Improving ridepooling reliability using shareability shadows

Abstract

A major limitation of ridepooling systems is travel-time unreliability: users may receive updated arrival times when vehicle itineraries are modified to accommodate additional passengers. This paper introduces a passenger-specific predictive model to estimate this unreliability. The model uses shareability shadows and expected demand rates to compute a Shareability Score, which captures the expected shareable demand along each passenger’s trip and indicates their likelihood of experiencing an increase in the initially estimated arrival time. Using this prediction, we apply targeted operational rules for each passenger, and we guarantee no travel-time updates for low-score passengers. As a benchmark, we consider a system-wide policy that manages unreliability uniformly across all passengers. Simulation results using data from Manhattan and Utrecht show that the proposed approach improves both reliability and efficiency compared with the benchmark. Furthermore, at the same rejection rate of 29% between the proposed and benchmark methods, the approach reduces the percentage of passengers experiencing updates to travel time by up to 47.8% and guarantees a fully reliable trip for 83.9% of passengers at the time of assignment. The model performs consistently across different fleet sizes, vehicle capacities, user tolerance levels, and networks, highlighting the value of managing ridepooling unreliability at the individual passenger level.