private_flocking

Code accompanying the paper

"An Adversarial Approach to Private Flocking in Mobile Robot Teams"
in IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 1009-1016, April 2020
doi: 10.1109/LRA.2020.2967331

Requirements

Simulation environment

Unreal Engine (UE4) v4.18
AirSim

Learning libraries

Python >= 3.5
PyGMO == 2.10
PyTorch == 1.2.0
torchvision == 0.4.0

Instructions to compile and run AirSim with UE4can be found here

Then

$ git clone https://github.com/proroklab/private_flocking.git
$ cd private_flocking/

Discriminator pre-training

Before the co-optimization, the discriminator needed to be pre-trained.
This step can be skipped by using our pretrain-weights.py.

Collect pre-training data

$ cd pretrain_discriminator && python 1_collect_cnn_data.py

Pre-training

$ cd pretrain_discriminator/scripts
$ ./2_run_parse_cnn_data.sh
$ ./3_run_pretrain.sh
$ ./4_tensorboard.sh

• After pre-training, weights.pt need to be manually copied into folder private_flocking and renamed as pretrain-weights.py

Co-optimization

$ cd scripts && ./run.sh

Please note that arguments used for the co-optimization experiment are stored in config.py

Visualization

Visualize a single flocking simulation.

$ cd scripts && ./summary-plot.sh

Visualize and compare up to 4 flocking simulations.

$ cd scripts && ./summary-plot.sh

Visualize the co-optmization process.

$ cd scripts && ./evolution-plot.sh

where the optimization id and simulation id for visualization should be changed in each script.

Citation

If you like, please cite our paper as:

@ARTICLE{zheng2020adversarial,
  author={Zheng, Hehui and Panerati, Jacopo and Beltrame, Giovanni and Prorok, Amanda},
  journal={IEEE Robotics and Automation Letters}, 
  title={An Adversarial Approach to Private Flocking in Mobile Robot Teams}, 
  year={2020},
  volume={5},
  number={2},
  pages={1009-1016},
  doi={10.1109/LRA.2020.2967331}}