Deep Reinforcement Learning-Based Collaborative Data Collection in UAV-Assisted Underwater IoT

In complex oceanic environments, efficient data collection for underwater Internet of Things (IoT) is a challenging issue. Unmanned aerial vehicle (UAV) technology provides an emerging solution to this problem. This article designs a UAV-assisted underwater IoT data collection system architecture. In this architecture, UAVs collect data through relay cooperation, effectively reducing the flight time for remote UAVs to return to the seashore base station (SBS). We propose a hybrid collaborative data collection (HCDC) scheme to generate a collection strategy for UAVs. In this scheme, we first allocate a balanced workload to the UAVs and determine the initial traversal order for each UAV to collect data from various surface sink nodes (SSNs). Then, we introduce a relay matching algorithm to establish the relay relationships among UAVs. Finally, we present a trajectory planning algorithm based on hierarchical multiagent deep deterministic policy gradient (MADDPG). Extensive simulation experiments demonstrate that the proposed HCDC scheme significantly reduces both the flight time and data collection time of the UAVs, thereby minimizing the age of information (AoI). Moreover, compared with other data collection schemes, HCDC performs better in networks of various scales.