Skip to content

A reinforcement learning algorithm to navigate a group of drones (UAV) tracking one moving target

License

Notifications You must be signed in to change notification settings

RuoxiQin/Unmanned-Aerial-Vehicle-Tracking

Repository files navigation

Unmanned-Aerial-Vehicle-Tracking

As Unmanned Aerial Vehicle’s (UAV) battery life and stability develop, multiple UAVs are having more and more applications in the uninterrupted patrol and security. Thus UAV’s searching, tracking and trajectory planning become important issues. This paper proposes an online distributed algorithm used in UAV’s tracking and search- ing, with the consideration of UAV’s practical need to recharge under limited power. We propose a Quantum Probability Model to describe the partially observable target positions, and we use Upper Confidence Tree (UCT) algorithm to find out the best searching and tracking route based on this model. We also introduce the Teammate Learning Model to handle the nonstationary problems in distributed reinforcement learning.

Problem and Algorithm

The explanation of the problem and the algorithm can be found here.

Requirements

  • Python 2.7

Experiments

To run the experiments, first set up the parameters of the algorithm and testing environment in test_file.py:

  • all_work: a dict containing a batch of experiment parameters.
  • name: The name of this experiment. The experiment record will be saved in name.txt.
  • T: Number of search times in each UCT planing, as explained in the paper.
  • D: The depth of the search tree, aas explained in the paper.
  • test_time: Repeat this experiment test_time times.
  • size: The size of the search area.
  • base_lists: A list of the locations of the charging bases. You can set any number of charging bases in the experiment.
  • plan_lists: A list of the initial positions of the drones. You can set any number of drones in the experiment.

Then you can start the experiment using command:

python test_file.py

The simulation of each step will be printed to stdout. The log of each experiment will be stored in the corresponding name.txt.

About

A reinforcement learning algorithm to navigate a group of drones (UAV) tracking one moving target

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages