Learning based collaborative task allocation
Loading...
Date
2024-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of New Brunswick
Abstract
In the digital era, Mobile Crowdsensing Systems (MCS) utilize mobile and wearable devices for large-scale data collection, forming participatory sensor networks. A primary challenge in MCS is collaborative task allocation, where multiple workers must coordinate to complete tasks. Our approach integrates workers’ social connections, recognizing that those with similar backgrounds collaborate more effectively. We frame this task allocation problem as a graph-based combinatorial optimization task, complicated by spatial, temporal, and social constraints. To ad dress the limitations of traditional heuristics, we propose a Heterogeneous Graph Attention Network-based Double Deep Q-Network (HGDQN). The HGDQN agent autonomously explores and learns to address complex scenarios by capturing nuanced worker-task relationships through subgraphs within the heterogeneous graph. Experimental results show that HGDQN surpasses traditional heuristic methods such as Greedy and ACO methods in scalability, adaptability, and generalization, providing a robust solution for collaborative task allocation in MCS.