/tvt2022

Oracle-Guided Deep Reinforcement Learning for Large-Scale Multi-UAVs Flocking and Navigation

Primary LanguageC++

tvt2022

  • A quick guide to install the environment (Flocking - RL).
  1. apt update && apt install swig (The version SWIG Version 4.0.1 is ok.)
  2. cd env_wrapper
  3. ./reinstall
  • A quick guide to train td3 agent.

Some python packages may be needed, pls install these packages by pip install. (e.g. pip install torch), and then

  1. python td3_fast.py --agents=16 --folder=./logs --global_="GLOBAL"

If you find this work is helpful at some aspects, please considering cite our paper:

W. Wang, L. Wang, J. Wu, X. Tao and H. Wu, "Oracle-Guided Deep Reinforcement Learning for Large-Scale Multi-UAVs Flocking and Navigation," in IEEE Transactions on Vehicular Technology, vol. 71, no. 10, pp. 10280-10292, Oct. 2022, doi: 10.1109/TVT.2022.3184043.

The c++ implementation of the environment is placed in env_wrapper.

The uav_2d_ma_fast.py is a python wrapper of the environment.

td3_fast.py is the first stage training script.

sl_train.py is the second stage training script.

Some results can be found in the evaluation-fast.ipynb.

Our models (parameters of the Neural Networks) can be downloaded from https://box.nju.edu.cn/f/b668839eec3b478e9fd3/?dl=1.