studies showing varied levels of acceptance and conflicting results regarding its impact on private car usage. We hypothesize that these inconsistent findings are due to insufficient consideration of the dynamic interplay between bundle subscriptions and daily mode choices, influenced more by perceived than actual travel costs. In this regard, this study examines the evolution of MaaS subscribers' bundle subscriptions and daily mode choices to test our hypothesis. Specifically, the static choices are captured by the stochastic user equilibrium principle, and their dynamic evolution is described using a day-to-day (DTD) dynamical model. Preliminary insights reveal patterns in travelers' adoption of different MaaS bundles and their impact on vehicle-based travels. Further analysis will focus on MaaS operators' optimal control strategies, aiming to demonstrate how effective implementation of MaaS can enhance resource allocation and inform the design of future MaaS strategies.

ABSTRACT. The advent of shared autonomous vehicles (SAVs) offers new avenues to enhance multimodal public transit services, particularly as feeders in less dense areas. With lower operating costs, they can address the first-mile-last-mile problem and bring passengers to other transit modes more effectively than conventional ride-sharing. We investigate semi-on-demand (SoD) hybrid-route services, which combine the cost efficiency of fixed-route buses in denser areas with the flexibility of on-demand services in less populated regions, thereby enhancing convenience and attracting more passengers. In a SoD route, SAVs first serve all fixed-route stops based on a schedule, then drop off and pick up passengers in the flexible route portion (pre-determined), and return to fixed-route scheduled stops and the terminus. With the theoretical formulations of costs and benefits derived in the previous works, this study focuses on the service design and simulation of the SoD hybrid-route transit feeders using SAVs. First, we conceptualize the service and determine the schedule and fleet size analytically considering detours, service guarantee, and peak/off-peak service level. Second, we conduct agent-based simulations on ten existing bus routes in Munich, Germany, optimize the flexible route portion and headway, and examine the benefits of converting these fixed routes to hybrid routes (in terms of access, waiting, and riding times for users and vehicle distances and requirements for operators). Third, by analyzing the simulation results, we identify SoD use cases with respect to demand and service characteristics, contrast theoretical predictions, and study variations in user experience to investigate the equity impacts.

ABSTRACT. The unprecedented changes in working patterns induced by the COVID-19 pandemic have put into question the future of conventional office set-ups and, in turn, of dense urban centres. A key theme underlying this discussion is: Why is densification important? While the productivity benefits of densification remain well known, more recently, the literature has sought to understand if densification contributes to a greener future. To that end, this paper evaluates the impact of densification on the operational energy usage of two major modes of commuting, urban rail transit (metro) and road-based private vehicular travel. We develop a novel causal model, grounded in economic theory, to study the relative demand for energy in transport operations. The model for metros is calibrated using a unique panel dataset related to the annual operations of twenty-seven metro operations worldwide. Our results show that metro operations with a high density of usage are highly energy-efficient --- a 10 percent increase in the passenger kilometres travelled on a fixed network length decreases the energy usage per passenger kilometres by 3.45 percent. Conversely, our analysis of private vehicular travel on road networks, conducted using the Millennium Cities Database for Sustainable Transport, suggests no statistically significant energy savings for high-density private car travel. These results have important implications for transport planning and policies to limit the future increase in urban energy use, thereby mitigating climate change.

ABSTRACT. Regular maintenance activities are essential to ensure the resilience and prevent unexpected disruptions from the urban rail transit (URT) systems operation. However, the need for non-traffic hour (NTH) maintenance window is expected to exceed the available time window in the future, and adjustments to train service schedules by ending services earlier at night and/or starting later in the morning may be necessary. This study aims to optimize the schedules of the last few late-night train services and the first few early-morning trains while meeting the requirement for NTH maintenance window. Firstly, we present closed-form solutions for analyzing the optimal late-night train schedules, service end time, early-morning train schedules, and service start time on a single line. Secondly, we analyze service coordination between these periods to meet the NTH maintenance window requirement numerically with passenger transaction data on metro line 2 in Chengdu, China. Our findings provide innovative solutions to the pressing issue of extending the NTH maintenance window on URT systems, striking a balance among the objectives of different stakeholders, including late-night passengers, early-morning passengers, and the maintenance work. Additionally, this study provides a novel aspect in understanding the coordination between late-night and early-morning train service operation with respect to NTH maintenance window.

ABSTRACT. The rise in demand for both passenger and freight services results in congestion on the rail network, consequently leading to disruptions. In recent years, there has been a growing focus on resilience in the transport sector due to its crucial role in facilitating recovery efforts and maintaining socio-economic stability. There have been limited studies on assessing the performance of rail transit networks under the varied nature, size, and effect of disruptions. This study investigates on the resilience of rail networks, with a particular focus on analysing the impact of infrastructural changes within station areas. Given the high costs associated with infrastructure modifications, we aim to provide better metrics that aid in identifying optimal strategy for enhancing network resilience. Therefore, the study offers a valuable guidance for infrastructure managers to strengthen rail networks against future uncertainties.

ABSTRACT. High-speed railway (HSR) has experienced significant expansion and has become the primary choice for passengers traveling between inter-cities in recent years. In HSR, passenger demand varies during different time periods. Thus, the rail managers, suffering from financial pressures in low seasons, are actively exploring the potentials for using a part of vehicles in some trains to transport both passengers and fast-delivery goods, while these is few research along this line. Our study formally addresses this problem for the integrated planning of train schedule and set allocations for passengers and freight. The objective of our model is to minimize the total travel time and operation costs of trains while satisfying the demands of both passengers and freight. In addition, we carefully consider some new constraints in our formulation, involving the time window limitation of freight, the freight loading/unloading work time at platforms, which are practically significant while were not fully investigated in the existing literature. Then, we prove that the constructed model is an NP-hard problem, indicating its inherent difficulty in being solved. To tackle this challenge, We develop an decomposition-based approach to solve the model, where the passenger-freight allocation is formulated as the master problem, and the scheduling of each train is formulated as one subproblem. Based on the decomposition scheme, we further construct an L-shaped cut for iteratively solving the model. Real-world case studies on Wuhan-Guangzhou high-speed railway corridor are conducted to verify the effectiveness of our approach.

ABSTRACT. As urbanization progresses, traffic congestion increasingly undermines economic and environmental health in cities. Effective traffic signal control (TSC) strategies are essential to mitigate these adverse effects. Traditional methods, which often rely on fixed traffic variables, struggle to address the complexity of modern traffic patterns. Recent developments in model-free multi-agent deep reinforcement learning (MARL) offer dynamic solutions that adapt to changing traffic conditions, demonstrating potential for enhanced traffic management. However, MARL faces significant challenges, including a misalignment between the immediate goals of the RL agent and the broader objectives of TSC, which aim to improve overall traffic conditions by reducing network delays or total travel time. Additionally, the real-world applicability of the methods based on the short time-step is limited by high operational costs and safety concerns.

To address these challenges, we propose a novel RL-based strategy using the Cross-Entropy Method (CEM) to hierarchically optimize traffic signal variables such as offset and phase split. This approach prioritizes control targets sequentially by Level of Service (LoS), starting with offset at the upper level and phase split at the lower level. Unlike previous studies, this method integrates a global reward focused on network delay time, aligning local actions with the broader traffic management goals and effectively bridging the gap between local rewards and the ultimate TSC objectives. Our method is designed to seamlessly integrate into existing traffic signal systems and can be effectively applied across the real network by observing the existing standard policy of offset control.

ABSTRACT. Multi-Agent Reinforcement Learning (MARL) has been empirically demonstrated as a highly promising paradigm for the Cooperative Traffic Signal Control (CTSC) of urban road networks. However, some recent literature shows that many well-designed MARL schemes are less effective than independent control schemes in multi-intersection scenarios. This paper analyzes the phenomenon and proposes a hypothesis that the setting of surveillance zone length may determine whether a MARL-based CTSC algorithm is effective or not. We prove this hypothesis both qualitatively and quantitatively and find that the intersection interactions are time-lagged. Faced with the incomplete surveillance zone, we propose ENB-RL, a MARL model with explicit neighborhood backtracking to handle the lag in impacts from neighbors. The core of our proposal is an ENB module, which is consist of a neighborhood backtracking stack to store and update neighborhood intersections' historical throughput in a segmented weighted way, and a multi-head attention model for spatio-temporal differentiated input. Such explicit and precise inputs can improve the agent's observations in incomplete perceptual environments. Considering that historical backtracking information may lead to convergence instability, we introduce random Gaussian noise for Double Deep Q-Network (DDQN) to generate uncertainty and improve the efficiency and stability of exploration. Experimental results show that ENB-RL has the best convergence performance on both synthetic and real-world datasets, and outperforms other state-of-the-art MARL models. Ablation experiments confirm the efficacy of each component in the framework. Moreover, the proposed ENB module can also be plugged and played in mainstream RL-based models.

ABSTRACT. This paper proposes the use of the Unjam API for actuator selection and control in a unified data-driven framework for urban traffic management through variable speed limit (VSL) adjustment. Focusing on an urban network, our approach identifies critical streets for implementing variable speed limits to alleviate traffic congestion. Leveraging the Data-Enabled Predictive Control (DeePC) algorithm, traffic lights in these streets serve as actuators for the urban traffic system. Our study demonstrates the versatility of this approach, as the same theoretical framework efficiently handles both tasks with minimal modifications, presenting a comprehensive solution for urban traffic control challenges. Validation through extensive simulations on the SUMO traffic simulator confirms the efficacy of our scheme in selecting key streets for control, leading to overall travel time reduction.

ABSTRACT. We introduce a fully adaptive multimodal traffic signal controller with safety constraint to resolve priority request conflicts and ensure traffic safety. The multimodal signal controller is built upon the concept of constrained Markov decision process in the connected and human-driven vehicle environment, termed c3DQN. The algorithm consists of a component that utilises Dueling Double Deep Q-network (3DQN) for initial policy learning, and a hazardous costs estimator that uses Double Deep Q-network to assess the unsafe costs generated by emergency braking related to the yellow signal dilemma zone, which is caused by random phase-switching actions. When selecting actions, the DRL agent generates two sets of values through these two components. The agent chooses the best action by a weighted sum of the two set of values using a gradient ascent mechanism to maximise rewards while minimising critical hazardous costs. Its performance is evaluated by simulations in SUMO under three traffic demands, fixed schedules and random passenger occupancy of public transportation vehicles. Results demonstrate that c3DQN substantially reduces average waiting queue length and passenger delays compared to an optimised multimodal NEMA controller via mixed integer linear program, and other model-free DRL-based controllers. The integration of the hazardous costs estimator proves effective in mitigating such unsafe costs during the learning process.

ABSTRACT. In recent years, both individual and public transport sectors have increasingly electrified globally, aligning with the European Union's objectives set in October 2022 to cease sales of new combustion- powered vehicles by 2035. However, this massive electrification increases the electric power needs of urban infrastructures, intensifying and complexifying interactions between electricity and transport networks. Consequently, disruptions in one network can significantly impact the other. This emphasizes the need for a comprehensive study of these critical infrastructures, considered jointly rather than independently, which is an operational challenge due to the lack of interactions among system operators.

Resilience and robustness are widely discussed concepts in the scientific literature when it comes to assessing the performance of public transportation networks. Resilience refers to the network's ability to absorb and adapt to unexpected disruptions, as well as its capacity to recover quickly. This notion requires dynamic approaches and simulations to capture the network's response to disruptions. For example, in (Goldbeck et al., 2019), the authors study the resilience of London’s interdependent metro and electricity networks, by dynamically redistributing passenger flows and deploying repair resources after a power breakdown. On the other hand, robustness is usually defined as the network's ability to maintain functionality in the face of failures. It is a narrower concept compared to resilience and it can be studied through static approaches. For instance, analyzing the topology of transport networks can provide valuable insights into their robustness, by measuring their resistance to disruptions or highlighting vulnerable elements.

However, transportation robustness studies rarely account for electricity networks in their model. Recent topological studies often focus on isolated properties of single-mode transportation networks and neglect their interconnectedness with power grids. For example, the robustness of rail networks in Paris and China was calculated in (Adjetey-Bahun et al., 2016) and (Fang et al., 2020) by using betweenness centrality, which is a metric that quantifies the proportion of the graph shortest paths between all pairs of nodes that run through a given node or link. (Cats, 2016) and (He et al., 2021) use metrics based on travel time delay caused by transit network disruptions to assess the robustness of the Stockholm’s metro, commuter and light rail trains and the Netherlands’ inland waterway, road and railway freight transport. The robustness of interdependent transport and power networks is thus a recent and complex area of study. The objective of this study is precisely to address this gap, by applying topological metrics to evaluate the robustness of a multimodal public transport network under various electric power breakdown scenarios. The methodology was applied to Lyon, France.

ABSTRACT. In this study, we propose an augmented ε-constraint-based optimization framework for multi-objective multi-modal traffic management. This framework is bi-level and can accommodate various traffic models and objectives that reflect the diverse interests of multiple stakeholders. Thus the multi-modal traffic management problem can be formulated as a multi-objective nonlinear optimization problem. The augmented ε-constraint method is employed to efficiently address the multiple objectives, and the multi-start sequential quadratic programming method is used to solve the nonlinear optimization problems, such that the Pareto front is obtained. We validate the effectiveness of our framework through a case study, whose preliminary results show that our method improves the traffic performance and provides insights into the trade-off among different objectives.

ABSTRACT. This presentation will describe technologies from data-driven Bayesian networks, Benders’ branch-and-cut and emissions estimation technologies that together are used to predict the environmental impacts of changing global maritime freight flows from thawing Arctic sea ice.

ABSTRACT. In this paper, we present a study where the net benefits of sensor-based recyclables’ collection planning are analyzed. The two-phase methodology we have develop for this purpose is explained and exemplified through an application to a major municipal solid waste management company operating in Portugal.

ABSTRACT. Walking accessibility is a fundamental component of urban design, which plays a significant role in creating livable, sustainable and inclusive cities. This study aims to investigate potential disparities in walking accessibility between new and old development areas. To achieve this, a utility-based walking accessibility measure is developed, incorporating the impacts of street-level walking attributes, the spatial distribution of points of interest (POIs) and pedestrians’ heterogeneous behavioral preferences. Walking accessibility disparity or inequity among different age groups is further quantified using Gini coefficients. The comparative case study focuses on Kwun Tong, an old urban area, and Kai Tak, a new development area in Hong Kong. The results show that the new development area exhibits lower walking accessibility and a higher level of inequity compared to the old development area. This disparity can be attributed to the concentration of POIs within a few business clusters and insufficient pedestrian facilities. Additionally, the elderly have the worst walking accessibility among the three age groups. The findings highlight the necessity to incorporate pedestrians’ diverse preferences in planning new development areas to create a pedestrian-friendly environment for all.

ABSTRACT. Curb spaces has been increasingly critical in recent years due to the widespread e-commerce with on-demand delivery and expansion of ride-sourcing services. The intensive usage of the limited public resources by multi-modal travelers without adequate regulation leads to competition between various users and illegal parking behaviors. This situation is exacerbating the congestion not only on curbs but also the broader transportation system. Some existing research investigates the curb usage pattern and develops targeted pricing and space allocation strategies but the research gap lies on designing optimal control policy for curb spaces in dynamic network models which integrates both multi-modal travelers' behavior patterns and spatio-temporal network flow dynamics. To this end, this study proposes a framework to design optimal curbside pricing and space allocation policies for managing multi-modal curb users in general dynamic networks. Our framework combines curb-aware flow dynamic of multi-class traffic and travelers' route/curb choice behavior in mesoscopic modeling with the optimization of curbside operations.

ABSTRACT. Shared autonomous vehicle (SAV) systems could be a promising transportation mode in the near future. Optimization schemes for SAV systems such as congestion pricing have been extensively studied in the literature by using complicated methods such as reinforcement learning. However, to the authors' knowledge, mathematically tractable analysis and general insights based on it on this problem is still limited. In this study, we mathematically analyze dynamic operation of an SAV system and derive several properties on the optimal pricing for it. Specifically, we develop a model of SAV systems where the behaviors of travelers and SAV operators follow the dynamic user equilibrium principle in a congestible many-to-many network. By analyzing the model, we mathematically derive dynamic system optimal pricing for the SAV system. Then, we prove several theorem on the optimal pricing and SAV system. Some of the findings are as follows. In the optimal state, travelers may pay congestion toll to SAV operators, and SAV operators may pay another congestion toll to the road authority. The optimal SAV fleet size is automatically maintained by toll from passengers.

ABSTRACT. See the attached document.

ABSTRACT. Please see attached.

ABSTRACT. Modern-day transportation systems involve many stakeholders who influence the overall impacts of each other’s decisions. Therefore, it is imperative to model all pertinent stakeholders to reasonably predict scenarios and facilitate informed decision-making among them. In this abstract, we present a Game-theoretical model addressing these motivations. It can be applied to pricing decisions of public and private players/stakeholders in multi-modal transportation systems with different types of travelers. The travelers can be differentiated based on their sociodemographic characteristics and mobility resource ownership with each group having a different Willingness To Pay and Value of Time. Additionally, the public and private players may implement link or path-based instruments targeting each traveler group separately. In addition to finding the optimal pricing decision of players when they are optimistic, we also do so in the case when they are pessimistic. In the case studies, we apply this model to a pseudo real city inspired from Leuven and find equilibrium pricing decisions of public and private players involved in multiplayer games.

ABSTRACT. Perimeter control, based on aggregate dynamics modeling using network Macroscopic Fundamental Diagrams (MFDs), has been shown effective in congestion mitigation and throughput maximization for urban networks comprised of a single or multiple homogeneous regions. However, in dense urban areas, local pockets of congestion might form, resulting in traffic heterogeneity, which diminishes the effectiveness of perimeter control. To this end, an integrated framework that regulates both the inter-regional exchange flows (viz., perimeter control) and intra-regional traffic signals is proposed, wherein the upper-level perimeter control helps maintain regional accumulations around the critical levels while the lower-level signal control combats local congestion to improve traffic homogeneity. Both levels are controlled by reinforcement learning agents, and m multi-timescale multi-agent training approach is presented, with its effectiveness evaluated in simulated single-region networks.

ABSTRACT. In this paper, we propose a closed-loop learning framework for robust adaptive perimeter traffic control of traffic networks. Different from previous algorithms that learn traffic dynamics solely from open-loop data, ignoring the effect of controller deployment, we choose the closed-loop data generated from a microscopic simulation platform after the control deployed to learn the traffic dynamics. The iterative operation of controller design and traffic dynamics learning based on closed-loop data facilitated obtaining the optimal control strategy. Through our preliminary experiments, we verify the feasibility of the proposed method and the necessity of injecting the underlying control objective into the perimeter control framework.

ABSTRACT. As urban centers become more densely populated, the rise in vehicular traffic has overwhelmed existing transportation infrastructure. To mitigate the adverse effects of traffic congestion, various traffic control methods have been developed and implemented. Perimeter control is a promising direction which aims to regulate vehicle inflows at the boundaries of protected areas to prevent congestion within. Most studies in literature have developed MFD-enabled perimeter control strategies showcasing excellent performance in single and multi-region systems. More recently, the success of AI control methods has led the researchers to utilize deep reinforcement learning (DRL) methods. In general, there have been limited studies introducing RL-based perimeter control methods that are trained through direct interaction with a microsimulation environment, which is considered the most accurate representation of real-world traffic dynamics. In this work, we formulate the multi-region perimeter control task as a model-free DRL problem trained directly on a microsimulation environment and a 2-stage method is proposed that combines DRL with classic optimization to account for the queue formation on the boundaries of the multi-region system.

ABSTRACT. In disaster situations, unique trip patterns distinct from regular traffic emerge, necessitating network-wide traffic control to avoid heavy congestion. Previous research lacks a realistic, rapid prediction of dynamic traffic flow during a disaster, and insufficiently considers a gradual transition sequence from traffic control in ordinary times. In this study, we developed a dynamic evacuation behavior model using the Recursive Logit model to predict evacuation departure and destination choice. Additionally, we integrated a zone-based traffic assignment model employing MFD to predict and evaluate dynamic traffic flow during disasters swiftly. Using this model, we determined the configuration of inter-zone control links to minimize congestion during evacuations. To facilitate a smooth transition from regular traffic control to optimized evacuation control, we introduced a combinatorial reconfiguration problem. We successfully devised a rapid computational solution using Zero-suppressed binary Decision Diagrams for an exhaustive search.

ABSTRACT. Traffic speed estimation is a critical research area for effective traffic management and operations. Despite numerous model-based and data-driven methodologies proposed over the past decades, challenges such as data unavailability, map-matching complexity, and limited network scale persist in link-based methods. To address these challenges, this study introduces an area-based mean speed estimation system for large-scale urban networks, leveraging the widespread availability of loop detector data and training on sparse probe data. By addressing the common challenges in link-based traffic speed estimation, our approach utilizes a spatiotemporal data encoder combining graph convolutional neural network (GCN) and gated recurrent units (GRU) models to capture spatiotemporal interdependencies. A random forest regression model is subsequently utilized for speed estimation. Case studies conducted with loop detector data and sparse probe data from Munich, Germany, evaluated the performance of speed estimation using various input features. Additional experiments were performed in the city center and across different area unit sizes. The results demonstrate the potential of our area-based approach for reliable and accurate urban traffic speed estimation.

ABSTRACT. An extended abstract is attached.

ABSTRACT. United States (U.S) transportation policy is shifting from a 20th-century automobile-oriented market towards a 21st-century multimodal transportation-oriented market - choices including walking, bicycling, and public transportation. This focus is particularly amplified by emerging mobility options and technologies (e.g., ride-hail services and alternative fuel vehicles) leading to various decarbonization pathways that encourage efficient energy mobility systems (CARB, 2021). As a result, transportation planning agencies embark on developing advanced transportation models, wherein aspects of budget, time and other constraints act as barriers for agile decision making and policy analysis. We propose to fill this gap by developing an agile sketch-planning neighborhood-level work commute mode share model that leverages nationally available data sources from the U.S. Census Bureau and Environmental Protection Agency (EPA) and OpenStreet Maps. In this study, we make use of the flexible fractional split multinomial logit (FMNL) model formulation as proposed by Sivakumar and Bhat (2002) that predicts shares of commute modes (or proportion of commuters by mode) at the neighborhood-level (at block-group) for the entire U.S. Also, FMNL model application in transportation planning and modeling literature has been scarce with a few exceptions as applied to activity time-use patterns and aggregate land-use planning (Ye and Pendyala 2005; Schnieder et al., 2018). Although this structure provides a great advantage in predicting homogenous block-group scale commute mode shares. It is evident from past research that commute shares are both spatially correlated and have a heterogeneous distribution due to first and second order effects that impact travel behavior of the commuters. However, there is very little work-to-date that addresses these issues within FMNL models in a holistic framework. As a result, we propose to leverage bagging-based methods, i.e., utilizing an ensemble of FMNL models aimed at reducing model instability, capturing spatial heterogeneity and improving model goodness-of-fit and performance. This work contributes to the field in two ways - First, we demonstrate that bagging-based fractional split multinomial logit model (BAG-FMNL) combines desirable properties of FMNL of being stable and theoretically sound, along with bagging, which is noise resistant and applicable to large feature spaces in model building. Second, we describe a feature selection algorithm for BAG-FMNL based on statistical properties of FMNL that ensures computational time is not wasted on statistically insignificant features. In order to demonstrate the effectiveness of the proposed algorithm, we make use of the 2019 U.S. Census Journey-to-Work (J2W) flow data as compiled for the New York Metropolitan Transportation Commission (NYMTC) region - consisting of roughly 20 million in population. For this exercise, we specifically build on open-source data to make the sketch-level planning tool and BAG-FMNL algorithm accessible to planners and policymakers. For this purpose, we define block-group level commute mode shares (e.g., drive alone, carpool, bus, subway, telecommute, walk, bike and other) as the outcome variable, wherein their proportion sums to unity. We also conduct a comprehensive analysis based on literature review to perform feature engineering that represents various influencing aspects of commute behavior such as sociodemographic characteristics and built-environment characteristics (e.g., accessibility, density and diversity). For instance, Figure-1 shows a sample box plot distribution of differences between observed and predicted (on Y-axis ranging from -1 to +1) commute shares by mode (as on X-axis) at the block-group level in the NYMTC study area. Overall, the distribution showcases that a preliminary specification of BAG-FMNL with an ensemble of 10 training models matches the observed mode shares value very closely with significantly lower variation for modes related to bike, walk, telecommute (work from home) and public transportation (bus and subway). Similarly, Table-1 documents the performance results of the BAG-FMNL model with observed data along the prediction accuracy and mean absolute error measurements. The results (although still preliminary) demonstrate the model stability and predictive nature, wherein model stability and goodness of fit is robust. As an application, we also propose to demonstrate in this study the policy-case of COVID-19 restrictions, wherein aspects of shelter-in-place and operational constraints on businesses were imposed early in the pandemic. Finally, we propose to demonstrate the capability and efficiency of the BAG-FMNL model with other model specifications and we plan to compare this model against a generalized FMNL model and a series of ensemble-based machine learning models (e.g., gradient boosting and random forests). For this purpose, we will document and report the model performance results against standardized benchmarking (e.g., RMSE and MAPE).

ABSTRACT. In ride-hailing systems, drivers decide whether to accept or reject ride requests based on factors such as order characteristics, traffic conditions, and personal preferences. Accurately predicting these decisions is essential for improving the efficiency and reliability of these systems. Traditional models, such as the Random Utility Maximization (RUM) approach, typically predict drivers' decisions by assuming linear correlations among attributes. However, these models often fall short because they fail to account for non-linear interactions between attributes and do not cater to the unique, personalized preferences of individual drivers. In this paper, we develop a method for learning personalized utility functions using hypernetwork and ensemble learning. Hypernetworks dynamically generate weights for a linear utility function based on trip request data and driver profiles, capturing the non-linear relationships. An ensemble of hypernetworks trained on different data segments further improve model adaptability and generalization by introducing controlled randomness, thereby reducing over-fitting.

We validate the performance of our ensemble hypernetworks model in terms of prediction accuracy and uncertainty estimation in a real-world dataset. The results demonstrate that our approach not only accurately predicts each driver’s utility but also effectively balances the needs for explainability and uncertainty quantification. Additionally, our model serves as a powerful tool for revealing the personalized preferences of different drivers, clearly illustrating which attributes most significantly impact their rider acceptance decisions.

ABSTRACT. In this paper, we propose a new perturbed utility model with a semi-nonparametric form to approximate the unknown perturbation function. Specifically, we consider that the semi-nonparametric perturbation function is represented by a sequence of sigmoidal basis function integrals. We show that the proposed perturbed utility model with semi-nonparametric perturbation function is identifiable, namely, the parameters in both the systematic utility function and the semi-nonparametric perturbation function can be uniquely determined under identification conditions. We also develop an efficient estimation approach using the sieve least square estimator. Our Monto Carlo simulation experiment confirms the model identifiability and the asymptotic property of the estimator, by showing that the preset parameters can be recovered with finite samples, and the RMSE decreases with sample size at the rate of root-N.

ABSTRACT. This study proposes a Bayesian Network activity model that can flexibly build model structures in a data-driven way. Then, we validate the model using real data from the Tokyo metropolitan area. Using a dynamic object-oriented BN, the model effectively estimated the differences in dependencies among variables caused by the sequence of activities. It successfully generated plausible activity chains by incorporating resampling that accounts for time constraints.

ABSTRACT. Road transportation systems are usually designed for optimizing transportation efficiency. A pure focus on efficiency overlooks, that passengers are not equal in their urgency; delays cause them different harm. Vehicle prioritization is a promising countermeasure to this equity issue. Existing strategies for prioritization can be grouped into three categories: dedicated lanes (e.g. public transport, legislative prioritization of blue light vehicles (e.g. ambulance), and economic instruments (e.g. high occupancy toll lanes). There is no dedicated instrument that allows for prioritization at intersections in practice yet, even though intersections are a major source of delays in urban contexts. In this work, we analyze, to which extent it is possible to expedite a certain share of vehicles at intersections, without causing arbitrary delays for the other vehicles or affecting transportation efficiency de trop. We propose the Priority Pass, a needs-based signalized intersection management. Entitled vehicles shall be prioritized at intersections, resulting in shorter delays. The Priority Pass is an intelligent transportation system, that builds upon auction-controllers, and existing, urban, vehicle-identifying infrastructure. Experiments with varying number of prioritized vehicles, total traffic flows, and symmetry of demand, are conducted. The results demonstrate, that this concept generates significant benefits to the drivers, which do not come at the cost of transportation efficiency or arbitrary delays.

ABSTRACT. This study introduces a novel evaluation index for traffic signal control accommodating to various traffic conditions, and presents two distributed adaptive control methods. Leveraging shockwave theory, traffic dynamics at intersections are captured by integral in the index referred to as '\textit{synthetic delay}', which automatically evaluates both delay and throughput with flexible significance. Taking the synthetic delay as optimization objective, the first control method selects optimal phases in real-time, while the second method coordinates green time and advisory traffic speed based on fixed phase sequence. Numerical experiments conducted in a grid network demonstrate that the proposed methods significantly reduce average delay and increase network throughput under different conditions, thereby enhancing traffic efficiency.

ABSTRACT. Efficient traffic signal control is essential for managing urban transportation, minimizing congestion, and improving safety and sustainability. Reinforcement Learning (RL) has emerged as a promising approach to enhancing adaptive traffic signal control (ATSC) systems, allowing controllers to learn optimal policies through interaction with the environment. However, challenges arise due to partial observability (PO) in traffic networks, where agents have limited visibility, hindering effectiveness. This paper presents the integration of Transformer-based controllers into ATSC systems to address PO effectively. We propose strategies to enhance training efficiency and effectiveness, demonstrating improved coordination capabilities in real-world scenarios. The results showcase the Transformer-based model's ability to capture significant information from historical observations, leading to better control policies and improved traffic flow. This study highlights the potential of leveraging the advanced Transformer architecture to enhance urban transportation management.

ABSTRACT. The rapid increase of car ownership in recent decades has led to the increase of air pollutants in road networks, degrading the air quality and public health. Traffic signals are components of the urban networks able to be modified to promote efficiency and sustainability. This study develops a real-time signal control system for a single intersection that is part of a signalized arterial, i.e., vehicles arrive in platoons. This system optimizes signal control timings by minimizing the emissions of auto and transit vehicles. The model can handle both oversaturated and undersaturated flow conditions, taking into account delays experienced from residual queues. The final mathematical model is a mixed integer linear programming model that is characterized by low computation times, sufficient to allow for real-time optimization and applicability in the real-world.

ABSTRACT. Urban traffic signal timing separates incompatible traffic flows in time, but normally cannot intervene in the utilization of road resources when the bi-directional (e.g. east and west bound) traffic demands are much unbalanced, because such demands receive the same right of way. A novel traffic control measure called internal boundary control (IBC) was recently proposed to address a similar concern on freeways for lane-free traffic of fully connected automated vehicles (CAVs). This paper applies the idea of IBC to maximize the utilization of intersection resources for improved urban traffic efficiency. The paper focuses on the joint optimization of roads’ internal boundaries and signal timing for an isolated intersection in the paradigm of lane-free traffic of CAVs. The optimization problem is formulated as a binary-mixed-integer-quadratic-programing problem. The results show that the joint optimization has a potential of significantly improving traffic situations at an intersection, typically those intractable via signal timing optimization alone.

ABSTRACT. On-demand meal delivery has become a prevalent feature of last-mile delivery in urban landscapes, spurred by digital platforms and the temporary restaurant closures during the pandemic. This study extends the utility of a Markov model, which has been employed to model the circulating behaviour of couriers, to the strategic location of ghost kitchens. Utilising a Markov model, we examine the strategic location effects of placing kitchens. The model includes parameters for demand at each kitchen and customer location, highlighting an urgency parameter β. The calibrated model, demonstrated using a publicly available dataset from Grubhub representing typical meal orders, facilitates the evaluation of potential shifts in demand and delivery performance consequent to relocating a kitchen. The paper will explore the use of the Weiszfeld Algorithm, and make modifications to it for optimally locating ghost kitchens. In this way, we will tailor the Weiszfeld Algorithm to more closely correspond to the Markov model

ABSTRACT. Carsharing systems are considered as a convenient transportation mode by providing short-term access to private vehicles. One-way carsharing systems, a type of carsharing system, offer more flexible journeys by allowing users to pick up and leave the vehicles at stations that are not necessarily the same. As the demand for different stations varies throughout the day, one-way carsharing systems suffer from imbalances between vehicle supply and trip demands. These systems often integrate vehicle relocations between stations to decrease vehicle accumulations or shortages. Relocation of the vehicles requires personnel involvement which comes with a major operational cost. The need for relocation can be reduced by promoting alternative trips which can decrease the mismatches that occur in the vehicle stock imbalances.

In this study, we present a reservation-decision framework that determines the trip offer that is provided to the user upon receiving the trip request. The framework decides on whether to serve the user with their original trip request or to offer them an alternative route by considering the acceptance and rejection rates. The framework consists of a simulator and a mixed integer linear programming (MILP) model to decide on the trip offer that maximizes the expected profit, considering the acceptance probabilities of the offers. Due to the intractability of the MILP model, we propose two types of heuristic algorithms to reduce the number of variables created. The first heuristic algorithm reduces the number of relocation variables by considering the need for vehicles or parking spots. The second heuristic algorithm creates variables related to relocation activities using a graph spanner based network. We have tested the framework using the MILP model and both of the heuristic algorithms on real-life one-way carsharing system data. The results illustrate that the heuristic algorithms performing well compared to the MILP model. Our analysis suggests that the proposed framework generates a considerable increase in the profit of the system.

ABSTRACT. We constructed a simulation-based capacity allocation optimization (SOCA) problem to investigate if the existing network structure of the transportation system can be better configured to make it more resilient to traffic jams during daily operations, unexpected extreme demand, and supply disruptions. We presented a second-order mixed simultaneous perturbation stochastic approximation (2MSPSA) algorithm to solve this kind of mixed network design problem (MNDP). The preliminary results showed that the algorithm can effectively address the SOCA problem, and transportation networks indeed can be more resilient by improving the network capacity allocation.

ABSTRACT. The present work aims at analyzing and comparing different cooperative schemes for last-mile deliveries in an urban area with an historical center. For each scheme a mathematical programming model is proposed in order to find the optimal delivery solutions for a set of couriers which have to serve both the external area of the city and the historical center in which only Electric Vehicles (EVs) can enter.

ABSTRACT. In recent decades, the ridership of urban buses in many cities has seen a significant decline due to low levels of service and competition from emerging travel modes. To upgrade mass transit systems, it is necessary to improve operational flexibility to cater to time-varying travel demand. This study focuses on real-time dispatching of a flexible mobility-on-demand bus (FMDB) system, where multi-type buses operate along fixed bus lines and stops. Flexible dispatching strategies, such as stop-skipping, speed adjustment, and bus holding, can be implemented to enhance service quality by assigning collected requests to specific bus trips. In our previous work, passengers who book trips in advance are responded in batches and bus dispatching schemes are optimized in a rolling horizon framework. This study aims to further address real-time travel requests by investigating dynamic dispatching methods and extending the operational scenario to a network level. Matching multiple real-time requests to bus trips of different bus lines is vital for fully utilizing bus capacities, especially in overlapping bus lines. In this study, the FMDB network is modeled as a heterogeneous graph to capture the complex spatial correlations among bus lines, stops, and requests. A Deep-Q-Network (DQN) algorithm is employed to solve the real-time ride-matching and vehicle dispatching problem dynamically. Attention-based Graph Neural Networks (GNN) are embedded for feature extraction and value function approximation. Numerical studies validate the performance of the proposed dispatching algorithm compared to a well-known on-demand ride-sharing (ODRS) system.

ABSTRACT. In many metropolises, parking plays a central role in mode choice. Thus, transport authorities have increasingly used parking restrictions as a lever to enforce changes in mobility. At the same time, the cruising traffic in search of parking contributes to congestion and pollution in city centres. Recently, we introduced a generic modelling framework, valid for any street network and for a wide range of drivers' behaviours, which clarifies the determinants of parking search and quantitatively captures their effect on the search time. It is solved theoretically by leveraging the powerful machinery of graph theory and statistical physics. In our presentation, we will illustrate how these theoretical insights can shed light on two distinct parking-related policies enacted for example in two cities in the South of France, Lyon and Montpellier: (i) restricting the parking supply and (ii) providing a smart-parking solution to guide users towards probably-vacant parking spaces near their destination. In particular, we will outline how the recent theory can be exploited to find optimal routes in a smart-parking solution and to gauge the potential efficiency of such solutions depending on the context.

ABSTRACT. What would be the time and emissions savings if navigation apps routed drivers in need of parking to the best available location(s), rather than straight to their destination? This work adopts a value-of-information approach to investigate this question. Ultimately, this will feature formal analyses at three levels of parking availability information: (i) no information, (ii) distributional information (e.g., the driver knows there are 20% odds of finding a spot in lot i), and (iii) true availability information (e.g., lot i has 2 spaces available). To this end, this extended abstract makes two contributions: (1) a dynamic programming framework for characterizing the problem and (2) a closed-form analysis for the distributional information setting, delineating when it is optimal to wait at a specific parking lot as opposed to when it may be better to visit other lots, as well as identifying the expected cost in each case.

ABSTRACT. Parking reservation, with the booming of smart parking technologies, has been applied in cities and is becoming more and more popular. It can reduce parking cruising time, alleviate illegal parking, and improve parking service. To address the issues caused by drivers’ unpunctuality, this paper considers the reservation matching and reallocation problem of reserved parking and proposes a two-phase framework. A 0-1 quadratic programming (0-1 QP) model is proposed to maximize the net profit of parking system to determine which requests to be accepted during the reservation phase. During the service phase, a real-time rolling horizon framework is designed to reallocate the requests with confirmed reservations, which aims to minimize the service failure rate. The Phase-type (PH) distribution is used to fit characteristics of drivers’ arrival and parking time distribution, which is based on real short-term parking lot data form Beijing and further applied in the numerical experiment. Numerical experiments demonstrate that the proposed framework significantly reduces the service failure rate and improves parking utilization rate compared with that at the basic model. This framework can be applied to short-term parking reservation systems facing supply-demand imbalances, which can better utilize scarce parking resources.

ABSTRACT. This extended abstract proposes a dynamic macroscopic model for two-region mixed autonomy networks incorporating park-and-ride (P&R), with a focus on morning peak-hour commuting scenarios. The model serves as a quantitative tool for evaluating and optimizing urban P&R policies. Key contributions include a novel two-region mixed autonomy network model featuring an adaptive macroscopic fundamental diagram (MFD) and an extended multi-pool representation that considers P&R behavior, the cruising-for-parking phenomenon, and self-parking CAVs. An experiment conducted on a two-region mixed autonomy network in the Melbourne metropolitan area demonstrates the efficacy of the model. Under optimal pricing, congestion in the city center is alleviated, leading to reduced accumulation and improved trip completion rates. The model's versatility facilitates informed decision-making on urban parking policies, thereby enhancing mobility management in mixed autonomy networks.

ABSTRACT. The development of autonomous driving systems (ADS) has primarily focused on technical advancements to prevent accidents and enhance overall safety performance. While significant strides have been made in improving the safety of autonomous vehicles (AVs), there remains a substantial disconnect between the intelligence of AVs and user acceptance, especially under extreme conditions. This study aims to illuminate the differences in decision-making between the ADS and passengers in AVs during severe crash scenarios, identifying the specific factors that prompt user interventions. We have recreated three typical road accidents from Australian roads, as documented by Australian Car Crash Media, within a high-fidelity virtual reality (VR) environment. In these simulations, vehicles involved in the original crashes are replaced with finely-tuned AVs to evaluate whether the incidents would occur in a similar manner. We engaged 60 participants from diverse demographic backgrounds in a human-in-the-loop analysis, immersing them in these scenarios via the simulated AVs to gather insights into their perceptions and reactions. By exploring the factors that influence riders' intervention behaviors and underscoring the decision-making disparities between ADS and human drivers, this study sets the stage for further research aimed at improving the safety, efficiency, and public acceptance of AVs.

ABSTRACT. Accurately and safely predicting the trajectories of surrounding vehicles is essential for fully realizing autonomous driving (AD). This paper presents the Human-Like Trajectory Prediction model (HLTP++), which emulates human cognitive processes to improve trajectory prediction in AD. HLTP++ incorporates a novel teacher-student knowledge distillation framework. The ``teacher'' model equipped with an adaptive visual sector, mimics the dynamic allocation of attention human drivers exhibit based on factors like spatial orientation, proximity, and driving speed. On the other hand, the ``student'' model focuses on real-time interaction and human decision-making, drawing parallels to the human memory storage mechanism. Furthermore, we improve the model's efficiency by introducing a new Fourier Adaptive Spike Neural Network (FA-SNN), allowing for faster and more precise predictions with fewer parameters. Evaluated using the NGSIM, HighD, and MoCAD benchmarks, HLTP++ demonstrates superior performance compared to existing models, particularly in challenging environments with incomplete input data. The proposed dual-model approach meticulously replicates the human visual and cognitive process, providing essential perceptual cues for accurate prediction. This marks a significant stride in the journey towards fully AD systems.

ABSTRACT. Roundabouts play a substantial role in urban traffic, facilitating enhanced efficiency at light traffic while potentially becoming a bottleneck during peak periods. Hence, the proficient management of roundabouts can contribute to the enhancement of traffic flow in the surrounding area; however, they present challenges because of their geometrical complexity and frequent major conflicts among vehicles. In this paper, we employ a new nonlinear controller to guide vehicles moving on large lane-free roundabouts. The feedback law was designed for curvy boundaries that facilitates incorporating roundabout and Origin-Destination (OD) corridors; however, it has some limitations that do not allow its direct implementation for the case of roundabouts and necessitates some modifications which are discussed in this paper.

ABSTRACT. The existing literature on takeover impact of autonomous vehicles (AV) is primarily focused on the impact of takeover maneuvers on the subject vehicle itself, particularly on aspects such as increased time-headways. A crucial research gap exists in understanding how human driver takeover maneuvers in AVs impact subsequent traffic flow dynamics, arising from the complex experimental design, which requires considering multiple factors, including varied traffic scenarios, human psychology, and the unknown underlying control mechanisms in commercial AVs, integrating mentally engaging tasks before takeover requests raises safety concerns in field experiments. To bridge the gap, we conducted a comprehensive, reproducible driving simulator experiment, placing human factors and AVs at the forefront. This enabled us to investigate critical questions regarding the dynamics between driver takeover maneuvers and the subsequent AVs' traffic flow dynamics. In this study, we quantitatively evaluate the impact of driver takeover maneuvers, following engagement in non-driving tasks, on the traffic flow dynamics of subsequent AVs, using a comprehensive driving simulator experiment.

ABSTRACT. Vehicle repositioning in the ride-hailing market addresses the significant spatiotemporal imbalance between supply and demand. Previous studies highlight that a large portion of orders go unserved and drivers spend extensive periods without passengers. This has spurred research into developing effective repositioning algorithms. Traditional grid-based methods, which direct drivers from one grid to another using the shortest route, often fail to optimize for critical metrics such as driver utilization or platform profit, focusing instead on minimizing travel time or distance. In this context, repositioning should aim primarily at securing the next passenger efficiently. Although Monte Carlo Tree Search (MCTS) has been effective in various applications, it struggles with temporal adaptability and demands extensive training time. To address these challenges, we introduce a novel Top-to-Bottom Reposition Method (T2B-RM) integrating Reinforcement Learning (RL) and MCTS. The first stage leverages RL to guide the regional movement of empty vehicles, deciding whether to stay or move to a new area. The second stage applies MCTS to determine the optimal routes within the targeted area, maximizing potential passenger pickups. This dual-layer approach ensures that vehicles are not only directed efficiently but also positioned optimally within target zones to enhance service availability. Our extensive experiments in Manhattan verify the efficacy of our method, demonstrating significant improvements in key performance metrics like platform total revenue (+2.4%) and order matching rate (+2.9%). These results substantiate our approach's potential to enhance operational efficiencies in the ride-hailing industry.

ABSTRACT. A ride-sharing system allows drivers and riders with overlapping trips to travel together and share costs. In addition to economic incentives for both participants, ride-sharing holds great potential to alleviate traffic congestion and reduce emissions. At the core of a ride-sharing system is the matching problem. Different from previous studies that propose matching mechanisms to maximize total trip utilities (Tafreshian et al., 2020), this study takes the perspective of the traffic manager who aims to minimize system-wise traffic congestion. Hence, it is also different from existing traffic equilibrium models that generalize ride-sharing behaviors (Di et al., 2017, Li et al., 2020). The proposed matching scheme is largely inspired by several recent studies (Zhang & Nie, 2018, Chen et al., 2020, Farahani et al., 2021), which show that the traffic network can deviate from user equilibrium (UE) and closely approach system optimum (SO) by rerouting a small fraction of travelers. Instead of assuming fully controllable autonomous vehicles (Zhang & Nie, 2018, Chen et al., 2020) or arbitrarily introducing intermediate checkpoints to trips (Farahani et al., 2021), we consider travelers detoured from their UE paths as ride-sharing drivers. Accordingly, the detours hold particular meanings of picking up riders and trips after dropping off riders.

In this study, we consider three groups of travelers: ride-sharing riders, ride-sharing drivers, and solo drivers. The problem is formulated as a Stackelberg game (Stackelberg, 1952), or mathematically, a mathematical program with equilibrium constraints (MPEC) (Dempe, 2003). Specifically, the traffic manager is considered the leader, who performs ride-matching to minimize total traffic congestion while producing reasonable matching outcomes. The leader’s decision essentially manipulates the demand pattern of travelers, or the follower, whose routing behaviors are described by a static traffic equilibrium. The key advantage of this formulation over existing ride-sharing traffic equilibrium models (Di et al., 2017, Li et al., 2020) is that it decomposes the matching and routing problems through the bi-level framework and thus largely reduces the modeling complexity. Furthermore, we show that the upper-level ride-matching problem can be reformulated as an assignment problem over a hyper-network with link costs specified using the equilibrium sensitivities derived from the lower-level given current solution. Therefore, the subproblems at both levels can be solved using classic algorithms for static traffic assignment.

This study contributes to the literature from several aspects. First, we propose a novel traffic management strategy via ride-sharing and explore its potential to push the system from UE toward SO. Secondly, we present a simpler formulation for the ride-sharing traffic equilibrium, which can be easily extended to consider different incentives for riders and drivers. Last but not least, we develop a solution approach that decomposes and reformulate the original MPEC into two network assignment problems that can be solved efficiently with existing solution algorithms.

ABSTRACT. The emergence of technology-driven ride-hailing services like Uber and Lyft has transformed urban mobility, yet they bring along negative impacts such as increased vehicle miles, congestion, and reduced transit usage. Conversely, mass transit offers efficient passenger movement with high-capacity vehicles but faces challenges in low-density areas. This motivates the development of a novel transportation system that integrates ride-sharing with mass transit to leverage the strengths of both modes. In this paper, we delve into the operational aspects of such a system and present an algorithmic framework for operational optimization. Through simulations conducted in the city of Chicago, we showcase the scalability of the proposed integrated system and quantify its benefits.

ABSTRACT. Ride-hailing system is an emerging service within urban scenarios which shows its potential to improve service quality for customers. A critical operational challenge is the problem of imbalance between vehicle supply and customer demand. In this paper, we investigate a hierarchical control framework for repositioning empty vehicles. For the upper level, we introduce an MFD-based model to describe the dynamics of both the taxi and private vehicles and an MPC controller is designed to instruct the transferring repositioning of vehicles between regions; for the lower layer, the fleet location optimization problem is solved as a coverage control problem, which can be carried out by each vehicle to generate its own position guidance. To bridge between the upper and lower layers, an assignment problem minimizes the rebalancing cost of complying with the objectives of both layers. The effectiveness of the proposed method is verified by a discrete space simulator modeled the real road network of Shenzhen, China. The proposed hierarchical framework yields its advantages by serving more passengers with less waiting time.