GRL_CAVs is the source code for our paper: Graph Reinforcement Learning Application to Co-operative Decision-Making in Mixed Autonomy Traffic: Framework, Survey, and Challenges.
GRL_CAVs is an all-round improvement and optimization source code based on our previously repository TorchGRL. GRL_CAVs is a modular simulation framework that integrates different GRL algorithms, flow interface and SUMO simulation platform to realize the simulation of multi-agents decision-making algorithms for connected and autonomous vehicles (CAVs) in mixed autonomy traffic. You can design your own traffic scenarios, adjust the implemented GRL algorithm or do your mprovements for a particular module according to your needs.
Before starting to carry out some relevant works on our framework, some preparations are required to be done.
Our framework is developed based on a laptop, and the specific configuration is as follows:
We suggest that our program should be reproduced under the Ubuntu 20.04 operating system, and we strongly recommend using GPU for training.
Before compiling the code of our framework, you need to install the following development environment:
Please download our GRL framework repository first through git or directly download the compressed files:
git clone https://github.com/Jacklinkk/GRL_CAVs.gitThen enter the root directory of GRL_CAVs:
cd GRL_CAVsand please be sure to run the below commands from /path/to/GRL_CAVs.
The FLOW library will be firstly installed.
Firstly, enter the flow directory:
cd flowThen, create a conda environment from flow library.
The name of the conda environment is defined in "environment.yml" in the flow folder, you can change the name accordingly. Here, we choose GraphRL as the name of our environment.
conda env create -f environment.ymlActivate conda environment:
conda activate GraphRLInstall flow from source code:
python setup.py developSUMO simulation platform will be installed. Please make sure to run the below commands in the "GraphRL" virtual environment.
Install via pip, here we choose the 1.12.0 version of SUMO.
pip install eclipse-sumo==1.12.0Setting in Pycharm:
In order to adopt SUMO correctly, you need to define the environment variable of SUMO_HOME in Pycharm. The specific directory is:
/home/…/.conda/envs/GRL_CAVs/lib/python3.7/site-packages/sumoYou can define the pycharm environment variable through the following steps. Click "Run" in the menu bar, then click "Edit Configurations"; find the installation path of SUMO, and add the path in "Environment->Environment variables".
Setting in Ubuntu:
At first, run:
gedit ~/.bashrcthen copy the path name of SUMO_HOME to “~/.bashrc”:
export SUMO_HOME=“/home/…/.conda/envs/GraphRL/lib/python3.7/site-packages/sumo”Finally, run:
source ~/.bashrcPlease make sure to run the below commands in the "GraphRL" virtual environment.
Installation of Pytorch:
We use Pytorch version 1.11.0 for development under a specific version of CUDA and cudnn.
pip3 install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.htmlInstallation of pytorch geometric:
Pytorch geometric is a graph neural network (GNN) library upon Pytorch. You need to install the corresponding version of pytorch geometric according to the pytorch version you have installed.
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cu113.htmlThe flow folder is the root directory of the library after the FLOW library is installed through source code, including interface-related programs between the developed GRL algorithms and SUMO platform.
The Flow_Test folder includes the related programs of the test environment configuration. This folder includes test code for ring network, highway network, etc.. If this programs runs successfully, the environment configuration of the source code successful.
The programs in the GRL_Envs folder are used to define the environment configuration for mixed autonomy traffic constructed in our modular framework. Here we have constructed two mixed traffic scenarios, highway ramping scenario and Figure-Eight scenario.
A scenario is constructed by several python files described as follows:
You can construct your own traffic scenarios in GRL_Envs folder by referring to the above file structure.
The GRL_Experiment folder contains the programs for simulation configuration. Relative parameters for simulation are defined in each python file.
This folder contains two sub-folders: Exp_FigureEight folder and Exp_HighwayRamps folder. Each folder contains simulation files for different GRL algorithms for this traffic scenario. If you design a new scenario, or a new GRL algorithm, you need to create a new python program in this folder for simulation configuration.
It should be noted that in each python file, the model save path and load path need to be set and create the corresponding folders in advance.
The GRL folder is the core of the modular framework, which includes different GRL algorithms. It consists of two sub-folders: agent folder and common folder.
The agent folder contains several GRL agents for solving multi-agent decision-making problem in mixed autonomy traffic. We have divided the GRL algorithms into continuous and discrete algorithms, depending on the type of action space that the developed GRL algorithm can handle.
Each folder contains programs for several GRL algorithms with detailed parameter descriptions and comments. You can find detailed descriptions in the python files of each GRL algorithm for easy code reproduction and secondary development.
The common folder contains several generic programs of different GRL algorithms. The python files are described as follows:
You can design your own GRL algorithms in this folder as required.
The GRLNet folder contains the GRL neural network built in the pytorch environment. The networks are divided into continuous network and discrete network according to the categories of action space. The sub-folders are illustrated as follows:
You can modify the source code as needed or add your own neural network.
The GRL_utils folder contains basic functions such as model training and testing, data storage, and curve drawing. The files are illustrated as follows:
Before using these functions, please set the path for saving and reading the relevant data and curves. In addition, You need to select the corresponding "Train_and_Test" program according to the GRL algorithm to be verified.
The GRL_Simulation folder contains the main program to run the simulation of different traffic scenarios.
You can simply run python files in "/GRL_Simulation/main" in Pycharm to simulate the GRL algorithm, and observe the simulation process in SUMO platform. You can generate training plot such as reward curve.
If you want to verify other algorithms, you can develop the source code as needed under the GRL_Library folder. Don't forget to change the imported python script in "main.py", and define your own experiment file in the GRL_Experiment folder. In addition, you can also construct your own network in GRL_Net folder.
If you want to verify other traffic scenario, you can define a new scenario in GRL_Envs folder. You can refer to the documentation of SUMO and FLOW for more details.
To cite our publications, please cite our paper currently on arxiv, the library on which Graph_CAVs is based:
@article{liu2022graph,
title={Graph Reinforcement Learning Application to Co-operative Decision-Making in Mixed Autonomy Traffic: Framework, Survey, and Challenges},
author={Liu, Qi and Li, Xueyuan and Li, Zirui and Wu, Jingda and Du, Guodong and Gao, Xin and Yang, Fan and Yuan, Shihua},
journal={arXiv preprint arXiv:2211.03005},
year={2022}
}