Tsukada Lab. Computer Network Group

A Feature-Aware Elite-Imitation MARL for Multi-UAV Trajectory Optimization in Mountain Terrain Detection

Quanxi Zhou, Ye Tao, Qianxiao Su, Manabu Tsukada

Drones, Sep 13, 2025

Summary in three lines

  • FA-EIMARL plans coordinated UAV sensing routes over varied mountainous terrain.
  • It builds transferable features from partial terrain information and periodically shares strong agents’ policies through elite imitation.
  • In simulation it improves reward, convergence and computational efficiency while adapting to different terrain shapes.

Figures from the paper

Figure 2: Terrain data, modeling and policy learning
Figure 2. Terrain data, modeling and policy learningSource: Figure 2 from Zhou et al., Drones (2025) (authors’ figure from the manuscript)The flowchart shows how terrain observations and sensing, energy and communications models supply the learning environment.
Figure 7: Training rewards compared across learning algorithms
Figure 7. Training rewards compared across learning algorithmsSource: Figure 7 from Zhou et al., Drones (2025) (authors’ figure from the manuscript)The plot compares how rewards improve during training for FA-EIMARL and the tested baselines.

Video

Related video: Trajectory Planning for UAV-Based Smart Farming Using Imitation-Based Triple Deep Q-Learning

Related research and software

How to cite

Q. Zhou, Y. Tao, Q. Su, and M. Tsukada, “A Feature-Aware Elite-Imitation MARL for Multi-UAV Trajectory Optimization in Mountain Terrain Detection,” Drones, 2025, doi: 10.3390/drones9090645.

BibTeX

@article{Zhou2025b,
  title         = {A Feature-Aware Elite-Imitation MARL for Multi-UAV Trajectory Optimization in Mountain Terrain Detection},
  author        = {Quanxi Zhou and Ye Tao and Qianxiao Su and Manabu Tsukada},
  url           = {https://www.mdpi.com/2504-446X/9/9/645/pdf},
  doi           = {10.3390/drones9090645},
  year          = {2025},
  date          = {2025-09-13},
  urldate       = {2025-09-13},
  journal       = {Drones},
  keywords      = {Other},
  pubstate      = {published},
  tppubtype     = {article}
}