FairSense: Fairness-Aware Urban Sensing with Submodular Spatio-Temporal Reward Maximization

Urban challenges like air pollution, traffic congestion, and road quality demand precise and timely detection for effective city monitoring. Traditional stationary sensors, costly with limited coverage, have prompted the emergence of mobile sensing (e.g., vehicular urban sensing), which utilizes ubiquitous mobile devices (e.g., sensors on vehicles) for sensing tasks. While commercial vehicles (e.g., taxis) are favored for urban sensing due to easy data collection, biased spatiotemporal distributions often result in uneven sensing coverage. Despite efforts to maximize task completion and incentivize participation, fairness in coverage distribution remains neglected. To address this issue, we propose a hybrid approach integrating Dedicated Sensing Vehicles (DSVs) and commercial vehicles to optimize coverage while accounting for fairness constraints. Routing these DSVs for fairness-aware spatiotemporal coverage poses unique challenges, including achieving high coverage with limited DSVs, prioritizing underserved areas, and ensuring cost efficiency for financial sustainability. To overcome these challenges, in this paper, we design FairSense, a system, which combines the long-term reward optimization capabilities of deep reinforcement learning (DRL) with submodular maximization. FairSense integrates a submodular reward function with spatio-temporal fairness constraints into a DRL algorithm to prioritize long-term gain and fairness, enabling informed path planning for DSVs, thereby enhancing efficiency and fairness in sensing coverage. We evaluate our FairSense with real-world large-scale vehicle GPS data from two cities (i.e., Shenzhen and Beijing), and extensive experiments show our FairSense can effectively improve sensing coverage efficiency and also improve sensina fairness.