RLC: A Reinforcement Learning-Based Charging Algorithm for Mobile Devices

Wireless charging has been demonstrated as a promising technology for prolonging device operational lifetimes in Wireless Rechargeable Networks ( WRNs ). To schedule a mobile charger to move along a predesigned trajectory to charge devices, most existing studies assume that the precise location information of devices is already known. Unfortunately, this assumption does not always hold in real mobile application, because the activities of the vast majority of mobile devices carried by mobile agents appear dynamic and random. To the best of our knowledge, this is the first work to study how to wirelessly charge mobile devices with non-deterministic mobility. We aim to provide effective charging service to them, subject to the energy capacity of the mobile charger. We formalize the effective charging problem as a charging reward maximization problem ( CRMP ), where the amount of reward obtained by charging a device is inversely proportional to the residual lifetime of the device. Then, we prove that CRMP is NP-hard. To derive an effective charging heuristic, an algorithm based on Reinforcement Learning ( RL ) is proposed. The evaluation results show that the RL-based charging algorithm achieves excellent charging effectiveness. We further interpret the learned heuristic to gain deep and valuable insights into the design options.