ABSTRACT. Discrete choice models (DCM) are essential tools in travel demand analysis, providing a powerful theoretical econometric framework for understanding and predicting choice behaviors. These models are formed as random utility models (RUM), with their key advantage of interpretability. However, a core requirement for the estimation of these models is a priori specification of the associated utility functions, making them sensitive to modelers' subjective beliefs. Recently, machine learning (ML) approaches have emerged as a promising avenue for learning unobserved non-linear relationships in DCMs. However, ML models are considered "black box" and may not correspond with expected relationships.

To address this issue, this study presents a novel framework designed to enhance data-driven approaches for DCM by developing interpretable models that incorporate domain knowledge and prior beliefs through constraints. The proposed framework includes pseudo data samples that represent required relationships and a loss function that measures their fulfillment, along with observed data, for model training. The framework's independence from model structure allows integration with off-the-shelf DNN architectures, combining ML's specification flexibility with econometric principles and interpretable behavioral analysis to improve model interpretability.

The proposed framework was applied on the Swissmetro mode choice dataset, examining the incorporation of domain knowledge in Deep Neural Network (DNN) models and its effect on economic information. The methodology was applied to two model architectures: a standard DNN and the alternative specific utility deep neural network (ASU-DNN) (Wang, et al., 2020), in both unconstrained and constrained versions, with a Multinomial Logit (MNL) model estimated for comparison.

The case study demonstrated the framework's effectiveness in discrete choice analysis, illustrating how it combines the strengths of ML and econometric approaches to produce more accurate and interpretable models, thus better informing decision-making processes. The constrained models (C-DNN and C-ASU-DNN) demonstrated improved generalizability, while unconstrained models (DNN and ASU-DNN) showed better empirical fitting to the data. Market share predictions revealed the constrained models outperforming unconstrained models in unseen data. Choice probability functions illustrated the constrained models' ability to avoid inconsistencies present in unconstrained models. It also underlined the range of differences in analyses (e.g., predicted values of time and choice probabilities) that can arise from the different models. The case study suggests further research to include additional domain knowledge (e.g., magnitudes of elasticity and values of time) and test it on richer datasets to explore further the implications of incorporating domain knowledge for choice analysis.

ABSTRACT. This study aims to understand the factors that influence individual's travel mode choices to their main activities. The dataset for the analysis is an app-based household travel survey conducted in the Tel-Aviv metropolitan area collected information from 13,485 households with 39,090 household members and 333,724 trips, which was enriched with zonal variables such as population and employment. The raw dataset is organized as a trip-by-trip fashion, where each individual trip is treated separately. To account for trip dependency, we created a tour-based dataset in which each individual’s travels are chained. We trained different classification models to predict the main mode of travel for individuals. In line with the literature, the possession of a car license was the most significant feature, which formed the basis for data partition. By examining the feature important scores, we identified the most influential features for classifying each travel mode and investigated their relationship with personal characteristics or area constraints. The findings provide insights for policymakers and urban planners to make informed decisions about transportation infrastructure.

The objective of this study is to further analyze the most important features that help predict each class of the target variable (i.e., private car, public transportation, and active transportation) and interpret them in terms of personal characteristics or area constraints. This analysis will provide insights into the factors that influence people's travel choices. To achieve this objective, we pre-processed and normalized the data before using it to train several classification models, including Decision Trees, Random Forest, XGBoost, and MLP. We tuned the hyperparameters of each model using GridSearch and evaluated their performance using accuracy metrics. We also examined the importance of each feature in predicting transportation modes using feature importance techniques. In addition, we found that the feature car license was present in the top 5 features for every model. For this reason, we decided to partition the dataset into two subsets according to the car license possession. We then run the best two models (RF, XGB) on the partitioned datasets. The results of the study indicate that the XGBoost algorithm outperformed the other algorithms with an accuracy of 88.2% on the full dataset and 90.7% on the partitioned dataset with the absence of a car license. we found that the most important features for predicting each class of the target variable were related to tour characteristics, such as car license, tour distance and number of cars in the household. Relatively few important features were related to personal characteristics such as age, and household characteristics such as the institution type. We found that the dataset without car licenses achieved the highest accuracy on the classification task. This result may be related to the fact that a person with a driver license and available vehicle will choose to drive it instead of using PT, even with traffic congestion. This is because the PT alternative in the Tel-Aviv metropolitan area is not very attractive. This paper focused on comparing classification models and their results such as feature importance and data partitioning based on the most important feature.

ABSTRACT. In semi-autonomous transportation systems, vehicle autonomy capabilities are utilized in a partial manner, so as to adhere to current regulations. Hasbe et al. (2017b), presented a station-to-station last mile transportation service which is provided by platoons composed of human-driven lead vehicles and several autonomous cabins. While the lead vehicles travel non-stop between the stations of the system, the cabins have the capability to detach from the convoys at the proximity of stations and autonomously travel in and out of the stations, in order to pick-up and drop-off passengers. Repoux et al. (2021) termed this system as the “Multi-Layered Personal Transit System”, defined the operational problem arising in this system and proposed an operational concept for it . In particular, their concept determines fixed routes for the lead vehicles. In this study we examine the potential of dynamically shortcutting the routes of the lead vehicles so as to better respond in real-time to the current system state and the upcoming demand. For this purpose, we develop an approximate model of the system and formulate it as a Markov Decision Process (MDP). In particular, the MDP represents: (1) the system state, i.e., the cabin workload on each route; (2) an action, i.e., how many lead vehicles should shortcut on each route; and (3) the reward, which is modeled as a penalty based on the number of cabins waiting to be served. Due to the so-called ‘curse of dimensionality’ in both the states of the system and the potential actions, the MDP cannot be solved straightforwardly. Therefore, we develop an event-based simulation to represent the system’s dynamics and employ a Reinforcement Learning (RL) approach in order to take good shortcut decisions dynamically. We conduct a numerical experiment based on a network consisting of 20 stations, 50 cabins, and 12 lead vehicles, designed to examine the potential benefits of using RL to improve the performance of the system. At a first step, we train the RL model using 5000 simulation runs. Then, at the second step, we use the Q-table generated by the RL model to derive two explainable policies which are easier to communicate and operate: "Two Q Policy" and "Fix Policy B". Testing the four policies using 500 different demand scenarios show that the RL policy records the greatest average reduction in the waiting time at 3.96%, followed by the Two Q Policy at 2.36%, and Fix Policy B at 1.76%, as compared to the original fixed route policy.

ABSTRACT. Deep neural networks (DNNs) outperform manually-crafted systems and are increasingly integrated into autonomous vehicles, traffic incident inference systems, traffic signal timing systems, and more. Due to their ability to solve complex problems with excellent performance, DNNs will be increasingly used in smart transportation systems. However, this technology still suffers from severe limitations, since DNNs can still make errors when faced with previously unseen inputs. These mistakes can be catastrophic for a DNN-powered safety critical system, because they could risk lives. Safety and assurance of smart transportation systems must therefore be enhanced.

One promising approach to addressing this challenge is by extending DNN-powered systems with hand-crafted override rules, which override DNN output when certain conditions are met. We advocate crafting such override rules using the well-studied scenario-based modeling (SBM)} paradigm, which produces rules that are simple, extensible, powerful enough to ensure DNN safety, and can also be formally verified for correctness. Our long-term goal is to develop a public SBM framework that will enable users to identify cases when an override rule is required, and then implement it quickly and efficiently.

Our first research plan was to conduct case studies, which demonstrate our approach's feasibility. For this end we introduced override rules to two well known DL projects, using the existing SBM tools. Through this work, we identified what additional functionality is needed to facilitate override rules. We plan to implement these capabilities and empirically validate the package’s usability. Finally, we will measure the value of the new package through an empirical evaluation in smart transportation.

We recently extended a DL system aimed at solving the problem of mapless navigation. The system contains a DNN agent whose goal is to navigate a Turtlebot 3 (Turtlebot) robot to a target without collisions. Fig.1 shows that enabling the override rule leads to a significant reduction in collisions. These results increased our confidence that our approach is viable. We hope this work will promote the safe usage of DNNs in smart transportation systems, and will constitute a step forward towards the goal of zero casualties.

ABSTRACT. The imminent penetration of low-altitude passenger and delivery aircraft into the urban airspace will give rise to new urban air mobility (UAM) systems. Future UAM systems exemplify a new class of modern large scale engineering systems. Hence, realistic simulation tools are needed to pave the way towards future investigations in UAM systems. The current white paper aims at setting the foundation for such simulation tools, establishing general guidelines, specifying the fundamental concepts, and proposing methodologies for implementation

ABSTRACT. As global cities suffer from increasing traffic congestion and especially in the Tel Aviv metropolitan area there is a noticeable increase in congestion levels, shared mobility has the potential for reducing congestion and other externalities and enhancing transportation systems. To shift com-muter behaviors toward shared mobility (e.g., ridesharing, carpooling, work shuttles, and vanpools) we need to identify the underlying factors that can cause such a shift. The study aims to understand the factors affecting the choice to carpool including the distinction between carpooling with household versus non-household members, using discrete choice models. We develop a multinomial logit model based on the Tel Aviv Metropolitan Travels' Habits survey. The survey was conducted in two phases: phase one in 2014 and phase two in 2017 with more than 13,500 households, 39,000 individuals, and 300,000 trips/activity. The survey trips were transformed into tours, and we focus on tours that started and finished at home. The derived factors from the data include household and individual sociodemographic characteristics, and commute and activity pattern attributes. Spatial and built environment variables are also included based on commuters' origins and destinations. Furthermore, the level of service data (e.g., transit time, auto time, park cost, walk distance, waiting time, access time) for tours within the metropolitan area are used based on The Tel Aviv activity-based model run. The level of service data for tours that arrive or depart outside the Metropolitan are calculated in real-time through the directions service in the Google Maps Platform, which offers its own programming interfaces (APIs), which can return information regarding distances or travel duration between different places given a travel mode. To provide more insight into the actual individual choice behavior, we integrate a car allocation model into the mode choice model. The suggested choice set for the mode choice model are: Solo Driver, Driver with a rider, A Rider, Transit, and Non-motorized (The shared choices are designed to consider household carpools and non-household carpools separately). An extensive multinomial logit model based on a massive amount of Revealed Preference (RP) data (Total home-based tours sample size 80,000), allows us to estimate the probability that an individual will choose to share a ride showing the impact of the various variables: demographic, commute, spatial, car allocation, and level of service factors. The information will be analyzed to extract com-muter actual behavior patterns in the Tel Aviv Metropolitan area, the behavioral details of com-muting mode choice and carpooling will help service providers to enhance the popularity of ride-sharing systems.

ABSTRACT. 1. Introduction

The prospect of automated mobility on-demand (AMoD) services has garnered significant research, policy and economic interest. However, the prognosis on AMoD availability is mixed, while its interaction with current mobility pattern types and impacts on system performance has yet to be fully determined. Various experiments have demonstrated how AMoD could impact urban mobility under various cost, demand, supply (fleet control and management) and policy scenarios in either specific cities (Azevedo et al., 2016; Vazifeh et al., 2018; Alonso-Mora et al., 2017; Santi et al., 2014) or on virtual cities (Basu et al., 2018; Wen et al., 2018; Zhang and Guhathakurta, 2017; Gross et al., 2019). These studies highlight the importance of understanding how differences in urban form and behavior affect moderate the impacts of new modes such as AMoD (Nahmias-Biran et al., 2020).

2. Methods

Our objective is to examine the environmental impacts of future AMoD strategies in dense transit-oriented cities, in order to provide insights for sustainable implementation. Our simulation experiments are performed on a futuristic Tel Aviv metropolitan area for 2040, which will include massive investments in public transportation. Tel Aviv metropolitan area was modeled using a simulation framework which is a hybridization of SimMobility MT and Aimsun next. Further details are provided in (Nahmias-Biran et al., 2022; Dadashev et al., forthcoming).

2.1. Scenario Design To analyze the impacts of AMoD across Tel-Aviv metropolitan area, we consider four scenarios. Through these, we explore plausible AMoD futures in which passive or active strategies are employed to manage AMoD across the city.

2.1.1. Base Case The Base Case represents future conditions for a futuristic Tel Aviv for 2040 in terms of mode availability and choice, and network performance.

2.1.2. AMoD Intro Earlier studies investigating the cost implications of AMoD provide context for this cost starting point (Pavone, 2015; B¨osch et al., 2018). The AMoD service offers both single and shared ride (pooling) options to allow for additional fare and energy consumption reductions. Shared AMoD rides are 30% cheaper than single AMoD rides.

2.1.3. AMoD Transit Integration The potential of the integration of on-demand with transit to improve connectivity has received considerable attention (Scheltes and de Almeida Correia, 2017; Shen et al., 2018; Wen et al., 2018) as a viable urban policy. Therefore, we explore this strategy whereby AMoD is subsidized by 20% for shared access-and-egress connectivity to rail stops.

2.1.4. AMoD + Car Reduction Recent studies have indicated that AMoD could encourage a reduction in car ownership (Firnkorn and Mu¨ller, 2015; Giesel and Nobis, 2016). We thus test such an approach where we simulate the effects of restricted car ownership (by 25%) along with the use of a low- cost AMOD service in lieu of traditional MoD.

3. Results Our results indicate that AMoD increases congestion and passenger VKT in dense, transit-oriented city such as Tel-Aviv metropolitan area. Integrating AMoD with transit or introducing it in conjunction with reducing household car ownership mitigates the impact on VKT. More critically, these two interventions reverse the cannibalization effect of AMoD on transit, as well as resulting in significant energy gains, under a fully electrified fleet. None of the AMoD strategies considered reduce the impact of AMoD on congestion, compared to base case. We find, however, that a policy that results in lower car ownership is more effective at reducing energy consumption and greenhouse gas emissions.

References

Azevedo, C.L., Marczuk, K., Raveau, S., Soh, H., Adnan, M., Basak, K., Loganathan, H., Desh- munkh, N., Lee, D.H., Frazzoli, E., et al., 2016. Microsimulation of demand and supply of autonomous mobility on demand. Transportation Research Record: Journal of the Transporta- tion Research Board , 21–30.

Basu, R., Araldo, A., Akkinepally, A.P., Biran, B.H.N., Basak, K., Seshadri, R., Deshmukh, N., Kumar, N., Azevedo, C.L., Ben-Akiva, M., 2018. Automated mobility-on-demand vs. mass transit: A multi-modal activity-driven agent-based simulation approach. Transportation Research Record 0, 0361198118758630. URL: https://doi.org/10.1177/0361198118758630, doi:10.1177/0361198118758630, arXiv:https://doi.org/10.1177/0361198118758630.

B¨osch, P.M., Becker, F., Becker, H., Axhausen, K.W., 2018. Cost-based analysis of autonomous mobility services. Transport Policy 64, 76–91. URL: http://www.sciencedirect.com/science/ article/pii/S0967070X17300811, doi:10.1016/j.tranpol.2017.09.005.

Dadashev, G., Levi, Y., Nahmias-Biran, B. )forthcoming(. Implications of de-carbonization policies using an innovative simulation tool. Transportation Research Part D: Transport and Environment.

Firnkorn, J., Mu¨ller, M., 2015. Free-floating electric carsharing-fleets in smart cities: The dawn- ing of a post-private car era in urban environments? Environmental Science & Policy 45,30 – 40. URL: http://www.sciencedirect.com/science/article/pii/S1462901114001774, doi:https://doi.org/10.1016/j.envsci.2014.09.005.

Gross, E., Oke, J., Akkinepally, A., Nahmias-Biran, B.h., Azevedo, C.L., Zegras, P.C., Ferreira, J., Ben-Akiva, M., 2019. Accessibility and energy consumption evaluation under different strategies of mobility on-demand deployment, in: 98th Annual Meeting of the Transportation Research Board.

de Lima, I.V., Danaf, M., Akkinepally, A., Azevedo, C.L.D., Ben-Akiva, M., 2018. Model- ing framework and implementation of activity- and agent-based simulation: An applica- tion to the greater boston area. Transportation Research Record 0, 0361198118798970. URL: https://doi.org/10.1177/0361198118798970, doi:10.1177/0361198118798970, arXiv:https://doi.org/10.1177/0361198118798970.

Mu¨ller, K., Axhausen, K.W., 2012. Multi-level fitting algorithms for population synthesis. Arbeits- berichte Verkehrs-und Raumplanung 821.

Nahmias-Biran, B., Oke, J.B., Kumar, N., Akkinepally, A.P., Azevedo, C.L., Ferreira, J., Zegras, P.C., Ben-Akiva, M., 2020. Who benefits from AVs? equity aspects of autonomous vehicles policies in full-scale prototype cities, in: 99th Annual Meeting of the Transportation Research Board. In review.

Nahmias-Biran, B., Oke, J.B., Kumar, N., Basak, K., Araldo, A., Seshadri, R., Akkinepally, A., Azevedo, C.L., Ben-Akiva, M., 2019. From traditional to automated mobility on demand: A comprehensive framework for modeling on-demand services in simmobility. Transportation Research Record 0, 0361198119853553. URL: https://doi.org/10.1177/0361198119853553, doi:10.1177/0361198119853553, arXiv:https://doi.org/10.1177/0361198119853553.

Nahmias-Biran, B., Dadashev, G., Levi, Y. (2022). Demand Exploration of Automated Mobility on-Demand Services Using an Innovative Simulation Tool. IEEE Open Journal of Intelligent Transportation Systems. vol. 3, pp. 580-591, 2022, doi: 10.1109/OJITS.2022.3197709. Pavone, M., 2015. Autonomous mobility-on-demand systems for future urban mobility, in: Au- tonomes Fahren. Springer, pp. 399–416.

Scheltes, A., de Almeida Correia, G.H., 2017. Exploring the use of automated vehicles as last mile connection of train trips through an agent-based simulation model: An applica- tion to elft, Netherlands. International Journal of Transportation Science and Technology 6, 28 – 41. URL: http://www.sciencedirect.com/science/article/pii/S2046043016300296, doi:https://doi.org/10.1016/j.ijtst.2017.05.004. connected and Automated Vehicles: Ef- fects on Traffic, Mobility and Urban Design.

Wen, J., Chen, Y.X., Nassir, N., Zhao, J., 2018. Transit-oriented autonomous vehicle oper- ation with integrated demand-supply interaction. Transportation Research Part C: Emerg- ing Technologies 97, 216–234. URL: http://www.sciencedirect.com/science/article/pii/ S0968090X18300378, doi:10.1016/j.trc.2018.10.018.

Zhang, W., Guhathakurta, S., 2017. Parking spaces in the age of shared autonomous vehicles. Transportation Research Record: Journal of the Transportation Research

Board 2651, 80–91. URL: https://doi.org/10.3141/2651-09, doi:10.3141/2651-09, arXiv:https://doi.org/10.3141/2651-09.

ABSTRACT. This paper presents a new ridesharing simulation platform that accounts for dynamic driver supply and passenger demand, and complex interactions between drivers and passengers. The proposed simulation platform explicitly considers driver and passenger acceptance/rejection on the matching options, and cancellation before/after being matched. New simulation events, procedures and modules have been developed to handle these realistic interactions. The capabilities of the simulation platform are illustrated using numerical experiments. The experiments confirm the importance of considering supply and demand interactions and provide new insights to ridesharing operations. Results show that increase of driver supply does not always increase matching option accept rate, and larger matching window could have negative impacts on overall ridesharing success rate. These results emphasize the importance of a careful planning of a ridesharing system.

ABSTRACT. On-board chargers for PHEVs and EVs are AC-DC converters capable of one or three phases, incorporating a Power Factor Correction Rectifier (PFCR) as their primary stage. As modern grid codes impose strict power quality requirements, the primary goal of such converters is to shape the mains current to be nearly sinusoidal and in phase with the grid voltage to comply with Total Harmonic Distortion (THD) and Power Factor (PF) standards. PFCRs are supplemented by short-term energy storage components (large high-voltage electrolytic capacitors) to deal with pulsating power components, active power mismatches during transients and provide hold-up time if necessary. Electrolytic capacitors required for discussed applications suffer from well-known issues related to reliability, physical size, and relatively high cost. Fig. 1 illustrates a typical structure of an onboard charger with a commercial product featuring four electrolytic capacitors connected in parallel. As an alternative to traditional electrolytic capacitors, Electronic Capacitors (ECs) are gaining popularity in recent years in various industries, including smart transportation systems, charging stations for electric vehicles and plug-in hybrid electric vehicles, and onboard charging systems. ECs can primarily be used to improve the overall performance of these systems, reducing ripple and smoothing out voltage fluctuations. This helps enhance the systems' efficiency, reliability, and safety while reducing costs. Additionally, electronic capacitors are considered more environmentally friendly than traditional electrolytic capacitors due to their longer lifespan and ability to withstand higher temperatures, reducing the risk of fire. This reduces the need for frequent replacements, making them a more sustainable option. An Electronic Capacitor (EC) is a bidirectional DC-DC converter that utilizes a small auxiliary capacitance and mimics the low-frequency behavior of a different (typically much higher) capacitance. Since onboard chargers typically do not encounter dynamic transients and do not require hold-up requirements, a two-fold reduction in capacitance is utilized in such chargers. The reduction in capacitance, however, results in a two-fold increase in DC-link voltage ripple, thereby imposing double stress on electrolytic capacitors. The utilization of the proposed Electronic Capacitor resolves this issue, as its capacitance is an adjustable parameter and can be set to the required value. The advantages, estimated cost, size, and lifespan of electrolytic capacitors for onboard chargers will be presented at the conference.

ABSTRACT. Reducing the fuel consumption of ground transportation systems is one of the central answers to the challenge of reducing greenhouse gas emissions and air pollution. Aerodynamic shaping flow control can reduce the drag and consequently fuel consumption. In this study, road tests were conducted comparing the fuel consumption of an unmodified, blunt trailer aft section with a modified reflex profiled aft section attached to the trailer rear-end. The fuel consumption was measured using the fuel economy system of the towing SUV. A reduction of 10% of the combined car-trailer fuel consumption was measured with the reflex profile. Additional reduction of 7% was achieved due to active flow control.

ABSTRACT. In these days, the variety of portable electronic devices and electric vehicles is growing. Development of high energy density electrochemical energy storage and conversion technologies, especially batteries, became crucial for the progress of these technologies. Today, the leading technology is Li-ion battery due to its energy and power densities, safety issues and mass production ability. However, this technology depends on resources like lithium, graphite, copper, and some transition metals that are either available in limited quantities and/or geographically unequally distributed.1 In recent years it came to understanding that batteries based on metallic anodes has the potential to deliver higher energy density than that of Li-intercalating anode materials.2 Due to their low reduction potential and high volumetric and gravimetric capacities, Lithium and Magnesium are suitable to serve as an anode material. Except its high volumetric capacity and low reduction potential, Mg metal possesses several additional advantages over Li metal in terms of: non-dendritic growth, low price, high abundance, environmental issues and safety.3 Despite the advantages of magnesium as an anode, the development of secondary Mg batteries encountered two major challenges:4 1) High charge density of Mg2+ creates strong interaction, forming strong Mg complexes with the solvents, causing energy-lost during de-solvation, and strong interaction between the magnesium anions and the solid intercalation host resulting with slow solid-state diffusion of Mg2+ ions, and thus low efficiency. 2) Most of the electrolyte solutions, supporting high reversibility of Mg deposition contain chlorides. Cl- causing corrosion of the cell components, that limits the cell cyclability,5 and force us to work with noble metals as cell components only. Chlorides have two important roles in Mg-based systems. Cl anions complexes with Mg-ions that facilitate the dissolutions process of the Mg2+-ions.6 Moreover, chlorides form surface film on the cathodes the reduces the activation energy of Mg2+ ions transition between the solution and the cathode material.7 A new class of halogen-free electrolyte solutions is investigated for the last years8, but the electrochemical performances of them is not satisfying.9 The objective of this research is to study the effect of chlorides on halogen-free electrolyte solutions, in order to develop a new, improved, chloride-free solution for magnesium batteries. To do so, Mg[B(HFIP)4]2/DME electrolyte solution is used as a case study. Its electrochemical a performances were fully evaluated.9 The effect of chlorides on the system is a work in progress, and is yet to be published. The next step is to test the high-potential cathode/Cl-free electrolyte solution/Mg-metal anode compatibility. Current results show that Mg[B(HFIP)4]2/DME support reversible Mg deposition-dissolution, but performs high over-potential for magnesium deposition, low coulombic efficiency, and it must go through pre-treatment, ‘conditioning’, process to present its best performance. Moreover, it seems that chlorides have positive effect on over-potential, efficiency, and full-cell capacities. At the end of this work, it will be clearer whether developing chlorides-free electrolyte solutions is the correct way or should the academy focus on developing stainless cell components for magnesium batteries.

ABSTRACT. Many new technology initiatives that rely on vehicle autonomy capabilities have emerged in recent years. One prominent concept is based on Autonomous Mobile Robots (AMRs). In such systems, small-wheeled robots provide point-to-point deliveries on sidewalks at pedestrian speed. They are powered by small batteries, limiting their service range to around 3 km. The system considered in this work consists of a fleet of vehicles distributed in multiple mini-depots along the service area. These robots are assigned to service requests characterized by pick-up and delivery locations and corresponding service time windows. We examine the potential of enhancing the service by public transit. That is, allowing the robots to fulfill parts of their journey on board public transit vehicles. As the robots do not discharge while traveling on board the public transit vehicles, this extension comprises multiple opportunities. First, the service range can be extended and in some cases, service durations can be shortened. Second, the overall energy consumption can be reduced. The operational planning problem in the studied AMR based services represents a special case of the well-known Pick-up and Delivery Problem (PDP), with Full Truck Load and multiple modes of transport. We develop two mixed integer programming formulations for the problem: an arc-based formulation and a route-based formulation. The arc-based formulation explicitly represents each robot's potential leg and decides upon the legs to be traveled. The latter considers complete feasible routes with the aim of selecting the best route (and robot) for each request. While route-based formulation has a more compact structure, the number of routes that may be considered grows exponentially with the number of requests. To overcome this, we develop a column generation approach. Specifically, we define an initial set of potentially good routes for each robot-request pair and then formulate the underlying sub-problem of finding new promising routes as a resource constrained shortest path problem. We develop a four-stage dynamic programming algorithm to solve the sub-problem problem. Subsequently, by exploiting robot-request symmetries we are capable of reducing significantly the number of sub-problems solved at each iteration. In addition, as the robot-request sub-problems are independent, we apply parallel computing to solve multiple sub-problems simultaneously. These actions allow us to reduce the computing time required for each column generation iteration. A numerical experiment is conducted using 700 problem instances with varying numbers of requests, robots, and public transit nodes. Results show that while the arc-based formulation only enables solving instances with up to 15 requests, the column generation approach can solve instances with up to 150 requests in less than five minutes.

ABSTRACT. Restaurant meal delivery has been rapidly growing in the last years. The main challenges in operating it are the temporally and spatially dispersed demand from different customers all over town as well as the customers expectation of timely and fresh delivery. To overcome these challenges new business concepts emerge, called “Cloud kitchens”, “Ghost kitchens”, or “Dark kitchens”. These concepts propose food preparation of several restaurants in a central complex in the city, thus, exploiting pooling benefits.

In this research, we propose operational strategies for the effective operations of ghost kitchens. We model the problem as a sequential decision process with the aim to minimize the average delay. For the complex, combinatorial decision space, we propose a large neighborhood search that is based on properties that we developed for the optimal solution. Within the large neighborhood search, decisions are evaluated via a value-function-approximation, enabling anticipatory and real-time decisions. Last, we conduct numerical experiments to demonstrate the performance of our heuristic and to derive managerial insights.

ABSTRACT. The small parcel delivery industry has grown significantly over the last few years. From an economic perspective, logistic operations in urban areas, notably last-mile delivery, tend to be the most expensive segment of the logistic process. In this study, we adopt a novel framework to economize the delivery of small parcels using the Physical Internet (PI) concept. In a PI, objects are carried from their origin to their destination in a path that may contain several legs and interchange (transfer) nodes. The additional flexibility obtained by the possibility of transferring objects between vehicles can economize the delivery process. However, these transfer operations require additional resources and time for loading and unloading the objects in the interchange nodes. Thus, the optimal routing of objects must consider the effect of these transfers, on both their travel and service times (the time required for loading and unloading). Most of the routing literature assumes that service times are unrelated to the number of loaded and unloaded objects, even though they constitute a significant share of logistic operations. In a PI setting, endogenizing the service time into the model affects the coordination between the visits of the vehicles at the transfer nodes. This coordination complicates the routing and scheduling problem significantly but is crucial for the applicability of the model. We consider a hyper-connected network of service points(SPs) that can be used for drop-off, pickup, and transfer locations, and parcels may be transferred from their origin to their destination in several legs. We present a method to route and schedule the parcels while considering the service times of the vehicles at the SPs. The objective is to minimize the parcels' steady-state expected delivery time. As a first step in analyzing this intricate problem, our model assumes that the routes of the vehicles are fixed and determined in advance. We present a dynamic routing policy for the parcels in a given network and a simulation-based search algorithm to improve the vehicles' schedule, with the aim of minimizing the parcels' steady-state expected delivery time. A preliminary experiment demonstrated the merits of this approach.

ABSTRACT. We address the problem of allocating resources of limited supply to several agents whose demand is uncertain and revealed only upon arriving at them. The challenge is to determine the allocation amount to a given site upon arriving at it when the demand of the subsequent sites along the route is unknown. The goal is to maximize both effectiveness, measured by the total number of units distributed, and the fairness (equity) among the agents. This objective arises in many applications of urban logistic settings such as food distribution to agencies or a massive vaccine campaign after a sudden outbreak. We formulated the problem as an efficient linear program that represents all possible realizations compactly. We then developed structural properties and a heuristic solution method based on them. In practice, the allocation decisions are likely to be controlled via simple rules of thumb, and in this work, we suggest implementing optimization-based decisions that are shown to provide better performance.

ABSTRACT. This work presents an algorithm for real-time junction scheduling towards the non-stopping junction. We demonstrate the results that imply road safety as actions are remotely controlled, by using the SUMO simulator [1]. The main idea of the non-stop VTL is to synchronize the vehicles coming to the junction in such a way that vehicles do not stop at all. They will not stop to let others cross the junction or wait for others to evacuate the junction. The timing principle is “first come, first served”. That is, the vehicles that are closer to the junction by a given distance D are the ones that are timed first crossing the junction. In other words, the order of priority between the lanes entering the junction is equal. That is why there is no starvation; this principle produces a junction in which the vehicles cross in analog to a zipper.

ABSTRACT. This research aims to extend the velocity obstacles (VO) method to account for robot dynamics. The VO is a real-time obstacle representation in the velocity space, which allows motion planning in dynamic environments, especially for autonomous vehicles (AV). The VO will be extended by integrating it with a dynamical model, which will contain the dynamic constraints of a car-like vehicle. The extension for achieving the restrictions of the vehicle dynamical model is building the vehicle’s feasible range of speeds and accelerations (FSA), which represents the feasible accelerations of the vehicle for various speeds along its path. The FSA will consider the vehicle's dynamical structure (bicycle model) and the acting forces between its tires and the ground. The feasible accelerations and speeds that will be found by the FSA, will be incorporated into a VO-based controller for real-time motion planning in dynamic environments. The extended algorithm will be tested and verified by conducting a series of experiments in a ROS-based simulator, under specific tracks and various dynamic environments for a vehicle with a VO-based controller that contains the suggested dynamical model. The research provides better obstacles avoidance capabilities alongside safer dynamic behavior for VO-based controllers, which may accelerate the use of the method in the AV industry.

ABSTRACT. In recent years, the development of a new mode of urban air transport, the low-altitude passenger and delivery aircraft, emerges from the advancement of aviation and communication technologies. This advances the concept of urban air mobility (UAM), leading to the invention of new urban air transport systems, namely low-altitude air city transport (LAAT). Such systems will include aircraft operated with or without pilots, transferring passengers and goods in urban areas using low-altitude levels of urban airspace.

Connectivity and digitalization will enable new control measures in aviation operations and open new ways for integrating these measures in real-time urban traffic management. Hence, new control strategies can be designed to regulate LAAT demand and supply. This can be achieved by manipulating aircraft departures, aircraft routings, aircraft speed, etc. \cite{Haddad2021} proposed that aircraft flows can be controlled at the boundary between urban regions. In this research, new model-based control strategies will be designed, where aircraft departure management and boundary control will be integrated. While the urban ground systems are still limited in controllability, the aviation operation can benefit from the proposed flow-oriented control paradigm, which can balance the LAAT system's \emph{supply} and \emph{demand}. It should be noted that LAAT controlled systems might create system queues, which greatly impact the system's efficiency and can lead to environmental and economical challenges. Therefore, it is worth investigating the interaction and synchronization of controlling both the \emph{input} (demand) and \emph{transfer} traffic flows (supply) of LAAT systems.

This research investigates the development of traffic flow management and control strategies for LAAT systems. For this reason, a new LAAT framework is developed that couples modeling and controls LAAT systems, and integrates the two aggregation levels, i.e., microscopic and macroscopic, as shown in Fig.~\ref{fig:LAATframework}. The developed framework is an extension of the LAAT simulation that was recently developed by \cite{Safadi2023TRC}. This research contributes to the literature with a novel LAAT framework that (i) models the air traffic flow and (ii) implements model-based traffic control strategies.

ABSTRACT. We study a fundamental NP-hard motion coordination problem for multi-agent transport systems: We are given a graph $G$ and set of agents, where each agent has a given directed path in $G$. Each agent is initially located on the first vertex of its path. At each time step an agent can move to the next vertex on its path, provided that the vertex is not occupied by another agent. The goal is to find a sequence of such moves along the given paths so that each agent reaches its target, or to report that no such sequence exists.

The problem models guidepath-based transport systems, which is a pertinent abstraction for traffic arising in a variety of contemporary applications, ranging from smart intersections and autonomous vehicles operating in dedicated infrastructure (e.g., guideways, rails, or special lanes) to industrial material handling and robotics. As such constrained autonomous systems evolve beyond simple topologies and fixed schedules, i.e., to cater on-demand service, they will require more complex motion coordination.

We provide a fine-grained tractability analysis of the problem by considering new assumptions and identifying minimal values of key parameters for which the problem remains NP-hard. Our analysis identifies a critical parameter called \emph{vertex multiplicity} (VM), defined as the maximum number of paths passing through the same vertex. We show that a prevalent variant of the problem is NP-hard even when VM is 3. This contrasts previous hardness constructions, in which VM was unbounded. On the positive side, for VM $\le$ 2 we give an efficient algorithm that iteratively resolves cycles of blocking relations among agents. We also present a variant that is NP-hard when the VM is 2 even when $G$ is a 2D grid and each path lies in a single grid row or column. By studying highly distilled yet NP-hard variants, we deepen the understanding of what makes the problem intractable and thereby guide the search for efficient solutions under practical assumptions.

ABSTRACT. Over the last few years, we have seen a rise in the use of GPS and internet-enabled navigation systems for drivers. A few options exist in the market today, and many of them use technology to provide additional benefits beyond simply finding a path from point A to point B and providing directions. Waze, for example, allows drivers to send updates to other users about car accidents, traffic, police presence, and even objects on the road. It also allows users to edit the map, add new roads as they open, and mark road closures. Google Maps, another highly popular option, also deals with pedestrian directions, public transport, and bicycles. Google Maps also provides real-time traffic information based on location data that it has from its users. Apps for navigation provide many benefits for drivers, but they can come at a cost to other users of the roads. Drivers seeking shortcuts are driven by the directions of a naviga-tion app to make use of roads that are smaller and not meant for passing traffic. For example, in the UK between 2008 and 2019 the number of cars rose by 12%, in that same time the number of cars making use of type ‘A’ roads (major roads intended to provide large-scale transport links within or between areas) rose only 1%. In contrast, traffic on minor roads rose by 36% in that same time. Similar complaints have risen in other countries such as Israel and the USA to name a few. Residents even resort to tactics like filing false traffic reports to prevent passing traffic. The rise in passing traffic in small neighborhoods has a few concerning side effects such as the increased risk of car accidents involving pedestrians, increased emissions that can affect the health of the local population, increased noise, and many other disruptions to daily life. In this project, our goal is to conceptually plan alterations to existing popular navigation systems to provide more ethical considerations and increase utility for the other ethical stakeholders. This could be accomplished by adding more time to spe-cific routes where added traffic is a danger or a nuisance to the local people, For ex-ample near schools, parks, in quiet neighborhoods, etc. By adding time to the map, we can make small impacts on the navigation algorithm without impacting the amount of time the algorithm takes to run. The modifications made should be determined by the needs and requests of the local community and government. This is especially important when it comes to traffic policy because these policies can change from country to country and state to state. We view the tools that drivers use to travel and drive to be just as integral a part of the entire transportation system as policies, infrastructure, and driving culture. This presentation aims to showcase some of the various ethical problems that exist in navigation apps today and to suggest some modifications that can be made to min-imize those. Our base premise is that the modifications need to be implemented care-fully and with consultation with the local community to ensure their values are pre-served and their quality of life is not negatively impacted.

ABSTRACT. This paper presents a combined mode choice and traffic assignment model that considers Autonomous Vehicles (AVs), Human Driven Vehicles (HDVs) and public transportation. In contrast to previous studies, in this paper the demand for AVs is endogenously considered. The developed model aims to study how the system equilibrium will evolve in different possible deployment scenarios of AVs. The model is flexible and can accommodate different route choice paradigms. In addition, the model is able to capture waiting times for shared use of AVs. The paper compares the effects of AVs as a private mode or as a shared service and measures their expected effects on the network.

ABSTRACT. On-street parking occupancy and effect of cruising for parking on urban traffic are defined by three major parameters – (1) the rate of car arrivals, (2) the dwell time of already parked cars, and (3) the willingness of drivers to continue their search for a vacant parking spot if they have failed until now. We investigate a series of theoretical and numeric models, both deterministic and stochastic, that describe parking dynamics in an area as dependent on these three parameters, over the entire spectrum of the demand-to-supply ratio, focusing on the case when the demand is close to or above the supply, and present the basic estimates of the system's characteristics, including formula for the distribution of the cruising time.