Institute of Transportation Studies Berkeley faculty and student affiliates travelled to the 2025 INFORMS Annual Meeting at the Georgia World Congress Center and the Omni Atlanta Hotel at Centennial Park in Atlanta, Georgia, joining more than 6,000 INFORMS members, students, prospective employers and employees, and academic and industry experts to share ways O.R. and analytics are fueling Smarter Decisions for a Better World.
ITS Berkeley highlights:
Congratulations to Chiwei Yan, Assistant Professor at UC Berkeley IEOR, was named Runner-Up in the INFORMS Case Competition for his project, “Calyber: A Ridesharing Game.” Developed and refined with the support of UC Berkeley IEOR students over the years, Calyber immerses players in the challenges of ridesharing logistics.
Calyber: A Ridesharing Game
Chiwei Yan, University of California Berkeley, Yifan Shen, Julia Yan
Motivated by the persistent challenges that ridesharing platforms face in offering shared rides (a service that pools riders into a single vehicle), this case offers a hands-on and engaging experience in developing real-time analytics solutions. Using a rich historical ridesharing dataset, students design and test policies for quoting prices and matching riders, competing for top performance on a test set. By completing this case, students gain a practical understanding of stochastic dynamic optimization, a concept often difficult to teach through traditional instruction. 2 - Reinforcement Learning for Clinical Decision Support
Operational Optimization for Delivery Logistics in Urban Air Mobility and Commercial Aviation
Invited Session
Air Transportation Section
The UAM Fleet Scheduling Problem with Non-Linear Charging Time
Chair and presenter: Shangqing Cao, University of California, Berkeley
To better assess the economic feasibility of a UAM service, we introduce a vehicle-routing formulation of the Urban Air Mobility (UAM) fleet scheduling problem. The proposed model, electric urban air mobility vehicle routing problem with non-linear battery charging time (eUAMVRP-NL), studies the optimal flight scheduling and charging policy for a given fleet of eVToL vehicles to maximize the operating revenue with the assumption of a non-linear charging model for the aircraft batteries. We design a Cluster-First Route-Second (CFRS) solution heuristic that first decomposes the k-vehicle problem into k number of 1-vehicle problems before optimizing the f light scheduling and charging decisions for each aircraft in the fleet. We validate our CFRS algorithm on a small, two-vertiport instance in which an optimal solution can be obtained. We found a loss of optimality less than 2\%. Finally, we demonstrate the use of eUAMVRP-NL in obtaining meaningful metrics for the estimation of capital cost and operating cost by considering the airport access use case of UAM at Los Angeles International Airport (LAX).
UCAirSim: Designing and Evaluating Cross-Campus UAM Network
Andrew Park, University of California, Berkeley
Urban Air Mobility (UAM) is emerging as a transformative mode of intra-regional transportation, enabled by electric vertical takeoff and landing (eVTOL) aircraft. This study presents UCAirSim, a discrete-event simulation framework designed to evaluate the performance of an inter-campus air taxi network connecting CITRIS campuses and major research hubs in Northern California. The framework integrates a Campus Interconnectivity Simulation (CIS)—a macroscopic, event-driven model of fleet scheduling, demand balancing, and energy dynamics—with a Guidance, Navigation, and Control (GNC) simulator that validates flight trajectories, aeropropulsive performance, and noise impacts using high-fidelity NASA eVTOL models. The analysis evaluates system capacity, fleet sizing, and performance under fluctuating demand and varying flight paths. Results show that a fleet of 31 aircraft could serve approximately 400 passengers per day under nominal conditions, with scalability up to 1,600 passengers under optimized scheduling. Average wait times were reduced by incorporating repositioning strategies, while nonlinear battery charging dynamics constrained operational efficiency. Environmental and stakeholder considerations were also addressed, including vertiport siting, multimodal connectivity, and static noise policies. By combining network-level scheduling with vehicle-level dynamics, UCAirSim provides a scalable platform for analyzing operational strategies, environmental impacts, and future optimization approaches. This dual-layer simulation offers decision support for fleet operators, policymakers, and researchers in planning sustainable and efficient UAM systems for intraregional connectivity.
ATS Best Student Competition
Award Session
Air Transportation Section
A Dynamic Model for Airline Fleeting and Schedulings
Jingxing Wang, University of California, Berkeley
The global airline industry slumped during COVID-19: demand was volatile, cancellations surged, and airlines struggled to align capacity with shifting demand, pushing load factors to historic lows. In the recovery, airlines have institutionalized several measures first adopted during the pandemic, such as offering passengers greater flexibility in booking and cancellation. As a result, airlines are increasingly required to be agile and adaptive in their operational decisions. We introduce a dynamic model to help airlines make adaptive fleeting and scheduling decisions based on stochastically evolving bookings. The model incorporates both arrivals (new bookings) and departures (cancellations). This significantly generalizes and strengthens previous modeling attempts. We develop a Lagrangian relaxation framework that decomposes the dynamic program on a large time–space network into separable flight-level problems, and we analyze the tightness of this relaxation under representative flight network structures. Owing to the complexities of this network, our Lagrangian dual problem is not straightforward to solve. We therefore develop a tailored yet simple projected subgradient algorithm that exploits the structure of this network for efficient solutions. We establish the correctness and convergence properties of this procedure. Our analysis yields new theoretical, algorithmic, and managerial insights into the dynamic fleeting and scheduling problem. We present computational experiments based on real-world airline booking data to demonstrate the potential benefits of this approach.
The Present and Future of Operational Resilience in Air Transportation
Panel Session
Transportation Science and Logistics (TSL)
Panelist: Mark Hansen, University of California, Berkeley
Recent, high-profile events have disrupted the national airspace system, resulting in significant delays and cancellations. Examples include winter storms in the southeast and the CrowdStrike software outage. These events demonstrate the need for resilient air transportation operations. When extreme weather, infrastructure outages, and other disruptive events degrade operations, it is critical to contain these events and avoid system collapse. This panel will discuss present and future efforts to improve operational resilience. Panelists will discuss what resilience means to their organization and the current efforts that are being taken to improve resilience. Panelists will also discuss current and future challenges, and the potential for operations researchers and management scientists to contribute to this topic.
Smart Transportation Systems
Job Market Showcase
Optimizing Freight Electrification: Stochastic Routing and Grid-Aware Infrastructure Design
Chair and presenter: Ruiting Wang, University of California, Berkeley
Electrifying heavy-duty trucks is critical for reducing emissions in logistics but remains challenging due to high energy demands, operational uncertainties, and infrastructure limitations. This talk presents innovative strategies to overcome these challenges, including advanced optimization for routing and infrastructure design. I will show our methods of energy-efficient routing for mixed fleets of electric and conventional trucks under energy consumption uncertainty. Reformulated as a second-order cone mixed-integer program, this approach reduces operating costs, cuts emissions, and greatly reduces route capacity violations under real-world energy consumption uncertainty. In addition, I will compare battery swapping and fast charging mechanisms for heavy-duty freight from different perspectives. Focusing on California’s logistics network, we introduce an optimization framework for station location, sizing, and battery sizing, leveraging real-world data on freight traffic, energy prices, renewable generation, and grid services. Through evaluations of transportation efficiency, battery utilization efficiency, and grid service potential, we provide pioneering insights into charging infrastructure networks for sustainable freight systems. Finally, I will discuss open challenges in operation and grid-aware infrastructure planning, and mobility operation, how scalable solutions are desired in this area, and my proposals to address them.
Advances in Air Transportation
Invited Session
TSL: Air Transportation
A Dynamic Model for Airline Fleeting and Scheduling
Jingxing Wang, University of California Berkeley, Chiwei Yan, Archis Ghate
COVID-19 has reshaped the global airline industry. Travel demands are volatile, and passengers have more flexibility in bookings and cancellations. More than ever, airlines have to be agile and adaptive while making operational decisions. We introduce a dynamic model to help airlines make adaptive fleeting and scheduling decisionsbased on stochastically evolving bookings. The model incorporates both arrivals (new bookings) and departures (cancellations). This significantly generalizes and strengthens previous modeling attempts. We develop a Lagrangian relaxation framework that decomposes the dynamic program defined on a large time-space network into separable flight-level problems. Owing to the complexities of this network, our Lagrangian dual problem is not straightforward to solve. We therefore develop a tailored yet simple projected subgradient algorithm that exploits the structure of this network for efficient solutions. We establish the correctness and convergence properties of this procedure. Our analysis yields new theoretical, algorithmic, and managerial insights into the dynamic fleeting and scheduling problem. We present computational experiments based on real-world airline data to demonstrate the potential benefits of this approach.
Operational Resilience in Aviation
Invited Session
Air Transportation Section
Identification and Case Studies of Operational Disruptions in the U.S. National Airspace System
Jing Xu, University of California, Berkeley
Disruptions in the National Airspace System (NAS) lead to significant losses to air traffic system participants and raise public concerns. We apply two methods, cluster analysis and anomaly detection models, to identify operational disruptions with geographical patterns in the NAS since 2010. We identify four types and twelve categories of days of operations, distinguished according to air traffic system operational performance and geographical patterns of disruptions. Anomaly detection results show good agreement with cluster results and further distinguish days in the same cluster by severity of disruptions. Results show an increasing trend in frequency of disruptions, especially post-COVID. Additionally, disruptions happen most frequently in the summer and winter.
Intelligent Fulfillment and Transportation
Job Market Showcase
Physics-Informed Machine Learning for Enhanced Traffic Control: from Megavandertest to Neural Finite Volume Methods
Zhe Fu, University of California, Berkeley
Strategically controlling a small number of "leader" vehicles can significantly enhance traffic flow, improving efficiency, fuel economy, and driver comfort for the entire traffic system with minimal intervention. This was validated in the MegaVanderTest, a large-scale field experiment involving 100 autonomous vehicles (AVs) operating inmixed autonomy traffic. We designed both model-based and machine learning-driven control strategies for the AVs to improve traffic congestion and reduce the system-wide fuel consumption by over 10%.
However, the effectiveness of these control strategies heavily relies on accurate traffic state predictions and reliable communication. To address these challenges, we developed the Neural Finite Volume Method (NFVM), a physics-informed machine learning framework that integrates neural networks with physics-based models to solve hyperbolic partial differential equations (PDEs). NFVM achieves high predictive accuracy while maintaining key physical principles such as conservation laws and entropy conditions, making it well-suited to capturing complex traffic phenomena like shocks and stop-and-go waves. NFVM not only outperforms traditional finite volume solvers in accuracy but does so with comparable computational cost. It also offers structural flexibility, enabling multi-timestep inputs and the ability to generalize from PDE-synthetic data to experimental field data. By embedding NFVM as the traffic state estimation layer within the AV control stack, we further enhance the performance and robustness of our control strategies.
Online Resource Allocation
Invited Session
Manufacturing and Service Operations Management (MSOM)
Matching Queues, Flexibility and Incentives
Chiwei Yan, University of California, Berkeley, Francisco Castro, Peter Frazier, Hongyao Ma, Hamid Nazerzadeh
Agents in online marketplaces (such as ridesharing and freelancing platforms) are often strategic, and heterogeneous in their compatibility with different types of jobs: fully flexible agents can fulfill any job, whereas specialized agents can only complete specific subsets of jobs. Convention wisdom suggests reserving agents that are more flexible whenever possible, however this may incentivize agents to pretend to be more specialized, leading to loss in matches. We focus on designing a practical matching policy that performs well in a strategic environment.
Market Design and Network Games (I)
Invited Session
TSL: Urban Transportation Planning and Modeling
Information Design for Strategic Driver Reposition on Ride-Hailing Platforms
Chair and presenter Manxi Wu, University of California, Berkeley, Ozan Candogan
In this paper, we study how ride-hailing platforms like Uber and Lyft can strategically share information about the demand distribution with drivers to influence repositioning behavior. On these platforms, drivers have the flexibility to move strategically to different locations for better payoffs, and the spatial distribution of service requests and drivers is crucial for platform efficiency and revenue. We model drivers' strategic repositioning decisions as a game. Using a Bayesian persuasion framework, we show that in many practically relevant cases, the optimal mechanism takes a simple threshold form: the demand realization is fully revealed when it is below or above certain thresholds, and not revealed in the intermediate region. We also develop a computational method to solve for optimal information mechanisms in general settings, and extend the framework to the joint design of information and pricing.