Deploying Mobility-On-Demand for All by Optimizing Paratransit Services

Sophie Pavia, David J. Rogers, Amutheezan Sivagnanam

Overview

Paratransit services provide essential door-to-door transportation for individuals with disabilities, as mandated by the Americans with Disabilities Act (ADA). Despite their importance, these services are often inefficient and costly. In Chattanooga, Tennessee, paratransit accounts for 22% of total service miles while serving less than 4% of total passenger trips. This paper presents a data-driven optimization framework developed in collaboration with the Chattanooga Area Regional Transportation Authority (CARTA) to address these operational challenges.

The Complexity of Paratransit Routing

The core challenge lies in solving a complex Vehicle Routing Problem with Pickup and Delivery and Time Windows (VRPPDTW). Unlike standard ride-hailing services, paratransit must manage variable vehicle capacity and strict legal constraints.

Daily paratransit ridership in Chattanooga, showing the breakdown between ambulatory and wheelchair users.

As shown in Figure 3, ridership varies significantly day-to-day, and the mix of ambulatory versus wheelchair passengers directly impacts vehicle capacity. A wheelchair takes up considerably more space than an ambulatory passenger, requiring dynamic capacity management. Furthermore, travel times must not exceed those of a comparable fixed-route bus trip, and pickup/drop-off windows are strictly regulated. Existing commercial "black-box" software often fails to find optimal groupings, leading to manual scheduling that can take dispatchers up to six hours a day.

Methodology: A Two-Tiered Algorithmic Approach

To optimize these services, the researchers developed a two-tiered solver:

  1. Offline Solver: For advance requests, a VRPPDTW formulation is solved using Guided Local Search via Google’s OR-Tools. This minimizes total Vehicle Miles Traveled (VMT) while respecting all capacity and time constraints.
  2. Online Solver: For same-day requests, an insertion heuristic slots new trips into existing schedules by minimizing the ratio of vehicle miles to passenger miles (VMT/PMT).

Human-in-the-Loop Interface

A critical component of the system is the "glass box" approach. Recognizing that human dispatchers possess tacit knowledge about passenger needs and driver preferences, the system includes a web-based interface for human supervision.

The SmartTransit interface allows dispatchers to view, edit, and validate schedules with real-time constraint checking.

As illustrated in Figure 5, the interface allows for drag-and-drop adjustments. A backend "constraint checker" provides immediate feedback, ensuring that manual overrides do not violate legal time windows or vehicle capacity limits. This maintains human control while leveraging AI for heavy computational lifting.

Key Results and Impact

The framework was evaluated using real-world operational data from CARTA, including GPS logs and road network data from OpenStreetMap. The results demonstrated significant improvements over the agency's existing commercial tools:

  • Efficiency Gains: Simulations showed a 21.3% reduction in the VMT/PMT ratio.
  • Increased Sharing: The AI-optimized routes achieved a shared ride rate of 85%, compared to the baseline of 53%.
  • Operational Speed: Generating a full day's manifest was reduced from hours of manual labor to just a few minutes of computation.
  • Reliability: The system served all passengers with zero time-window violations, whereas the baseline manifests frequently scheduled pickups outside the legal 15-minute window.

A live pilot deployment in August 2023 confirmed these results, reducing total vehicle miles by hundreds per day while serving the same number of passengers.

Conclusion

This research demonstrates that data-driven optimization can significantly improve the efficiency and reliability of paratransit services. By combining advanced VRP solvers with a human-in-the-loop interface, transit agencies can reduce costs and environmental impact while providing better service to passengers with mobility needs. Future work will explore scaling these techniques to larger metropolitan areas and integrating multimodal transfers between paratransit and fixed-route systems.

Summary updated

BibTeX
@inproceedings{pavia2024deploying,
  author = {Sophie Pavia and David J. Rogers and Amutheezan Sivagnanam},
  title = {{Deploying Mobility-On-Demand for All by Optimizing Paratransit Services}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2024},
  url = {https://openreview.net/forum?id=mS9g0JXD2l},
}
Figure