Learning Connectivity for Data Distribution in Robot Teams

Video Demonstration

Set up

To set this project up to work with Docker, see the docker/ directory for the Dockerfile and a bash script to start the container.

Dependencies

• graph_nets

pip install graph_nets "tensorflow>=1.15,<2" "dm-sonnet<2" "tensorflow_probability<0.9"

• gym==0.11.0

• progress

• stable_baselines==2.9.0

Model Training

python3 train.py -p cfg/nl.cfg where the first argument is the path of the experiment config file.

Model Testing

python3 test_all.py cfg/nl.cfg hops where the first argument is the config file that was used for training and the second argument is the filename for the output results.

Visualization

python3 test.py -p <path to model> -e StationaryEnv-v0 -v add -gif to save visualization to a GIF

Configuration files

The following configuration files were used to train the models for the paper:

• nl.cfg - varying the number of hops in the GNN
• n.cfg - stationary agents, varying the team size
• power.cfg - stationary agents, varying the transmit power
• flocking.cfg - varying the flocking initial velocity (variance of velocities cost)
• flocking_aoi.cfg - varying the flocking initial velocity (Age of Information cost)
• mobile.cfg - random walk mobile agents, varying the agent velocity
• n_mobile.cfg - random walk mobile agents, varying the team size

The following additional configuration files are only used for testing:

• gen_flocking_aoi_to_var.cfg
• gen_flocking_var_to_aoi.cfg
• gen_n.cfg
• gen_n_mobile.cfg