EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Hierarchical Deep Reinforcement Learning with Formal Collision Safety for WRSN Charging in Obstacle-Dense Environments

Jingjing Chen, Bo Yan, Rongjie Wang, Zhiming Huang

Electronics · 15.07.2026 · 15 (14) · 3103

DOI: 10.3390/electronics15143103 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Electronics, online erschienen am 15.07.2026. Verfasst von Jingjing Chen und 3 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Science Foundation of Fujian Province, China, Open Research Project of the Applied Technology Engineering Center for Intelligent Manufacturing in Fujian Provincial Universities.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Energy replenishment in Wireless Rechargeable Sensor Networks (WRSNs) is challenging in obstacle-dense environments. It tightly couples two sub-problems: charging scheduling and trajectory planning. Existing heuristics and end-to-end deep reinforcement learning (DRL) methods cannot jointly solve these tasks with formal safety guarantees. To address this, we propose a hierarchical DRL framework that decouples the two sub-problems. At the high level, a Deep Q-Network (DQN) selects the next charging target based on the global energy state. At the low level, an Improved A* planner uses obstacle dilation and Bézier smoothing to generate a provably collision-free and mechanically feasible trajectory. This dilation mechanism strictly ensures physical safety. Our ablation study shows that disabling this mechanism introduces 25.20 ± 13.31 collisions per episode. We evaluated our framework on networks with 100 nodes and 10 obstacles. Compared to the end-to-end DDPG baseline, our approach reduces the Node Dead Ratio (NDR) from 0.950 to 0.785 and completely eliminates collisions. Furthermore, it reduces the median cumulative turning angle by 96% across 20 randomized topologies, and achieves 3.8× the charging efficiency of the EDF heuristic. Ultimately, this framework occupies a distinct Pareto-optimal point on the efficiency–reliability–safety frontier.

Förderung

  • Natural Science Foundation of Fujian Province, China
  • Open Research Project of the Applied Technology Engineering Center for Intelligent Manufacturing in Fujian Provincial Universities

Zitieren

APA

Chen, J., Yan, B., Wang, R., & Huang, Z. (2026). Hierarchical Deep Reinforcement Learning with Formal Collision Safety for WRSN Charging in Obstacle-Dense Environments. Electronics, 15(14), 3103. https://doi.org/10.3390/electronics15143103

Vancouver

Chen J, Yan B, Wang R, Huang Z. Hierarchical Deep Reinforcement Learning with Formal Collision Safety for WRSN Charging in Obstacle-Dense Environments. Electronics. 2026;15(14):3103. doi:10.3390/electronics15143103

BibTeX
@article{Chen2026,
  author = {Chen, Jingjing and Yan, Bo and Wang, Rongjie and Huang, Zhiming},
  title = {Hierarchical Deep Reinforcement Learning with Formal Collision Safety for WRSN Charging in Obstacle-Dense Environments},
  journal = {Electronics},
  year = {2026},
  volume = {15},
  number = {14},
  pages = {3103},
  doi = {10.3390/electronics15143103}
}
Kostenlos starten →