Learning Based Dynamic Resource Allocation in UAV-Assisted Mobile Crowdsensing Networks

Unmanned aerial vehicles (UAV) assisted mobile crowdsensing (MCS) is an emerging paradigm that utilizes mobile user (MU) collaboration to complete sensing tasks. However, little attention has been paid to the issue of how to solve the resource allocation of sensing, communication, and computing processes, as well as the trajectory planning of UAV. Therefore, this paper focuses on maximizing the total completion of sensed bits in the UAV-assisted MCS network by jointly optimizing MU selection, resource allocation, and UAV trajectory planning. Considering the random mobility of MUs and the limited communication, computation, and energy resources, we formulate a nonconvex optimization problem that necessitates real-time decision-making. To tackle this demanding problem, we approach it by formulating it as a Markov decision process (MDP). In response, we propose a real-time solution based on proximal policy optimization (PPO) to obtain an approximate suboptimal solution for the problem. Numerical results show that the proposed PPO-based method yields a noteworthy improvement in the completion of sensed bits within when compared to other benchmark schemes.