| Title |
GNN-based Resource Allocation and Path Planning for Data Mule Data Retrieval in Extreme Cold Region DTN Environments |
| Authors |
황아리(Ari Hwang) ; 염선호(Sun-Ho Yum) ; 윤동진(Dong-Jin Yoon) ; 박수현(Soo-Hyun Park) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.2.54 |
| Keywords |
Extreme DTN environment; Graph neural networks; Deep reinforcement learning; Co-optimization; Data mule |
| Abstract |
Unlike the ground environment of temperate climates, polar regions such as Antarctica and the Arctic are environments where communication is easily disconnected due to climate and topographical features. In such an environment, it is essential to use delay/error tolerance network technology for stable transmission without data loss and to use mobile nodes such as ground mobile robots, UAVs, and drones as data carriers to overcome polar environments such as cryogenic temperatures, strong winds, snowfall, and crevasses. However, operating mobile nodes due to the environmental characteristics of polar regions creates path planning problems and limited resource (energy, buffer) management problems at the same time. In particular, in an environment where a small number of nodes are distributed in a wide area of several tens of kilometers, reinforcement learning agents are likely to fail in initial learning due to the sparse reward problem, which is difficult to obtain rewards only by random search. To solve this problem, this paper proposes a Path-Resource co-optimization framework that combines pre-training phase and Deep Reinforcement Learning (DRL) using driving data of rule-based algorithms. The proposed model maximizes initial exploration efficiency by learning verified path planning patterns in advance and adapts to the dynamic network phase through graph neural network (GNN). As a result of the simulation, it was confirmed that the proposed model secures the initial learning stability compared to the existing techniques and improves the data transfer rate and energy efficiency. |