ABSTRACT. Despite extensive research on e-commerce supply chain optimization, some operational constraints remain underexplored. One critical factor is case-breaking at warehouses, where goods shipped in large quantities (case-packs) are disaggregated into individual units to meet retailers’ preference for small, frequent replenishments. Limited processing capacity may prevent all cases from being broken upstream, making it important to decide which cases to disaggregate and how to distribute both individual units and intact cases. Motivated by this, we study a case-pack allocation problem with case-breaking capacity constraints. Assuming deterministic warehouse inventory and target inventory levels at retailers, we optimize a concave, nondecreasing surrogate objective capturing product urgency, current inventory, and expected demand. We formulate an integer program and propose a column generation algorithm with efficient pricing heuristics. Preliminary experiments show the method finds near-optimal solutions with an average optimality gap of 0.02% in four minutes, outperforming a commercial solver in solution quality and runtime.

ABSTRACT. Retailers can select fulfillment centers (FCs) to fulfill orders, enabling inventory pooling and product expansion. However, fulfillment decisions are complex and companies need strategies addressing fulfillment complexity at scale. Regionalization divides the fulfillment network into a set of regions: orders that originate within a region are primarily fulfilled by FCs associated with the region. This structure simplifies the fulfillment network, and has recently provided retailers with significant gains in cost and speed. We propose an optimization model to design the regions while simultaneously assigning fulfillment centers to match each region's demand; a novel problem not studied previously. The model's main challenges include a non-linear objective function and contiguity constraints. We develop a local search heuristic for scalable solutions and efficient lower bounds for benchmarking solution quality. We demonstrate that appropriate parameters for region design significantly affect solution quality and increase demand fulfilled within service guarantees.

ABSTRACT. We consider a multi-period biomass-biofuel supply chain network planning problem in which we determine biomass collection and biofuel production locations and capacities in conjunction with biomass pricing that affects biomass supply quantity and locations. We develop a profit-maximizing tactical framework to optimize biomass and biofuel shipment quantities, inventory levels, and farm-specific price offers across periods, while accounting for reductions in conversion efficiency due to biomass degradation during storage. To solve the model efficiently, we propose a Benders Decomposition algorithm enhanced with surrogate constraints to improve upper bounds. The model is evaluated on both randomly generated instances to assess computational performance and a realistic case study for the U.S. Midwest using data integrated through a Geographic Information System. Numerical results demonstrate the effect of key parameters on network design and validate the effectiveness of the proposed model and algorithm.

ABSTRACT. We investigate a dynamic inspection problem for infrastructure systems where multiple vehicles are assigned to monitor components with heterogeneous and stochastic failure risks. The problem is modeled as a multi-stage stochastic mixed-integer program that jointly optimizes routing and scheduling decisions over time under endogenous uncertainty, where future failure probabilities depend on prior inspection actions. To address the computational challenges of solving large-scale instances, we develop a stochastic dual dynamic integer programming (SDDiP) algorithm that integrates dual approximation, integer state reduction, and sampling-based forward simulation to approximate cost-to-go functions. In parallel, we design a reinforcement learning framework that learns job clustering and routing policies by interacting with simulated system dynamics, where component failures depend on inspection decisions. Comparative studies highlight the trade-offs between exact stochastic optimization and data-driven learning approaches in terms of computational scalability and policy interpretability under different failure risk configurations, offering insights for intelligent infrastructure monitoring and maintenance.

ABSTRACT. We address dynamic scheduling of inbound trucks at a warehouse with known service times and uncertain arrival times. Arrival time distributions are unobserved but inferred from estimated times of arrival (ETAs). Motivated by collaboration with Poste Italiane, which operates one of Italy’s largest logistics networks, the objective is to minimize expected total waiting time. We develop a dual bound using information relaxations and an information penalty. We show how a sequence of theoretical reductions transforms the resulting stochastic dynamic program into a compact mixed integer program. Computational results indicate the penalized dual bound is nearly 10 percent stronger than a bound based on perfect information. We propose a one-step lookahead policy that adapts decisions as ETAs are updated. When arrival distributions are observable, the policy gap is below one percent. Relative to industry practice, the policy reduces expected waiting time by 29 percent and is fast enough for real-time use.

ABSTRACT. Strategic maritime fleet composition requires long-term commitments under significant uncertainty regarding demand, freight rates, port congestion, and regulatory costs. This paper proposes a strategic multi-stage stochastic integer program (MSSIP) that bridges the gap between strategic stochastic programming and high-fidelity tactical simulation through an iterative calibration process. At the strategic level, the MSSIP optimizes fleet composition, time-charter contracts, and lane allocations using a mean-Conditional Value at Risk (CVaR) objective to manage downside risk. The methodology utilizes bidirectional data flows where an external tactical simulation environment generates empirical estimates of operational feasibility and costs. These realized capacities, voyage costs, and demand fulfillment patterns are fed back to recalibrate the strategic model’s parameters. The framework is designed to support both profit-maximizing and cost-minimizing variants to suit different business contexts.

ABSTRACT. Amid growing interest in the integration of drones into maritime logistics, this paper addresses the drone scheduling problem in shore-to-ship delivery (DSP-SSD), which is both significant and challenging. We introduce a mixed-integer programming model with time discretization that incorporates drone-related constraints, moving targets, and the need for multiple drone trips. While commercial solvers can handle this model in small-scale scenarios, we propose a tailored branch-and-price-and-cut (BPC) algorithm for larger and more complex cases. This algorithm integrates a drone-specific backward labeling algorithm, cutting planes, and acceleration methods to boost its effectiveness. Experiments show that the BPC algorithm substantially outperforms the commercial solvers in terms of solution quality and computational efficiency and that the inclusion of acceleration strategies in the algorithm enhances its performance.

ABSTRACT. Intra-city express delivery is an emerging service that enables customers to send and receive packages within the same city in a short time (typically 6-8 hours). Its efficiency relies on a dedicated intra-city service network, where packages are consolidated at and transported between urban satellite stations. A key challenge in designing such service networks lies in the limited docking capacity for unloading/loading packages at satellite stations, which may lead to congestion and delays. To alleviate this bottleneck, logistics companies can reroute a subset of intra-city packages using the spare capacity of the inter-city service network. We formulate this problem as a consolidation-based integer program and propose a solution framework that combines state-of-the-art column generation and integer-programming-based local search techniques. A numerical study confirms the effectiveness of our solution framework, highlights the impact of docking capacity, and demonstrates the benefits of the synergy between intra- and inter-city service networks.

ABSTRACT. The surge in e-commerce has heightened the need for efficient last-mile delivery, prompting the exploration of drone-assisted logistics integrated with public transportation. This paper focuses on optimizing a Drone-Transit Coordinated Delivery Problem by formulating mathematical models based on time-space networks. We use a multi-variable generation (MVG) algorithm that iteratively generates time-space arcs to identify promising paths. To solve the problem, we first develop an exact algorithm combining a branch-and-bound framework with MVG, accelerated by cut generation and heuristic primal bounds; A heuristic variant that solves a compact integer program after MVG pre-processing is also employed. Experiments based on real-world data from Shenzhen demonstrate that our approach significantly outperforms benchmark methods. Sensitivity analysis quantifies the effects of key parameters, including transit network scale, vehicle scheduling, drone charging rates and endurance, and network topology. The algorithm and model exhibit scalability, efficiency, and robustness, making them suitable for real-world urban logistics applications.

ABSTRACT. Home healthcare (HHC) agencies face complex operational decisions when assigning nurses to geographically dispersed patients under uncertain, dynamic demand while preserving continuity of care. We study a multi-period HHC planning problem in which patients request long-term services, may cancel over time, and must be continuously served by the same nurse, subject to nurses’ working-hour and routing constraints. Decisions are made sequentially, and some patients may be rejected due to limited capacity. The objective is to maximize the expected number of service months. We model the problem as a sequential decision process and propose a solution that combines cluster-based sampling to manage the exponential assignment space with a value function approximation (VFA). The VFA captures both the current post-decision state and potential future states. Preliminary numerical experiments show that the proposed policy outperforms practical geography-based heuristics by up to 8%, highlighting the value of anticipating future system dynamics in HHC planning.

ABSTRACT. In this presentation, we present routing optimization for field artillery units where travel decisions must be coordinated with mission execution. We formulate a new vehicle routing problem with scheduling for artillery operations (VRPwSA) featuring a dual-graph structure, cooperative multi-vehicle engagements, limited onboard inventories, and heterogeneous ammunition types with distinct range–damage–duration trade-offs. To solve large instances, we develop a multi-agent reinforcement learning approach, modeling decisions in a three-phase cycle (move, select ammunition, fire) with transformer-based encoding, inter-agent communication, and dynamic embeddings. Experiments show the RL method—especially sampling-based decoding—consistently outperforms MILP baselines and generalizes to instances up to 10× larger with sub-second inference.

ABSTRACT. We study a learning-augmented prize-collecting vehicle routing problem (VRP) where customer demand is uncertain, context-dependent, and revealed only upon service. We formulate the problem as a contextual combinatorial multiarmed bandit and propose a Bayesian policy integrating contextual demand learning, Thompson sampling, and scenario-based stochastic routing. Demand follows a linear–Gaussian model with Normal–Inverse–Gamma priors, enabling closed-form posterior updates and information sharing across customers with similar contextual features. Routes are selected by solving scenario-based prize-collecting VRPs under posterior-predictive demand samples. We evaluate the policy in a semi-synthetic environment derived from a large-scale retail dataset with heterogeneous, empirically estimated demand noise. Results over a multi-day horizon show stable learning behavior across random seeds, with regret dynamics reflecting both demand uncertainty and contextual generalization. Overall, the framework provides an approach for integrating sequential learning and routing decisions under partial demand information in last-mile logistics systems with limited observability and operational constraints.

ABSTRACT. The practice of transit planning involves generating candidate lines through the current service plan, public engagement, and the consultants’ experiential knowledge. In other words, riders and seasoned planners know which lines may work well individually, based on factors that may or may not be captured by automated line-generation techniques. However, their main bottleneck is to identify a subset selection that works well as a system. Therefore, in this work we pivot from generating a rich set of candidate lines to generating a rich selection based on established principles of transit planning. This re-focus on leveraging the given set of candidate lines rather than generating it allows us to reach a practical scale, with hundreds of thousands of OD pairs.

ABSTRACT. Network-based systems are inherently interconnected, with the design and performance of subnetworks being interdependent. However, the decisions of self-interested operators may lead to suboptimal outcomes for users and the system as a whole. We address this challenge using a game-theoretical framework that integrates both non-cooperative and cooperative game theory. In the non-cooperative stage, we propose a network design game in which subnetwork decision-makers strategically design their local infrastructures. In the cooperative stage, co-investment with payoff-sharing mechanism is developed to enlarge collective benefits and fairly distribute them. Case studies on the Sioux Falls network and the Zurich–Winterthur public transport systems demonstrate that the proposed mechanisms significantly enhance long-term sustainability, social welfare, and economic efficiency. The proposed framework provides a foundation for improving interdependent networked systems by enabling strategic cooperation among self-interested operators.

ABSTRACT. Urban transportation has seen a significant shift in recent years, yet our understanding of transportation network-efficiency has not evolved with this change. Prevailing network-efficiency metrics are limited to assessing the efficiency of single origin-single destination trips. Such metrics miss the operational core of modern transportation services---the fulfillment of multiple demand locations per trip under diverse service modality and demand patterns. In this work, combining the well-known BHH theorem and large-scale modern road network and transportation data, we introduce a new network-efficiency metric, routing efficiency, that captures the efficiency of modern transportation services on urban networks. Using real road networks of 200 cities, we compute at-scale routing efficiencies under varying service modalities and demand distributions, and identify micro- and macro-structural features that explain cross-city efficiency variations. Case studies of New York’s subway evolution and Amazon's last-mile delivery service illustrate how routing efficiency can evaluate long-term infrastructure investment and emerging service designs.

ABSTRACT. The Terminal Maneuvering Area is where most air traffic interactions occur, making the coordination of arrivals and departures particularly challenging. Current Time-Based Management practices first assign fixed waypoint crossing times to satisfy separation constraints and then optimize individual aircraft trajectories. However, existing flow-level methods typically minimize airborne delay without considering the environmental impacts of delay distribution, which can lead to fuel-inefficient trajectories. This paper introduces a bi-objective, fuel-aware, stochastic TMA flow-level optimization framework that jointly minimizes delay and fuel consumption. The framework integrates three components: a data-driven fuel consumption model with endogenous uncertainty, a matheuristic combining linear programming and a genetic algorithm for sequencing and scheduling, and a waypoint-time-constrained optimal control model for trajectory generation. Applied to a full day of operations at SIN Airport, the approach reduces total fuel burn by 6.2% compared to a First-Come-First-Served baseline and by 5.5% relative to delay-only optimization, improving sustainability without sacrificing efficiency.

ABSTRACT. Aircraft trajectory planning relies on up-to-date weather information to minimize travel costs such as time and fuel burn. However, observations in today’s airspace—whether from weather balloons or en-route aircraft—are spatially and temporally sparse, so useful weather information to identify better routes is often wasted. To collect weather information under operational constraints, we study optimal sampling schedules with weather balloons and aircraft to maximize information gain in airspace networks. We introduce a graph representation of routes for a single origin–destination pair and model the wind-induced excess cost on each arc as independent Ornstein–Uhlenbeck processes. We then analyze two complementary settings: (i) arc sampling with weather balloons and (ii) path sampling with aircraft. For each setting, we characterize optimal periodic sampling schedules that minimize systemic steady-state uncertainty, and relate optimal sampling frequencies to network geometry and temporal dynamics. We validate the theoretical insights by simulation.

ABSTRACT. The sustainment of aircraft fleets requires spare-parts supply chains that balance high readiness with limited budgets across geographically dispersed facilities. This study develops a bi-objective, two-stage stochastic optimization model to support strategic inventory placement decisions for aircraft maintenance networks. The model jointly determines hub locations and pre-positioned inventory levels to minimize systemwide inventory costs while maintaining readiness, measured through unmet-demand penalties under uncertain demand. Using historical aircraft maintenance and part-order data, demand uncertainty is represented through scenario-based realizations derived from observed variability. Computational results show that introducing regional hubs can reduce total inventory through risk pooling with little to no detriment to readiness. A Pareto frontier illustrates cost–readiness trade-offs, identifying operating regions that maintain near-zero unmet demand with substantially lower inventory investment than the current baseline. The proposed framework provides actionable insights for maintenance-intensive logistics systems facing stochastic demand and spatial complexity.

ABSTRACT. We study an inventory replenishment problem that relies on a sustainable but delay-prone transport mode, complemented by the option to expedite in-transit shipments. Transport delays occur according to localized and time-varying delay risks. We develop an inventory control model that assists in determining the ordering and expediting decisions given these transport delays. We show that, under a no order crossing condition, the optimal decisions follow an adaptive base-stock policy driven by inventory and delay information. Otherwise, we propose an approximate policy that extends these base-stock structures and is applicable to realistic problem sizes while delivering a near-optimal performance. Numerical experiments demonstrate that an expediting opportunity improves responsiveness and reduces inventory costs. Moreover, expediting in-transit shipments increases the sustainable mode's usage compared to having an expediting option at the origin and without incurring additional costs. Overall, our findings highlight that a delay-mitigating expediting strategy supports sustainable replenishment shipments while remaining cost-competitive.

ABSTRACT. This paper develops exact solution methods for two general shipment consolidation problems: Long-haul shipment (L) problem and combined Long-haul and Short-haul shipment (LS) problem. Both problems involve large numbers of orders and nonlinear, piecewise cost functions, making existing heuristics produce optimality gaps of up to 9%. For the problem L, we propose a Bound-then-Repair (BR) algorithm that first derives tight lower and upper bounds using dynamic programming and iterative search, then embeds customized checking constraints within a branch-and-bound framework to efficiently reach optimality. For the problem LS, we design a Logic-Based Benders Decomposition (LBBD) that separates long-haul decisions from short-haul quantity optimization, supported by a newly constructed tight convex lower envelope and hybrid analytical cuts that markedly accelerate convergence. Our BR algorithm solves all instances in 2 seconds and reduces costs over the state-of-the-art heuristics by up to 16.39%, while our LBBD algorithm cuts costs by up to 15.97%.

ABSTRACT. Stochastic clearing theory has widespread applications in the context of supply chain and service operations management. In this presentation, we consider applications arising in the context of outbound delivery operations involving active efforts for supply optimization under shipment consolidation. We develop a stochastic dynamic model to consider explicit efforts and costs associated with temporal shipment consolidation of outbound deliveries. We formulate the problem via a backward stochastic dynamic programming approach. We prove that clearing-type policies are not necessarily optimal due to the interplay between inbound and outbound fixed costs.

ABSTRACT. We study the evacuation of dense crowds from a single origin to multiple destinations. We propose a multimodal strategy that integrates a Scheduled Response Mode (SRM) with planned routes and a centrally dispatched fleet, and an Incentivized Response Mode (IRM) that mobilizes nearby vehicles via subsidies. This multimodal mass evacuation (MME) problem coordinates SRM and IRM through the joint optimization of SRM routing and fleet deployment, IRM subsidy pricing, and evacuee assignment. We propose an triple decomposition and correction algorithm to address the computational challenge. The method integrates a three-stage decomposition component with a tailored correction procedure to refine solutions. A case study on mass transit railway disruption in Hong Kong shows that the multimodal evacuation strategy significantly reduces cost and improves evacuation efficiency over the single-mode benchmark. Extensive computational experiments further demonstrate that the proposed method substantially outperforms baselines in runtime, solution quality, convergence speed, and scalability.

ABSTRACT. We study a bilevel incentive-guided routing problem motivated by mobile food vendor operations in underserved regions. A policymaker allocates incentives to improve food accessibility, while vendors select profit-maximizing service routes, formulated as a Profitable Tour Problem. The resulting model combines continuous incentive decisions with NP-hard routing constraints. We use a logic-based Benders decomposition that separates accessibility-driven visit selection from routing feasibility, and employ an oracle-based verification procedure to ensure follower optimality. Computational experiments based on a real-world case study show that modest incentive budgets can induce disproportionately large accessibility gains, highlighting the effectiveness of targeted policy interventions.

ABSTRACT. Emergency evacuation and relief distribution are typically planned as separate operations, despite relying on the same transportation infrastructure and time-sensitive shelter operations. This separation often results in infeasible plans where evacuees arrive before supplies or relief convoys are blocked by evacuation traffic. This paper proposes a time-dependent, integrated network design framework that jointly coordinates evacuation and relief flows across shared, capacity-constrained corridors. A time-expanded formulation enforces synchronized arrivals, shared arc capacities, and non-free waiting through congestion-aware holding arcs, ensuring that clearance and provisioning decisions are mutually feasible. To solve the resulting large-scale mixed integer program, we develop a tailored Benders decomposition scheme that prioritizes time feasibility through connectivity repair and separates evacuation and relief routing into tractable subproblems. Computational experiments demonstrate that the integrated approach outperforms standalone models and scales to realistic GIS-based coastal networks, producing time-feasible and capacity-consistent response plans.

ABSTRACT. We study California’s Bottle Bill, a deposit-refund policy that charges consumers a deposit on beverage containers and refunds it upon return. Currently, the redemption value has remained stagnant at 5 cents for decades; the number of recycling centers has decreased, and recycling rates have been in decline. To address these issues, we “spin” the Bottle Bill by jointly optimizing policy levers and the reverse supply chain in a bilevel mixed-integer nonlinear program. At the policy level, we choose the redemption value, manufacturers’ fees, and consumers' distance policy to recycling centers. At the network level, we determine the number of recycling centers in each county. To solve this problem, we develop a continuous-approximation and monotonicity-based solution approach with optimality guarantees. Applied to California, our methodology identifies implementable policy options to achieve its 80% recycling target by jointly balancing the redemption value and the number of recycling centers to promote circularity.

ABSTRACT. We study outbound shipping decisions in a two-echelon reverse-logistics network operated by a third-party provider. Customers return online purchases to local return bars, which consolidate items and ship parcels to a central hub. The hub sorts and aggregates returns and then ships consolidated loads to merchants using less-than-truckload and full-truckload services. We model the joint return bar–hub system as a discounted Markov decision process and analyze its structural properties. At the return bars, we show that the optimal policy has a state-dependent threshold form on a reduced state space. At the hub, the large state space makes exact dynamic programming intractable, so we propose two simple shipment heuristics based on inventory and lateness thresholds and benchmark them against the optimal policy and other heuristics from the literature. We prove asymptotic optimality of these heuristics under varying parameter settings and demonstrate substantial cost savings on operational data from our industry partner.

ABSTRACT. Reliable facility planning depends on both disruption patterns and customer decisions under uncertainty. This work studies a setting with correlated failures and limited real-time information about facility availability. A mixed-integer linear program is developed that jointly chooses facility locations and customer visiting sequences. The model minimizes facility opening costs and expected travel and penalty costs under correlated disruptions and imperfect information. To scale to large instances, a Lagrangian-relaxation–based solution approach is proposed. It includes efficient algorithms to compute both a feasible upper bound and a dual-based lower bound, with much lower memory use than the full MILP. Numerical experiments demonstrate the effectiveness and scalability of the proposed framework.

ABSTRACT. We study the vehicle routing problem with speed optimization (VRP-SO), in which vehicle routes and continuous travel speeds are jointly optimized to minimize total operational costs. To solve the VRP-SO, we develop an exact price-and-enumerate algorithm. We first present a novel speed-optimized labeling algorithm for solving the pricing subproblem, which integrates closed-form expressions for optimal speeds, efficient partial dominance rules, and tailored label pruning acceleration techniques. These eliminate the need to solve convex optimization problems during label extension and significantly improve computational performance while maintaining exactness. Additionally, we introduce a route enumeration phase that generates a pool of high-quality, speed-optimized routes. Computational results show that the proposed algorithm effectively solves benchmark instances with up to 75 nodes within one hour, achieves an average of three-fold speedup in runtime compared to state-of-the-art methods, and successfully solves 45 additional instances where existing exact approaches fail to reach optimality.

ABSTRACT. Many transportation problems can be modelled as path problems, where a set of elements must be covered by a least-cost set of paths. Column generation is the method of choice for such problems. A large number of novel problem variants feature a nested subpath-path structure that classical column generation algorithms fail to exploit. In this work, we propose an exact, bucket-based pricing algorithm that is specifically designed for computational efficiency in subpath-path settings, avoiding ex-ante enumeration of non-dominated subpaths. Instead, the algorithm partitions the space of subpaths into so-called buckets and limits the search to at most one subpath per bucket, which are then concatenated to paths. When this fails, the bucket partitioning is iteratively refined. We show that this pricing algorithm solves the pricing problem to optimality in finite time, and demonstrate its effectiveness on a series of robust railway crew scheduling and balanced multi-period vehicle routing instances.

ABSTRACT. We propose an integrated optimization model for the air cargo service network scheduling problem, specifically targeting next-day delivery services. By synchronizing dedicated cargo fleet with passenger belly capacity, the model optimizes flight scheduling, fleet routing and cargo flow, incorporating through cargo connections to eliminate intermediate handling. We develop a tailored column-generation-based algorithm to solve this large-scale problem. Experiments using real-world industrial data demonstrate that our algorithm rapidly yields near-optimal solutions for small instances and consistently outperforms commercial solvers on medium and large datasets. Furthermore, the inclusion of through cargo connections is shown to significantly reduce operational costs while improving flight load factors and service levels.

ABSTRACT. Out-of-home delivery networks are rapidly expanding, offering a growing number of parcel lockers and parcel shops. While lockers provide 24/7 access, capacity is strictly limited. If all compatible compartments are occupied, the provider must redirect parcels to other facilities. In this work, we explicitly quantify this risk by framing lockers as probabilistic products. More specifically, we dynamically display delivery probabilities to each request. The provider can either report truthful probabilities or manipulate the offer set to guide customers' choices and maximize customer welfare. To investigate both strategies, we design a novel choice model and formulate a recursive equation for the truthful probabilities. In our solution methodology, we tackle the curses of dimensionality by training an approximation of the latter with a modified deep Q-learning algorithm and embed it in dynamic assortment optimization problems. Our preliminary results illustrate that actively managing demand generates a substantial improvement of 6.7% in customer welfare.

ABSTRACT. We study how on-demand delivery providers should expand and allocate fleet capacity when demand is endogenous to service quality. Motivated by recent geographic expansions by Amazon, Walmart, and rapid-delivery startups, we model a provider operating across heterogeneous regions, some mature and others emerging with “cold start” demand. We develop a finite-horizon dynamic model where managers choose both tactical fleet allocation and strategic expansion. Using a BHH-based capacity proxy and a retention–acquisition demand process, we derive three main insights: (i) in emerging regions with convex growth, the optimal policy deliberately reallocates vehicles away from mature “cash cow” regions to jump-start new markets, (ii) there exists a viability threshold below which entering an emerging market is not optimal, and (iii) the optimal expansion policy takes a demand-dependent base-stock form and, in the steady-state limit, the optimal fleet size satisfies a square-root capacity scaling law that links geography and cost parameters.

ABSTRACT. Activity-based models (ABMs) offer behaviorally rich representations of travel demand but remain underutilized due to high data and computational requirements. This study proposes an optimization-based framework that generates activity-based travel demand directly from existing trip-based model outputs. By integrating legacy household travel survey data, synthetic population generation, and an integer programming–based spatial allocation model, the framework reconstructs disaggregate activity–travel patterns while remaining consistent with aggregate origin–destination flows. Individual activity chains are expanded across a synthesized population, and activity locations are assigned by minimizing discrepancies between reconstructed and observed trip-based OD counts under behavioral feasibility constraints. A case study for the Kentuckiana Regional Planning & Development Agency demonstrates the effectiveness of the approach. The proposed framework provides a practical pathway for enriching trip-based forecasts with activity-based detail without requiring full ABM implementation.

ABSTRACT. Urban e-commerce is increasing parcel flows while transit capacity is underused off-peak. We study scheduling of modular vehicles for passenger-freight co-modal operations with end-to-end freight delivery. Modular units can dynamically couple/decouple at defferent times and locations, share capacity, and enable in-vehicle transfers so passengers and freight can switch lines without re-boarding. On a system of a bus network and residential locations, we jointly decide fleet size, each unit’s scheduling, as well as the demand assignment, minimizing operating costs and penalties for unserved freight with time windows. We formulate a set-covering model and develop a column generation algorithm with a tailored label-setting method. To obtain high-quality integer solutions, a diving procedure is designed. Computational results show our algorithm finds better integer solutions within 20 seconds than Gurobi’s best incumbent after 1 hour, and the co-modal operations reduce objective values by 10-27% and fleet size by up to 27% versus separated transport.

ABSTRACT. Long school bus rides adversely affect student performance and well-being. In rural areas, lengthy rides incentivize parents to drive their children instead, exacerbating traffic congestion near schools and creating underutilized buses with higher per-rider costs. This paper addresses rural school bus routing and scheduling, a challenging problem due to mixed loading and irregular road networks. We formalize a model minimizing total student ride time and develop a road network-aware cluster-then-route heuristic that produces high-quality solutions, as demonstrated through extensive ablation studies. In real-world case studies, our approach reduces student bus ride times by 37-39% compared to status quo solutions. By making school buses more attractive, we improve bus utilization by 17-19% and reduce congestion-inducing car trips by 12-17%. Given that many rural districts share these operational characteristics, our approach offers broad applicability for addressing transportation-related barriers to student performance and well-being.

ABSTRACT. We propose a continuous approximation framework for robust estimation of a school’s minimum required bus fleet size when student locations are not known with certainty. We focus on the non-trivial case in which the maximum allowable route duration (and not the physical bus capacity) is the limiting factor. Prior experience with school administrators suggests that the ability to quickly evaluate ‘what-if’ scenarios in real time is critical during the planning process; therefore, our approach requires minimal computation after one-time setup. From a methodological perspective, we demonstrate how applying continuous approximation techniques to school bus routing entails unique challenges not found in typical TSP-type problems. Specifically, we show that the classical Beardwood-Halton-Hammersley Theorem does not produce appropriate approximations for a broad class of covering route problems, in which vehicles visit a set of planar facilities (e.g., school bus stops, parcel lockers) that collectively ‘cover’ demand points (e.g., students' homes, delivery recipients).

ABSTRACT. This paper studies a tactical–strategic planning problem faced by a motor carrier operating a fixed middle-mile hub network while serving clients with differing sensitivity to greenhouse gas emissions. Commodities purchased by retailers from multiple vendors must be transported from origins to destinations through a hub-based network, with all-or-nothing service decisions. The carrier must decide which shipments to accept, how to route accepted shipments, where to install electric vehicle charging infrastructure, and how many diesel and electric trucks to deploy on each lane, while accounting for client-specific emissions targets. Emissions are measured using the Global Logistics Emissions Council (GLEC) well-to-wheel framework and allocated proportionally among shippers. We formulate this problem as a mixed-integer linear program that integrates fleet composition, infrastructure placement, routing, and emissions considerations. Computational experiments on realistic network instances evaluate the tractability of the model and examine how emissions sensitivity affects optimal network design and fleet electrification decisions.

ABSTRACT. Independent planning by carriers in middle-mile delivery often leads to substantial unused capacity, resulting in higher costs, congestion, and CO₂ emissions. This work investigates the potential savings enabled by horizontal collaboration among regional and national carriers operating on overlapping networks. We formulate a time-dependent service network design problem and model it using an integer linear programming approach, comparing solutions from independent carrier planning with a fully centralized collaborative benchmark. Using 390 instances generated on realistic European road networks with up to 100 commodities, we observe average cost savings of 9% through collaboration, increasing to nearly 14% for larger instances. These findings indicate that decentralized collaboration mechanisms based on multi-phase combinatorial auctions, which allow carriers to exchange bundles of commodities while preserving planning autonomy, offer substantial potential for cost reduction and improved utilization in middle-mile delivery.

ABSTRACT. The use of underground tunnels for freight transportation is gaining attention in both industry and academia as a way to efficiently move packages between distribution centers. We propose a new problem for tunnel delivery in the middle mile, the middle mile tunnel dispatching problem (MMTDP). We model this problem as an integer program with the objective of minimizing the average delivery time for orders. In the MMTDP, vehicles carry pallets of goods between distribution centers in different cities. We evaluate the complexity of the MMTDP and provide a set of computational results to demonstrate the tunnel system's performance and the impact of different design parameters. We find that low demand rates, short distances between distribution centers, and faster tunnel vehicles increase the benefit of using delivery tunnels compared to traditional truck delivery.

ABSTRACT. In some military and humanitarian logistics applications, area clearing and path planning decisions need to be made together. For example, an army wants to transport troops, munitions, or equipment through a minefield. It is indispensable to coordinate the mine clearance decision and the subsequent transportation decision. As another example, since it is impossible to clear all affected areas in a short period of time after a hurricane, determining which areas to clear and which paths to use for transporting emergency supplies is crucial for successful humanitarian operations. Motivated by these logistics applications, we discretize a relevant region into a grid and propose an optimization model for joint area clearing and path planning decisions. Since clearing an area will reduce the risk of traversing the area, our objective is to minimize a total transportation risk. Two case studies are provided to illustrate the application and effectiveness of our optimization model.

ABSTRACT. Humanitarian organizations face a high degree of uncertainty in delivering aid to disaster-affected regions, particularly during the middle-mile where reliance on commercial air carriers constrains efficiency through limited reservable capacity, variable aircraft configurations, and latent carrier acceptance probabilities. To address these challenges, this work develops a predict-then-optimize framework that models the interaction between commercial air carriers and a humanitarian freight forwarder. Leveraging historical operational data, we estimate commercial carrier acceptance probabilities and aircraft configuration availability to drive a two-stage stochastic program. The model identifies optimal request strategies by balancing upfront carrier choice against the downstream recourse costs of operational outcomes, including rejected requests, rebooking with alternative carriers, and delays that arise from incompatible aircraft assignments. Overall, the framework aims to reduce expected costs, estimate outcomes, and handle uncertainty in air cargo transportation.

ABSTRACT. Wildfires require prioritization decisions to suppress fires over disperse areas with limited resources. This paper develops a predictive and prescriptive framework to jointly optimize crew assignments and wildfire suppression. The problem features a combinatorial resource allocation structure with endogenous wildfire demand and nonlinear fire dynamics. We formulate an integer optimization model that integrates a time–space–rest network for crew assignments with a time–state network for wildfire dynamics, linked by consistency constraints. We develop a branch-and-price-and-cut algorithm based on a two-sided column generation scheme that iteratively generates crew routes and suppression plans, a new family of cuts exploiting the knapsack structure of the linking constraints, and novel branching rules accommodating nonlinear wildfire dynamics. We propose a data-driven approach with double machine learning to estimate wildfire spread while controlling for endogeneity. Experiments show the algorithm scales to otherwise intractable instances and reduces the burned area by 12.7% relative to a greedy baseline.

ABSTRACT. Online grocery services require customers to be present at delivery, making the design of time window assortments, including choices between many or few, long or short, overlapping or non-overlapping options, an important task. Although interest in time window assortment design is growing, the value of overlapping windows remains insufficiently understood. Building on an established model that links time window assortments to demand and delivery efficiency, we extend the framework to capture demand variability across windows. Specifically, we examine how overlapping time windows can mitigate inefficiencies caused by demand peaks. We analytically derive ex-post necessary and sufficient conditions under which overlapping time windows reduce delivery costs relative to consecutive ones, and complement this with an ex-ante Monte Carlo simulation. Our findings advance understanding of how overlapping time windows influence delivery efficiency and identify the operational conditions and demand patterns under which their use is most beneficial.

ABSTRACT. Meeting delivery dates remains a major challenge for companies, as end-to-end lead times often fluctuate significantly. While many studies focus on reducing external delays (e.g., supplier issues), firms also face internal variability—such as processing or loading delays—that drives tardiness. Real-time digital monitoring promises greater visibility into job progress, enabling timely interventions, yet its economic value remains largely unquantified and uncertainty about benefits often deters adoption. This work quantifies the value of digital monitoring that flags lagging jobs to trigger expediting actions. We develop an influence-diagram framework integrating Bayesian inference with queueing dynamics to evaluate monitoring regimes. Using a intralogistics case study, we show that optimally tuned digital monitoring can substantially reduce operational costs relative to manual monitoring and no-monitoring scenarios. The results clarify when digital monitoring outperforms manual practices and help managers justify investments in real-time visibility by linking information accuracy, timeliness, reliability, and expediting capacity to measurable cost reductions.

ABSTRACT. Attended home delivery in grocery logistics often require customers to select delivery time slots at ordering. While incentives can influence these choices, the resulting selections greatly affect overall routing efficiency. We investigate the Time Slot Adjustment Problem (TSAP): the dynamic and interactive adjusting of customer delivery slots between the cut-off time and the delivery period. Our goal is to improve routing efficiency, while minimizing customer disruption. The TSAP raises two core questions: estimating the impact of given slot changes, and identifying which customers should be approached. We first focus on quantifying the number of changes required to obtain a target cost reduction, and developed a dynamic programming-based procedure that computes an upper bound on this number. Preliminary results over hundreds of routes indicate that only a fraction of customers might need adjustments, and establish this as a promising avenue for both operational improvements and further research.

ABSTRACT. In contract-based periodic deliveries to urban businesses (e.g., hotels, restaurants, and cafés), customers choose service offerings by trading off visit frequency, visit timing, promised time-window length, and price. To capture endogenous demand and explicit customer choice behavior, we propose a choice-based periodic vehicle routing with pricing (CB-PVRP). The problem is formulated as a two-stage stochastic mixed-integer program and approximated using sample average approximation (SAA). To represent supply–demand interactions, we introduce two linear customer best-response sets that are computationally more efficient than bilevel programming and KKT-based single-level reformulation. We develop a tailored logic-based Benders decomposition (LBBD) algorithm to solve the SAA problem and enhance it with several acceleration techniques. Computational results show that the proposed LBBD outperforms the state-of-the-art MIP solver in terms of solution quality and speed. A case study demonstrates the managerial benefits of the proposed CB-PVRP in terms of service profitability.

ABSTRACT. This work uses a stochastic two-stage optimization approach to study a generic vehicle routing problem (VRP) that is subject to vehicle breakdown, with the objective of minimizing the robust makespan. The problem of interest reflects a number of challenges present in general logistics practices. For instance, this problem contains a (naturally imposed) hierarchical decision-making structure, where the initial decisions must be made under uncertainty, an inherent characteristic of logistics operations, which generally transpire in stochastic environments. The preliminary results indicate an objective improvement of the proposed stochastic optimization approach when compared to a deterministic, rule-based solution. Current work focuses on devising a general stochastic initial solution optimality proof. While numerous additional VRP properties and disruption types could be incorporated in future work, another valuable research direction is to gain insight into (relationships of) static VRP components that contribute to a more general notion of system adaptability.

ABSTRACT. In this paper, we study a variant of the Vehicle Routing Problem with Flexible Deliveries (VRPFD) that incorporates stochastic travel times and soft time windows, referred to as the VRPFD-STT. In this setting, customers may be served at one of their preferred locations before or after their time window, which incurs an expected earliness or lateness penalty. Moreover, drivers may work beyond their regular working hours, in which case the expected overtime is also taken into account. The objective of the VRPFD-STT is to minimize the total expected cost, consisting of operational costs and service costs. The proposed problem belongs to the class of NP-hard problems, and we develop a metaheuristic solution approach based on Adaptive Large Neighborhood Search to address its computational complexity. Results indicate that the proposed approach is effective in obtaining high-quality solutions not only for the VRPFD-STT but also for the VRP with Roaming Delivery Locations.

ABSTRACT. Platforms with informational advantages can influence agent behavior in dynamic environments by strategically revealing information about an evolving system state. We study this problem in a dynamic traffic routing game with latent network states, where agents repeatedly choose routes based on public signals and update beliefs through learning. These equilibrium routing decisions shape both current congestion and the future evolution of the network. A central challenge is that the platform does not observe agents’ beliefs, which evolve endogenously from realized costs and past signals. We formulate the platform’s dynamic information design problem as a partially observable Markov decision process (POMDP) and show that, under mild conditions, the platform can asymptotically infer agents’ beliefs. Leveraging this result, we propose a two-stage learning algorithm that substantially reduces computational cost relative to direct POMDP methods while preserving efficient long-run performance.

ABSTRACT. Credit-based congestion pricing (CBCP) and discount-based congestion pricing (DBCP), which respectively allot travel credits and toll discounts to subsidize low-income users’ access to tolled freeway lanes, have emerged as promising policies for mitigating traffic congestion without exacerbating societal inequities. However, since the deployment of CBCP and DBCP policies in real-world traffic is nascent, the relative merits of each policy class remain unclear. Our work studies and compares the efficacy of deploying CBCP and DBCP policies to reduce travel cost and collect toll revenues. We present conditions, e.g., on the traffic network and user population, under which DBCP policies provably outperform CBCP policies in minimizing a given societal cost. Moreover, we present examples in which CBCP outperforms DBCP when the aforementioned conditions are violated. Finally, we validate our theoretical contributions on a case study of the 101 Express Lanes Project, a pilot CBCP program implemented in the San Francisco Bay Area.

ABSTRACT. Congestion pricing has emerged as an effective tool for mitigating traffic congestion, yet implementing welfare or revenue-optimal dynamic tolls is often impractical, with most real-world congestion pricing deployments, including New York’s recent program, relying on significantly simpler, often static, tolls. This gap between theoretically optimal and practical congestion pricing deployments motivates an evaluation of the performance loss in revenue and welfare when real-world systems use static rather than optimal dynamic pricing. We address this gap by studying two canonical frameworks that capture practical congestion pricing implementations: Vickrey’s bottleneck model and its city-scale extension based on the Macroscopic Fundamental Diagram. Our theoretical and numerical results demonstrate that simple static tolls achieve robust performance relative to their optimal dynamic counterparts on both revenue and welfare metrics, underscoring their practical effectiveness. A further contribution is the closed-form derivation of revenue-optimal static and dynamic tolls, which has received limited attention in prior work.

ABSTRACT. Shared micromobility systems (SMS) support transition to sustainable urban mobility. We present mathematical and computational framework for optimizing SMS network design by aligning capacity decisions with customer demand. We formulate mixed-integer non-convex optimization model to determine station locations, capacities, and sizes of the micromobility and rebalancing fleets. Our model incorporates customer preferences while approximating key operational and rebalancing dynamics. We develop spatial decomposition heuristic that separates model into station-specific problems and links them through iterative bilevel updates. Experiments with real-world data show that our approach delivers near-optimal solutions while cutting computational time by two orders of magnitude. Compared to current practice, our solutions serve up to 14% more demand, reduce rebalancing costs by up to 17%, and increase profits by up to 82%. Overall, our approach yields annual gains of 6–40 million for the SMS operator and cuts the daily environmental footprint by over 100 metric tons of {\rm CO}_2.

ABSTRACT. Many transportation systems, including public transit and urban parcel delivery, rely on vehicles operating on fixed lines. We study routing in line networks with handling times, where loading and unloading require per-unit time. Consequently, vehicle dwell times at stops are endogenously determined by routed volume, yet they also shape routing by affecting cycle lengths, service frequencies, and effective capacities.

We formulate the problem as a bilinear capacity–flow assignment and use optimality conditions to yield a non-separable, non-convex multi-commodity flow problem. We derive convex relaxations that compute near-optimal routings and provide dual gaps around 10% on congested systems. To certify global optimality, we develop a custom spatial branch-and-cut using McCormick relaxations, bound tightening, and tailored cuts. Computational experiments show that ignoring handling can produce infeasible routings, while modelling endogenous dwell reduces average travel time by up to 40% on large real-world instances.

ABSTRACT. The line planning problem in transit networks is a classical network design problem that is very computationally challenging to solve. We consider the related problem of large-scale line planning with heterogeneous vehicles fleets, i.e., those with different capacities and correspondingly operating costs. To address the inherent scalability challenges of the problem, we propose a novel Mixed-Integer Linear program and companion Column Generation algorithm. The approach implicitly constructs high-quality lines with appropriately chosen vehicle types. The efficiency of our approach is empirically validated through a numerical analysis conducted on the Boston network to showcase the value of the line planning with an optimal fleet mix.

ABSTRACT. Consolidation-based freight carriers operate large-scale hub-and-spoke networks and face tactical planning decisions that can be supported by the Scheduled Service Network Design Problem (SSNDP), a challenging optimization problem that quickly becomes intractable as the number of shipments grows. This paper proposes the Aggregated SSNDP (A-SSNDP), a relaxation of the SSNDP that leverages commodity aggregation, a scarcely studied technique in the literature. By partitioning and aggregating commodities in meaningful ways, the resulting model is significantly smaller while providing valid dual bounds for large-scale instances that are otherwise intractable. To strengthen the relaxation, three families of valid inequalities are introduced, along with a repair mechanism to recover feasible primal solutions. Computational experiments on realistic hub-and-spoke networks with up to 10,000 commodities demonstrate substantial reductions in model size and solution time, while delivering high-quality bounds and feasible solutions where the original SSNDP fails.

ABSTRACT. Packages are routed through the middle-mile network according to a flow plan and moved by trailers specified by a load plan. The outbound load planning problem determines how many loads to operate from a terminal to downstream terminals in the network. In large carrier networks, commodities may be routed onto alternate flow paths at intermediate hubs. While Ojha et al. (2024) jointly optimize primary and alternate flow paths to minimize transportation and sorting costs, this work focuses on a robust variant of the outbound load planning problem. We formulate this problem as a two-stage robust optimization model (ROLPP) with relatively complete recourse. We develop an exact column-and-constraint generation algorithm and scenario-generation heuristics for ROLPP. Computational results shows that the proposed algorithms can solve real instances to small optimality gaps; these heuristics exploit problem structure to yield tight lower bounds to the master problem and, produce high-quality warm-starts for the subproblem.

ABSTRACT. E-commerce expansion has intensified parcel distribution volatility, forcing logistics operators to deploy oversized fleets that generate substantial residual capacity. While the Physical Internet vision advocates dynamic reconfiguration, tactical planning remains largely static, disconnecting first-stage allocation from second-stage corrective mechanisms. This research proposes a two-stage stochastic optimization model that explicitly treats residual capacity as a redistributable asset within a hyperconnected parcel network. The framework integrates anticipative vehicle allocation, inter-hub pooling, and corrective subcontracting under demand uncertainty and green-coverage constraints. A representative case study from our industrial partner (La Poste) demonstrates that dynamic pooling reduces total subcontracting by 30% and operational stress peaks by 55%, with only 3 inter-hub transfers, compared to a rigid baseline. This confirms that modest rebalancing flexibility can effectively transform idle resources into a cost-effective and sustainable resilience buffer, advancing tactical planning practices in modern logistics networks.

ABSTRACT. Uncrewed aerial vehicles are widely used to monitor uncertain environments, but effective planning must balance coverage of known hazards with exploration of areas where hazards may be undiscovered. We present an integrated bi-level framework that connects global route optimization with local path planning. The upper level enhances spatial balance by augmenting a conventional routing solution with pseudo-nodes placed through an edge-based centroidal Voronoi strategy. The lower level transforms each route segment into a smooth B-spline trajectory that uses remaining endurance for targeted exploration guided by information value. A proportional allocation rule distributes exploration budgets according to the spatial influence of each segment. Simulated experiments in disaster environments show that the approach increases route coverage and improves hazard discovery compared with standard routing and baseline path planners. The framework is computationally efficient and suitable for real-world deployment in dynamic monitoring tasks.

ABSTRACT. A critical aspect of natural disaster response is providing timely guidance for evacuation and routing to at-risk individuals. Disaster management agencies often direct individuals to the nearest shelters, underutilizing information about individual needs and shelter locations and capacities. We propose a decision support system that integrates (a) simulation data from state-of-the-art physics-based models, (b) collection of data on individual-level needs and guidance dissemination via an LLM-based chatbot, (c) prediction of regional demand for critical resources, and (d) stochastic optimization for shelter assignment. Our proposed approach is novel because it leverages modern chatbot capabilities to aggregate information and provide centralized guidance to at-risk populations, while accounting for spatio-temporal complexities and the heterogeneous impacts of flooding across diverse populations. Using simulation data from a historic flooding event in Florida, we demonstrate that our approach significantly outperforms the currently adopted nearest-shelter assignment strategy.

ABSTRACT. This research proposes the development of a predictive and data-driven framework for Resilient Route Mapping, a next-generation system that integrates flood modeling, transportation vulnerability assessment, and freight logistics to ensure the continuity of critical infrastructure and supply chains under extreme weather conditions and a changing climate. The project combines high-resolution hydrodynamic simulations with routing optimization and economic impact modeling to quantify how present-day and future flood hazards disrupt road networks, freight delivery, commodity pricing, and greenhouse gas emissions through congestion, detours, and infrastructure degradation. The resulting platform identifies flood-resilient routes, evaluates infrastructure investment priorities across future climate scenarios, and estimates the monetary, operational, and emissions-reduction benefits of proactive resilience measures. Project outcomes enable federal, state, and local transportation agencies to anticipate climate-driven flood disruptions, safeguard emergency access, stabilize freight operations and contracts, and support the transition to a lower-carbon, climate-resilient transportation and logistics system.

ABSTRACT. Last-mile delivery is one of the most expensive parts of the supply chain, and parking scarcity often creates challenges for route implementation. With smart parking technology, companies can incorporate parking availability information into their route planning to provide robust solutions. This work incorporates parking availability information into route-planning. We formalize the routing problem as a Mixed Integer Program to optimize the service order of customers and parking spot selection with corresponding on-foot delivery. As a baseline, we fix the service order using a Traveling Salesperson Problem (TSP) solution based on driving times and only optimize parking spot selection and on-foot delivery service. Experiments show that also optimizing for the service order reduces tour time by about 10% and often makes changes at the beginning of the route. This research helps delivery providers improve efficiency and profitability, while guiding cities on the benefits and tradeoffs of investing in smart parking systems.

ABSTRACT. Same-day delivery faces the challenge of dynamic customer arrivals that need to be efficiently integrated into routing plans. If the customer can be served in the current route, both packages need to be on the vehicle which requires a preemptive return to the depot. Preemptive depot returns are often effective in high-density customer settings where parking also poses a challenge. When parking is difficult, returning for the package may provide additional advantages as the driver can park once to serve both customers on-foot. To understand when preemptive depot returns are cost-effective, we first introduce the Stochastic Parking Problem with Consolidation (SPP-C) and characterize structure for optimal parking and walking decisions to deliver to one or more customers on a line. Then, we identify scenarios where preemptive depot returns are cost effective. We provide insight into how operational costs and parking availability influence optimal operational decisions for same-day delivery.

ABSTRACT. This paper considers a disrupted infrastructure network where the repair crew knows the locations of service outages but not the locations of actual damage. Our goal is to determine a route for a single crew to visit and repair damage to restore service with minimum negative impact. We call this problem the Traveling Repairman Network Restoration Problem (TRNRP). Considering the dynamic nature of decisions due to information revelation on the status of the nodes, we model this problem as a finite-horizon Markov decision process. Our solution approach uses value approximation based on reinforcement learning, which is strengthened by structural results that identify a set of suboptimal moves. In addition, we propose state aggregation methods to reduce the size of the state space. We perform extensive computational studies to characterize the performance of our solution methods under different parameter settings and to compare them with benchmark solution approaches.

ABSTRACT. Dantzig-Wolfe and Arc-Flow reformulations are two of the most widely used approaches for solving large-scale integer optimization problems. While these reformulations are known to be related, formal results on their connection are limited and the two approaches are typically studied in isolation. Arc-Flow reformulations, in particular, can be derived from dynamic programming models, decision diagrams, or other problem-specific constructions, which makes a generic comparison with Dantzig-Wolfe nontrivial. This study clarifies the theoretical connections and computational trade-offs between the two reformulations. We further make their relationship explicit in settings with cutting planes and subproblem relaxations, two core techniques of modern exact methods. To empirically test our insights, we provide an experimental comparison on the Vehicle Routing Problem with Time Windows, highlighting which instance and methodological features make each approach preferable. Our results provide theoretical and empirical insights, and can help practitioners and researchers design more effective optimization methods for applications.

ABSTRACT. Neural policies trained with reinforcement learning have emerged as fast heuristics for routing problems, but handling non-monotone constraints such as time windows remains challenging. In this work, we study how different feasibility enforcement mechanisms affect the performance of neural policies on the Traveling Salesman Problem with Time Windows (TSPTW). We build on the attention-based framework of Kool et al. and consider a family of state-dependent action restrictions with increasing levels of lookahead, ranging from myopic feasibility checks to rule-based mechanisms, and relaxed decision diagram–based masking. Through a systematic empirical evaluation, we show that how feasibility is enforced during sequential decision making significantly affects both feasibility and solution quality. In particular, purely myopic feasibility pruning can bias the search toward poor solutions, while lightweight, problem-informed mechanisms achieve a more favorable balance between feasibility, solution quality, and computational cost. These results provide insights into the behavior of neural policies under non-monotone constraints.

ABSTRACT. We propose NEO-LRP, a neural-embedded mixed-integer optimization framework for the Capacitated Location-Routing Problem (CLRP). NEO-LRP approximates the routing cost induced by each opened depot using a learned, permutation-invariant neural surrogate based on the Deep Sets architecture. After a one-time offline training stage on independently generated VRP instances (12 hours), an off-the-shelf MIP solver produces depot-opening and customer-allocation decisions without solving routing subproblems during optimization. NEO-LRP then constructs feasible routes by solving a CVRP for each opened depot with a fast heuristic. Across four standard benchmark suites (with instances up to 600 customers), NEO-LRP achieves median gaps of 1.2% to 7.1% relative to best-known solutions while delivering 3x to 120x speedups over state-of-the-art heuristics. On the largest benchmark, more than 75% of instances are within 2% of best-known solutions. At the 600-customer scale, NEO-LRP attains a 1% median gap in 256 seconds, compared to over 15000 seconds for leading heuristics.

ABSTRACT. Curb space is an essential urban infrastructure that supports diverse transportation activities, yet its value of high accessibility makes this infrastructure vulnerable to misuse. This study introduces a game-theoretic framework for curb space management that models the strategic interactions among a curb operator, a regulator, and heterogeneous users within a multi-leader multi-follower game. By distinguishing between compliant and selfish users, the model jointly optimizes pricing, enforcement, and space allocation to balance revenue generation with legal compliance. The problem is formulated as a multi-objective bi-level program and solved through a duality-based single-level reformulation, linearization techniques, and the epsilon-constraint method to uncover how user behavior and enforcement shape policy outcomes. A robust extension incorporates behavioral uncertainty to evaluate worst-case policy performance. Experimental results highlight stakeholder interdependencies, revealing trade-offs between operator profit, enforcement intensity, and user behavior in shaping optimal curb management strategies.

ABSTRACT. Research on intersection operation with automated vehicles (AVs) largely falls into two paradigms: signal-based and reservation-based (signal-free) control. Yet neither paradigm achieves a practical balance between flexibility and tractability. To fill this gap, this paper proposes a new operational primitive for automated intersections, called service modes, defined as cyclic, conflict-free vehicle passages that efficiently utilize the intersection’s time–space resources. Service modes are constructed using a capacity-oriented principle that aligns their cycle-average service rates with the geometry-induced capacity frontier. Once a diverse and representative service mode library is available, intersection operation reduces to selecting among these modes according to observed traffic conditions. In this way, service modes cast automated intersection control into a coherent and structured framework with provable stability guarantees. By combining the structural simplicity of signal-based control with the flexibility afforded by fine-grained AV motion, service modes provide an interpretable and computationally efficient solution for automated intersection management.

ABSTRACT. Road crashes are one of the leading causes of mortality worldwide. We design, assess, and deploy real-time targeted alerts to nudge drivers toward safer behaviors, using an end-to-end approach spanning descriptive, predictive, and prescriptive analytics. Partnering with Waze, we assemble a large-scale global dataset, and build a deep learning model to build a road safety indicator at a granular spatio-temporal level. We use it to design proactive, targeted alerts on Waze's navigation platform. We conduct a large-scale, global randomized field experiment to evaluate the impact of these alerts. Our results show (i) a statistically significant decrease in average speeds and high-speed rates, (ii) a fatigue effect that can be attenuated by parsimonious nudges, (iii) externalities of treated drivers on other drivers, and (iv) heterogeneous responses based on road and driver features. The positive results from our experiment led to the global deployment of the targeted alert system.

ABSTRACT. Gig economy platforms such as Uber and Lyft operate in highly dynamic and competitive markets where drivers often receive job offers from multiple platforms. A key challenge for these platforms is to determine optimal compensation levels, taking into account that drivers’ valuation of a request may be highly heterogeneous, while competition status also influences a driver’s valuation of a request, but a platform does not have full knowledge of the competitor actions. In this paper, we study how a ride-hailing platform can make adaptive decisions on driver pay based on small samples and real-time reactions from drivers. We propose a mechanism where the platform sequentially makes job offers with varying pay levels, updates these offers based on drivers’ real-time decisions, and uses observed rejections to infer whether external competition. Our approach provides a structured way to optimize driver pay in uncertain and competitive environments.

ABSTRACT. Important pricing problems in centralized matching markets—such as carpooling matching, food delivery order batching, and less-than-truckload (LTL) freight shipment consolidation—often exhibit a bilevel structure. At the upper level, the platform sets prices for different types of demand. The lower level then matches converted demand to minimize operational costs. Motivated by these applications, we study the optimal value function of a linear programming model, originally proposed by Aouad and Sarıtaç (2022) for cost-minimizing dynamic stochastic matching under limited time. We demonstrate that this cost function is weakly concave if the optimal unmatch rate is non-zero, and identify conditions regarding matching efficiencies that guarantee this property. Building on these theoretical insights, we develop a Minorize-Maximization algorithm for the pricing problem that requires little stepsize tuning, and demonstrate substantial performance improvements over projected gradient-based methods on a real-world ridesharing dataset. This makes it a compelling choice for solving such pricing problems in practice.

ABSTRACT. We present an integrated simulation–optimization framework for the design and evaluation of ride-pooling services. The framework tightly couples a discrete mode choice model with a high-resolution operational simulator FleetPy through an iterative feedback mechanism, allowing demand, user preferences, and service performance to co-evolve until convergence. An outer optimization layer explores service design configurations such as fleet size, pricing, and operational constraints. The framework is evaluated through a real-world case study in Jerusalem, calibrated using multiple data sources. Preliminary numerical results highlight key trade-offs between user adoption, operational efficiency, and service quality, and reveal increasing returns to scale in fleet capacity alongside potential substitution effects from other transport modes.