AI Fleet Coordination Reveals Trip-Length and Spatial Inequity in Manhattan, Chicago, and San Francisco
A new study from arXiv (2607.24336) diagnoses delay inequity in autonomous vehicle fleet coordination. Using real road-network and taxi-demand data from Manhattan, Chicago, and San Francisco, researchers conducted a distributional audit revealing pervasive trip-length inequity that varies by city and coordinator. Spatial inequity intensifies with demand, especially when trips are grouped by origin. The paper proposes SPARE (Spatially Aware Rerouting), a budgeted online framework that assigns limited replanning capacity to delayed vehicles using observed waiting pressure. SPARE provides per-review decision guarantees and bounds online route updates. Experiments across all three datasets against six baselines demonstrate its effectiveness.
Key facts
- Study uses real-city road-network and taxi-demand datasets from Manhattan, Chicago, and San Francisco.
- Audit reveals pervasive trip-length inequity depending on city and coordinator.
- Spatial inequity becomes more pronounced as demand grows.
- Inequity is stronger when trips are grouped by origin rather than destination.
- SPARE framework assigns limited replanning capacity to delayed vehicles.
- SPARE uses recently observed waiting pressure for rerouting.
- SPARE provides a per-review decision guarantee.
- Experiments conducted against six baselines on all three datasets.
Entities
Institutions
- arXiv
Locations
- Manhattan
- Chicago
- San Francisco