Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
nuomizai 90f5078e57 | 1 year ago | |
---|---|---|
carla_perception | 1 year ago | |
carla_project/src | 1 year ago | |
config_files | 1 year ago | |
leaderboard | 1 year ago | |
nocrash_route | 1 year ago | |
nocrash_scenarios | 1 year ago | |
ppo_agent | 1 year ago | |
scenario_runner | 1 year ago | |
scripts | 1 year ago | |
statics | 1 year ago | |
utils | 1 year ago | |
.gitignore | 1 year ago | |
LICENSE | 1 year ago | |
README.md | 1 year ago | |
env_wrapper.py | 1 year ago | |
eval.py | 1 year ago | |
main.py | 1 year ago | |
requirements.txt | 1 year ago | |
simple_test.py | 1 year ago |
This is the code accompanying the paper: "CADRE: A Cascade Deep Reinforcement Learning Framework for Vision-based Autonomous Urban Driving" by Yinuo Zhao, Kun Wu, et. al, published at AAAI 2022.
CADRE is a novel CAscade Deep REinforcement learning framework to achieve model-free vision-based autonomous urban driving on CARLA benchmark. We also provide an environment wrapper for CARLA that is suitable for distributed DRL training.
Clone repo
git clone https://github.com/BIT-MCS/Cadre.git
cd Cadre
Create conda virtue environment and install dependent packages
conda create -n cadre python=3.7
conda activate cadre
pip install -r requirements.txt
Download the trained perception model from Google Driver under carla_perception/
Download the Carla 0.9.10 server from official website.
To quickly test the installation, we provide a simple script . To run this example, you need to first start the server and then start the client.
To start server, run the script under scripts/start_server.sh
. Make sure to replace the CARLA_ROOT
to your own directory.
To start training client, change the command in we provide a training script under scripts/simple_test.sh
, you need to change the CARLA_ROOT
and CHALLENGE_DIR
.
If the installation is successful, then you will see the following two windows on your computer.
To start server, run the script under scripts/start_server.sh
. Make sure to replace the CARLA_ROOT
to your own directory.
To start training client, we provide a training script under scripts/main.sh
, you need to change the CARLA_ROOT
and CHALLENGE_DIR
. The hyperparamters are configured under /config_files/agent_config.py
. We recommend you to change the hyperparameter num_processes
to 4
in order to get a more stable policy.
Models and log files are saved under result/
After the training process finished, we recommend you to use the script under scripts/kill_server.sh
to kill all servers running on the server.
#!/bin/bash
export CARLA_ROOT=[PATH TO YOUR LOCAL DIRECTORY WITH CaraUE4.sh]
export CHALLENGE_DIR=[PATH TO WHERE]
export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI/carla
export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI/carla/dist/carla-0.9.10-py3.7-linux-x86_64.egg # 0.9.10
export PYTHONPATH=$PYTHONPATH:$CHALLENGE_DIR/leaderboard
export PYTHONPATH=$PYTHONPATH:$CHALLENGE_DIR/scenario_runner
export HAS_DISPLAY='0'
python main.py
To evaluate the models, please refer to the scripts under scrips/eval.sh
. Please set the pretrained_path
and load_episode
in eval_cfg under config_files/eval_agent_config.py
. It is recommended to use 8 models from different episodes for a more stable policy. You can also change the amount of vehicles/pedestrians and routes in this config files.
After evaluation ends, you can find the evaluation results under ${pretrained_path}/eval/eval_completion_ratio.csv
.
This work was supported in part by Shanghai Pujiang Program and the National Research and Development Program of China (No.
2019YQ1700).
If you have any question, please email ynzhao@bit.edu.cn
/ linda.chao.007@gmail.com
.
This project includes some implementations of DANet and the overall evaluation framework follows CARLA secnario runner, carla_project (no license) and leaderboard.
If you are interested in our work, please cite our paper as
@inproceedings{zhao2022cadre,
author = {Zhao, Yinuo and Wu, Kun and Xu, Zhiyuan and Che, Zhengping and Lu, Qi and Tang, Jian and Liu, Chi Harold},
title = {CADRE: A Cascade Deep Reinforcement Learning Framework for Vision-based Autonomous Urban Driving},
booktitle = {Association for the Advancement of Artificial Intelligence (AAAI)},
year = {2022},
}
No Description
Python other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》