Abstract
AbstractUnmanned-Aerial-Vehicles’ (UAVs) inherent features such as high dynamicity, quick deployment, and line of sight communication have motivated the research of UAV-assisted IoT networks. In such networks, one critical issue is path planing scheduling, which unfortunately is a complex multi-objective optimization problem (MOP). Although there exist extensive traditional MOP algorithms, the efficiency is unacceptable due to the resource constrains and they are unscalable for dynamic scenarios. In order to achieve a more efficient yet scalable multi-objective path planing algorithm, we innovatively propose a framework integrating deep reinforcement learning (DRL) and transformer. We firstly decompose the MOP problem into a series of sequencing subproblems with weighted objectives, and then we present a modified transformer network to solve each sequencing subproblem and further a DRL algorithm to facilitate the subproblem network training. Experimental results demonstrate that the proposed algorithm is superior to NSGA-II, MOEA/D and pointer network in terms of robustness, convergence, diversity of solutions, and temporal complexity.KeywordsTransformerMulti objective optimizationDeep reinforcement learning
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have