# Load Balancing in Mobility-on-Demand Systems: Reallocation Via Parametric Control Using Concurrent Estimation

## TL;DR

This paper introduces a parametric control method with concurrent estimation for load balancing in mobility-on-demand systems, improving service consistency amid dynamic demand and traffic conditions.

## Contribution

It proposes a novel concurrent estimation approach to optimize parametric load balancing control in MoD systems, surpassing trial-and-error methods.

## Key findings

- Concurrent estimation improves load balancing performance.
- State information enhances system efficiency.
- The proposed method outperforms static and no-control policies.

## Abstract

Mobility-on-Demand (MoD) systems require load balancing to maintain consistent service across regions with uneven demand subject to time-varying traffic conditions. The load-balancing objective is to jointly minimize the fraction of lost customer requests due to vehicle unavailability and the fraction of time when vehicles drive empty during load balancing operations. In order to bypass the intractability of a globally optimal solution to this stochastic dynamic optimization problem, we propose a parametric threshold-based control driven by the known relative abundance of vehicles available in and en route to each region. This is still a difficult parametric optimization problem for which one often resorts to trial-and-error methods where multiple sample paths are generated through simulation or from actual data under different parameter settings. In contrast, this paper utilizes concurrent estimation methods to simultaneously construct many sample paths from a single nominal sample path. The performance of the parametric controller for intermediate size systems is compared to that of a simpler single-parameter controller, a state-blind static controller, a policy of no control, and a theoretically-derived lower bound. Simulation results show the value of state information in improving performance.