Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distributed Fair Assignment and Rebalancing for Mobility-on-Demand Systems via an Auction-based Method

Kaier Liang, Cristian-Ioan Vasile

TL;DR

The paper tackles fair and scalable assignment of complex, temporally constrained mobility-on-demand requests expressed as scLTL formulas to a fleet of vehicles. It introduces a distributed auction-based method with a weight-correction phase and a complementary rebalancing scheme, underpinned by automata-based routing via product automata and Dijkstra. The approach minimizes the total travel time $J = \sum_r \sigma_v(r)$ while enhancing fairness across vehicles, measured by max-min utility and utility deviation, all without a central coordinator. Through a large-scale, mid-Manhattan simulation, the authors demonstrate that the combination of rebalancing and weight correction improves fairness metrics without sacrificing servicing quality, and that the distributed method can match the performance of centralized ILP baselines on average.

Abstract

In this paper, we consider fair assignment of complex requests for Mobility-On-Demand systems. We model the transportation requests as temporal logic formulas that must be satisfied by a fleet of vehicles. We require that the assignment of requests to vehicles is performed in a distributed manner based only on communication between vehicles while ensuring fair allocation. Our approach to the vehicle-request assignment problem is based on a distributed auction scheme with no centralized bidding that leverages utility history correction of bids to improve fairness. Complementarily, we propose a rebalancing scheme that employs rerouting vehicles to more rewarding areas to increase the potential future utility and ensure a fairer utility distribution. We adopt the max-min and deviation of utility as the two criteria for fairness. We demonstrate the methods in the mid-Manhattan map with a large number of requests generated in different probability settings. We show that we increase the fairness between vehicles based on the fairness criteria without degenerating the servicing quality.

Distributed Fair Assignment and Rebalancing for Mobility-on-Demand Systems via an Auction-based Method

TL;DR

The paper tackles fair and scalable assignment of complex, temporally constrained mobility-on-demand requests expressed as scLTL formulas to a fleet of vehicles. It introduces a distributed auction-based method with a weight-correction phase and a complementary rebalancing scheme, underpinned by automata-based routing via product automata and Dijkstra. The approach minimizes the total travel time while enhancing fairness across vehicles, measured by max-min utility and utility deviation, all without a central coordinator. Through a large-scale, mid-Manhattan simulation, the authors demonstrate that the combination of rebalancing and weight correction improves fairness metrics without sacrificing servicing quality, and that the distributed method can match the performance of centralized ILP baselines on average.

Abstract

In this paper, we consider fair assignment of complex requests for Mobility-On-Demand systems. We model the transportation requests as temporal logic formulas that must be satisfied by a fleet of vehicles. We require that the assignment of requests to vehicles is performed in a distributed manner based only on communication between vehicles while ensuring fair allocation. Our approach to the vehicle-request assignment problem is based on a distributed auction scheme with no centralized bidding that leverages utility history correction of bids to improve fairness. Complementarily, we propose a rebalancing scheme that employs rerouting vehicles to more rewarding areas to increase the potential future utility and ensure a fairer utility distribution. We adopt the max-min and deviation of utility as the two criteria for fairness. We demonstrate the methods in the mid-Manhattan map with a large number of requests generated in different probability settings. We show that we increase the fairness between vehicles based on the fairness criteria without degenerating the servicing quality.
Paper Structure (13 sections, 7 equations, 4 figures, 2 algorithms)

This paper contains 13 sections, 7 equations, 4 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Formulation
  4. Vehicle, Environment, and Request Models
  5. Fairness
  6. Solution
  7. Fair Auction Based Assignment Scheme
  8. Automata-based Route Planning
  9. Fair Rebalancing
  10. Simulation
  11. Simulation Specifications
  12. Simulation Results
  13. Conclusions

Figures (4)

  • Figure 1: Different settings for the Mid-Manhattan Map: the colors in nodes represent the probability of a new request arrival.
  • Figure 2: rebalancing performance in different map settings
  • Figure 3: Performance comparison for rebalancing and weight correction (center map: 20 vehicles, 300 requests) from the minimum utility, utility deviation and average utility
  • Figure 4: Comparison between ILP and auction without using rebalancing and weight correction

Theorems & Definitions (4)

  • Definition 1: Finite Automaton
  • Definition 2: scLTL
  • Definition 3: Weighted Transition System
  • Definition 4: Weighted product automaton at time $t$