A Spiking Reinforcement Trajectory Planning for UAV-Assisted MEC Systems

This study aims to minimize the energy consumption of user equipments (UE) and unmanned aerial vehicles (UAV) in UAV-assisted mobile edge computing (MEC) systems through the optimization of UAV flight trajectories, user associations, and resource allocations. The problem is formally articulated as a mixed integer nonlinear programming (MINLP) problem. However, employing deep reinforcement learning (DRL) to address this challenge introduces computational complexity and convergence obstacles. In response to these challenges, we propose an efficient and robust approach known as spiking reinforcement trajectory planning (SRTP), which uniquely integrates spiking neural networks (SNN) with DRL. SRTP utilizes a Poisson encoding mechanism to seamlessly convert between spike and continuous signals, enabling collaborative learning between spike actor networks and deep critic networks. Additionally, a pseudo-spacetime backpropagation method is employed for accelerated training of the spike actor network. Experimental findings distinctly highlight SRTP’s advantages, demonstrating a more streamlined network structure and faster convergence compared to conventional methodologies. This innovative methodology holds promise in addressing the computational complexities and convergence challenges associated with implementing DRL in UAV-assisted MEC systems.