EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Lightweight Vision-Language-Action Policy with Progress-Aware Hybrid Execution for UAV Waypoint Navigation in AirSim

Yiqing Xu iD, Haifeng Lin iD, Yujin Yang iD, Ji'An Xia, Zidong Han iD

Sensors · 22.07.2026 · 26 (14) · 4655

DOI: 10.3390/s26144655 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 22.07.2026. Verfasst von Yiqing Xu und 4 weiteren Autor:innen. Beteiligte Einrichtung: School of Computer and Software/School of Artificial Intelligence, Nanjing University of Industry Technology, Nanjing 210023, China. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Jiangsu Autonomous Driving Technology Innovation and Application Engineering Research Center and Qing Lan Project, Jiangsu Autonomous Driving Technology Innovation and Application Engineering Research Center, Qing Lan Project.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Offline action prediction does not by itself guarantee reliable closed-loop flight for unmanned aerial vehicle (UAV) vision-language-action (VLA) models. We study a controlled AirSim Blocks waypoint task using 100 expert episodes and 3385 RGB-D, instruction, state, and action records. Checkpoints are selected only on val-seen data, after which val-unseen is evaluated once. A 132,840-parameter policy reaches 0.9278±0.0019 final-test action accuracy across three training seeds, yet raw VLA control fails in closed loop. We therefore embed its action proposals in progress-aware hybrid execution with explicit recovery and near-goal precision. The strongest checkpoint reaches 59/60 goals, but crossing three independently trained checkpoints with three target seeds yields a more conservative 147/180 successes (81.7%) with zero recorded collisions and marked checkpoint sensitivity. Substantial overrides and fallback-tagged steps further show that the reported closed-loop outcomes are properties of the hybrid system, not of the learned policy alone. These findings are restricted to the controlled AirSim Blocks benchmark and do not demonstrate real-UAV deployment, sim-to-real transfer, or field robustness.

Schlagwörter

Stichwörter der Autor:innen

AirSimRGB-D perceptionUAV navigationhybrid executionimitation learninglightweight neural networksvision-language-action model

Autor:innen und Institutionen

  • School of Computer and Software/School of Artificial Intelligence, Nanjing University of Industry Technology, Nanjing 210023, China.
  • College of Information Science and Technology & Artificial Intelligence, Nanjing Forestry University, Nanjing 210037, China.
  • School of Transportation Engineering, Nanjing University of Industry Technology, Nanjing 210023, China.
  • School of Aeronautic Engineering, Nanjing University of Industry Technology, Nanjing 210023, China.

ORCID-Profile: Yiqing Xu, Haifeng Lin, Yujin Yang, Zidong Han

Förderung

  • Jiangsu Autonomous Driving Technology Innovation and Application Engineering Research Center and Qing Lan Project
  • Jiangsu Autonomous Driving Technology Innovation and Application Engineering Research Center
  • Qing Lan Project

Zitieren

APA

Xu, Y., Lin, H., Yang, Y., Xia, J., & Han, Z. (2026). A Lightweight Vision-Language-Action Policy with Progress-Aware Hybrid Execution for UAV Waypoint Navigation in AirSim. Sensors, 26(14), 4655. https://doi.org/10.3390/s26144655

Vancouver

Xu Y, Lin H, Yang Y, Xia J, Han Z. A Lightweight Vision-Language-Action Policy with Progress-Aware Hybrid Execution for UAV Waypoint Navigation in AirSim. Sensors. 2026;26(14):4655. doi:10.3390/s26144655

BibTeX
@article{Xu2026,
  author = {Xu, Yiqing and Lin, Haifeng and Yang, Yujin and Xia, Ji'An and Han, Zidong},
  title = {A Lightweight Vision-Language-Action Policy with Progress-Aware Hybrid Execution for UAV Waypoint Navigation in AirSim},
  journal = {Sensors},
  year = {2026},
  volume = {26},
  number = {14},
  pages = {4655},
  doi = {10.3390/s26144655}
}
Kostenlos starten →