Sensing Strategy Optimization for Satellite-Assisted Vehicular Crowdsensing in Non-Terrestrial Networks

Abstract

Vehicular crowdsensing (VCS) is a paradigm that exploits vehicle mobility, on-board sensing capabilities, and drivers' smartphone sensors to collect large-scale, distributed information to provide intelligent, location-based services. Satellite-assisted VCS architecture can complement terrestrial networks by enabling wide-area and infrastructure-independent data collection. Incentivizing vehicles to participate in satellite-assisted VCS campaign remains a major challenge due to associated sensing and communication costs, requiring each vehicle to optimize their sensing level to maximize their received reward. Moreover, unlike conventional assumptions where all vehicles participate simultaneously, practical VCS scenarios are asynchronous, as vehicle may start and complete sensing tasks at different times. To capture this realistic setting, we propose an asynchronous multi-agent proximal policy optimization (A-MAPPO) algorithm within a centralized training and decentralized execution (CTDE) framework to optimize the sensing strategies of individual vehicles in a satellite-assisted VCS setting. A dynamic social network effect among vehicles is also incorporated to encourage vehicle participation driven by social benefits. Extensive numerical experiments are conducted to evaluate the performance of the proposed approach, demonstrating that A-MAPPO achieves superior performance compared with MASAC, DQN, Greedy-Q, and Random baselines.