Q-learning approach to automated Unmanned Air Vehicle (UAV) demining

This paper develops a Q-learning approach to Unmanned Air Vehicle (UAV) navigation, or path planning, for sensing applications in which an infrared (IR) sensor or camera is installed onboard the UAV for the purpose of detecting and classifying multiple, stationary ground targets. The problem can be considered as a geometric sensor-path planning problem, because the geometry and position of the sensor's field of view (FOV) determines what targets can be detected and classified at any given time. The advantage of this approach over existing path planning techniques is that the optimal guidance policy is learned via the Q-function, without explicit knowledge of the system models and environmental conditions. The approach is demonstrated through a demining application in which a UAV-based IR sensor is capable of determining the optimal altitude for properly detecting and classifying targets buried in a complex region of interest. © 2010 SPIE.

Full Text

Duke Authors

Cited Authors

  • Ferrari, S; Daugherty, G

Published Date

  • 2010

Published In

Volume / Issue

  • 7692 /

International Standard Serial Number (ISSN)

  • 0277-786X

Digital Object Identifier (DOI)

  • 10.1117/12.850135