A Vanderbilt University research team has won the Best Paper Award at the 2026 IEEE International Conference on Smart Computing (SmartComp 2026), held June 22–25 at the University of Messina in Italy. The award recognizes an AI-enabled decision framework that uses real-world transit data, optimization and simulation to operate demand-responsive transportation services—including microtransit and paratransit—under the practical constraints of streets, vehicles, passengers and curb space.
The paper, “Dynamic Pickup-and-Delivery Routing with Early-Arrival Waiting Limits and Station Relocation,” was authored by Agrima Khanna, Sophie Pavia, Fangqi Liu, Ayan Mukhopadhyay and Abhishek Dubey.
Demand-responsive transit services must continuously assign vehicles to riders as new requests arrive, all while honoring capacity limits, pickup and drop-off time windows, and existing commitments to riders already on board. Most academic models for this problem, known as dynamic pickup-and-delivery routing with time windows, assume a vehicle can wait indefinitely once it reaches a pickup location. That assumption breaks down agains
t real operating conditions: curb space is limited, vehicles can’t always wait at the curb, and idle vehicles need to be positioned somewhere useful for the next likely request rather than sitting still.
The project is led by Khanna, a Ph.D. student in computer science at Vanderbilt, together with fellow researchers Pavia, Liu and Mukhopadhyay, and Dubey, associate professor of computer science, electrical engineering and computer engineering and associate dean for research in Vanderbilt’s College of Connected Computing. Dubey directs the SCOPE Lab at Vanderbilt’s Institute for Software Integrated Systems, where the group’s research focuses on decision-making under uncertainty in cyber-physical systems spanning transportation, emergency response and electric infrastructure.
The team’s framework, called DVRP-S, builds vehicle repositioning directly into the routing plan instead of treating it as an afterthought. In most systems, deciding where to send an idle vehicle is a separate step handled after routing is done. But once a vehicle arrives too early and can’t sit at the curb, that separation breaks down — where to send it next becomes just as urgent as who to pick up next, so the team’s approach solves both problems together.
Real-time transit routing is computationally challenging because each decision must account for new requests, current vehicle locations, rider commitments and uncertain future demand. The team addressed this challenge with MC-DVRPS, an AI-enabled decision-making method that combines real-time optimization with data-driven simulation. One component quickly matches vehicles with riders who need service now, while another evaluates likely future demand and identifies useful locations for idle vehicles. Together, these components allow the system to act on current conditions while anticipating what the transit network may need next.
The team tested its approach against six months of real paratransit trip data from the Chattanooga Area Regional Transportation Authority, comprising nearly 26,000 requests. Across every fleet size and station setup tested, it served more riders and drove less empty mileage than every other method it was compared against. With a five-vehicle fleet, it served about 90% of requests, compared with roughly 88% for the strongest existing alternative and 65% for a simpler approach that doesn’t plan ahead. The gap widened with smaller fleets: at three vehicles, the team’s method served about 72% of requests, more than double the rate of the next-best method that doesn’t plan ahead for future demand.
“This award recognizes the extraordinary work of our students and collaborators, as well as the growing importance of combining AI, optimization and real-world transportation data,” Dubey said. “Recently, we participated in the Dagstuhl Seminar on Algorithmic Advances for a New Era of Data-Driven, Multi-Modal Transit Systems, part of a broader effort to identify the algorithms, partnerships and research directions that will shape the future of mobility. We are looking forward to a new era of data-driven, multimodal transit systems that can anticipate changing conditions, coordinate different modes of transportation and make mobility more reliable and accessible for everyone.”
This research was supported in part by the U.S. National Science Foundation and the U.S. Department of Energy, whose investments are advancing data-driven methods for resilient, efficient and accessible transportation systems. The work is part of the SCOPE Lab’s broader research activities, pursued through its SmartTransit.ai platform, the PATH-TN consortium—which supports transit planning and operations across Tennessee—and engagement with the international research community through activities such as the Dagstuhl seminar. By treating vehicle routing and idle-vehicle stationing as one integrated decision problem rather than two separate tasks, the team’s approach offers transit agencies and other mobility providers a more realistic and scalable way to serve more riders with the same limited fleet.