Deep Reinforcement Learning Based Data Collection with Charging Stations

Incorporating efficient charging strategies into the trajectory planning process for unmanned aerial vehicles (UAVs) data collection missions remains a difficult task. In this paper, we propose a reinforcement learning (RL) approach for training trajectory planning policies which jointly considers data collection and charging. Firstly, a trajectory planning optimization problem constrained by charging and other environmental constraints is formulated. Secondly, a Markov decision process is constructed based on the proposed optimization problem. Then, the deep RL algorithm DDQN is employed to obtain the optimal trajectory planning policies, where the convolutional layers in the Q-network are utilized to extract the charging and other environmental information for decision-making. Finally, a custom data collection environment is built, and the simulation results demonstrate that the UAV successfully learns to collect more data through charging as well as satisfying the safety constraints guided by the trained policy.