Scalable Perception-Action-Communication Loops with Convolutional and Graph Neural Networks

IEEE Transactions on Signal and Information Processing over Networks, 2022, Scalable Perception-Action-Communication Loops with Convolutional and Graph Neural Networks

The code is used to reproduce the result of "Scalable Perception-Action-Communication Loops with Convolutional and Graph Neural Networks"

Ting-Kuei Hu, Fernando Gama, Tianlong Chen, Wenqing Zheng, Zhangyang Wang, Alejandro Ribeiro and Brian M. Sadler

Prerequisite

• Before running the script, make sure you have install microsoft airsim on 4.23 version. The instruction for the installation is in https://microsoft.github.io/AirSim/
• After installing airsim, launching the airsim by GUI and play start. Make sure the setting.json is correctly located in the airsim folder.
• Pytorch 1.0.0
• If you would like to get the environment we use for airsim, please replace the project file with ours in airsim folder.

Quick start

• Run the batch file "exec_dagnn.sh"/"exec_grnn.sh"" for the training.
• replace the path of 'root_path' to change the path to store the dataset.
• change the setting according to the "Training hyper-parameters"

Training hyper-parameters

• n_times : how many steps for each trajectory.
• n_agents : number of agents for the group.
• n_exp : number of training data for the initial size of dataset.
• start_idx : set 0 for the starting point of each trajectory.
• filterLength : the temporal length for DAGNN/GRNN.
• n_vis : number of feature dimension for transmission.
• radius : radius for the disk model.
• vinit : the maximum velocity for initialization.
• mode : 'optimal' using centralized controller to collect ground truth. 'vis_grnn'/'vis_dagnn'/ using other controllers to collect ground truth.
• arch: 'vis_grnn'/'vis_dagnn' for the choice of archetecture.
• comm_model : 'disk' for disk model. 'knn' for knn model.
• n_exp_aug : number of datasize for the augmentation of dataset.

Quick test

• Run the batch "exec_test.sh"
• change the setting according to the "Testing hyper-parameters"

Testing hyper-parameters

• n_agents : number of agents for each trajectory
• arch : 'vis_grnn'/'vis_dagnn' for the choice of archetecture.
• seed : the random seed for the initialization.
• Other hyper-parameters in "Training hyper-parameters".