Multi-Space Crowd Sensing Task Allocation: A Dynamic Co-Optimization Framework With Fairness-Aware Reinforcement Learning

doi:10.1109/TMC.2025.3640127

Multi-Space Crowd Sensing Task Allocation: A Dynamic Co-Optimization Framework With Fairness-Aware Reinforcement Learning

2025-12-04

0

PRE

AI

DOI：10.1109/TMC.2025.3640127

Original

Original request for help

Save

Abstract

En 中文

Multi-space crowd sensing has emerged as a promising paradigm for 3D urban perception. However, it faces critical challenges including space coupling, task heterogeneity, and dynamic resource availability. To address these issues, the Multi-Space Fairness Task Allocation (MSFTA) problem is formulated, aiming to maximize task completion while ensuring fairness across spatial dimensions. The problem is proven to be NP-hard, and a dynamic collaborative optimization framework is proposed. Within this framework, a Multi-Space Clustering QuadTree Voronoi Partition (MCQVP) is developed for fine-grained multi-dimensional partitioning by leveraging DBSCAN and quadtree structures. In addition, a Group Urgency-Based Multi-Shortest Path (GUBMSP) scheduler is incorporated to prioritize time-sensitive task groups via urgency-aware critical paths. Furthermore, a Fairness-Aware Pareto Multi-Objective Ant-Q Learning (FA-PMOAQL) allocator is introduced to integrate Q-learning and ant-colony optimization under fairness-aware multi-objective guidance. These designs establish a unified framework that not only improves task allocation efficiency through multi-space partitioning and urgency-driven scheduling, but also ensures equitable resource utilization by embedding fairness into the learning process. Comparison experiments on Tokyo and New York datasets demonstrate that the proposed approach achieves up to 12.8% higher task completion rate compared with baseline algorithms, while maintaining relatively low runtime. In cross-layer scenarios, the completion rate improves by 20% when agent resources increase, and under heavy task loads it sustains competitive performance with only moderate decline.

Keywords:

Crowd sensing

task allocation

fairness

reinforcement learning

multi-space

Multi-Space Crowd Sensing Task Allocation: A Dynamic Co-Optimization Framework With Fairness-Aware Reinforcement Learning

Abstract

Journal

Organization