img_env

A customized grid map-based navigation simulation platform following the Gym API.

• Very fast, based on C++, customized for robot navigation.
• Accurate, not perform physical simulation and time simulation.
• Parallelization，support multi-scene and multi-robot.
• Pedestrian simulation，contains pedestrian simulation based on Pedsim and ORCA.
• Gym API, easy to use in conjunction with other reinforcement learning frameworks.
• Grid_map output， directly outputs grid_map or raw laser data.
• Parameterized configuration, easily configure environment information via yaml file, including robot shape and size, pedestrian strategy, environmental map, and more.
• Very lightweight, only depends on OpenCV for visualization. 200 robots and 200 obstacles have been simulated.

Quick Start

Recommended system environment: Ubuntu20 + ROS noetic + python3

sudo apt-get install libedit-dev

git clone [email protected]:DRL-Navigation/img_env.git

cd img_env

compile:

catkin_make --only-pkg-with-deps img_env

source

echo "source `pwd`/devel/setup.bash" >> ~/.bashrc

config yaml file

Enter envs/cfg to configure the yaml to be run. Note, set env_num parameter to run nodes simultaneously.

generate a launch file

python create_launch.py test envs/cfg/test.yaml

the first param is task name, other params is yaml file，if you want to run different yaml files together,

python create_launch.py test A.yaml B.yaml C.yaml

the generated file are located at src/img_env/tmp

start up ROS

open up a new terminal

roslaunch img_env test.launch

run python

python env_test.py

Explanation

State Instance

envs/state/state.py following valuse are all type of numpy, and the first dim means robot_num.

    vector_states,  [robot_num, state_dim] vector，state_dim=3 which means the x,y position and yaw to target
    sensor_maps, [robot_num, batch, 48, 48] egocentric sensor map.
    is_collisions, [robot_num] 0 means no collision, 1 means get collision with static obstacles, 2 means get collision with pedestrians, 3 means get collision with other robots.
    is_arrives, [robot_num] 1 means reach the goal while 0 is not.
    lasers, [robot_num, laser_range_total] original laser scan data.
    ped_vector_states, [robot_num, n] n = 1 + vector_info_number * max_ped_number, where 1 means the real ped number
    ped_maps, [robot_num, 3, 48, 48] egocentric 3-channel pedestrians map, which means x，y velocity and position of the nearby pedestrains.
    step_ds, [robot_num] the distance of getting close to the target each step.
    ped_min_dists, [robot_num] the closest distance of nearby pedestrains, used in reward function.

Wrapper

To modify reward, action space and other operations, please wrapper the package according to the Gym API.

!!! NOTE: The order of filling in wrappers in yaml is very important. The first one is enclosure in the innermost layer and executed first, and the last one is enclosured in the outermost layer and executed last. Imagine preorder traversal, postorder traversal of the binary tree.

Parameters

All parameters are set in yaml file, they are sent to C++

pedestrians model

the 3 regular behaviour of pedestrains:

pedsim, rvo, ervo, thanks to their codes are all opensourced, we converge them and put in 3rd_party

• pedsim is controled by SFM

• rvo is controled by orca

• ervo denotes the emitional version of rvo

to change pedestrain behaviour, set yaml file ped_sim:type with one of [ pedscene,rvoscene,ervoscene]

GUI

parameters server

# close GUI
rosparam set /image_ped_circle0/show_gui false
# open GUI
rosparam set /image_ped_circle0/show_gui true
# `image_ped_circle0` is the node name, which is configured in the yaml file, `env_name`+env_id
# type `rosnode list` show all nodes

please cite:

@Article{chen2020distributed,
title = {Distributed Non-Communicating Multi-Robot Collision Avoidance via Map-Based Deep Reinforcement Learning},
author = {Chen, Guangda and Yao, Shunyi and Ma, Jun and Pan, Lifan and Chen, Yu'an and Xu, Pei and Ji, Jianmin and Chen, Xiaoping},
journal = {Sensors},
volume = {20},
number = {17},
pages = {4836},
year = {2020},
publisher = {Multidisciplinary Digital Publishing Institute},
doi = {10.3390/s20174836},
url = {https://www.mdpi.com/1424-8220/20/17/4836}
};

@inproceedings{yao2021crowd,
  title={Crowd-Aware Robot Navigation for Pedestrians with Multiple Collision Avoidance Strategies via Map-based Deep Reinforcement Learning},
  author={Yao, Shunyi and Chen, Guangda and Qiu, Quecheng and Ma, Jun and Chen, Xiaoping and Ji, Jianmin},
  booktitle={2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={8144--8150},
  year={2021},
  organization={IEEE}
}

More DRL navigation work

@article{qiu2023training,
  title={Training A Non-Cooperator to Identify Vulnerabilities and Improve Robustness for Robot Navigation},
  author={Qiu, Quecheng and Zhang, Xuyang and Yao, Shunyi and Chen, Guangda and Hua, Bei and Ji, Jianmin and others},
  journal={IEEE Robotics and Automation Letters},
  year={2023},
  publisher={IEEE}
}

Reproduce our works

chen2020distributed Sensors-20

git clone [email protected]:DRL-Navigation/DDRL4NAV.git --recurse-submodules

cd DDRL4NAV/USTC_lab/env/img_env
catkin_make --only-pkg-with-deps img_env
echo "source `pwd`/devel/setup.bash" >> ~/.bashrc
python create_launch.py sensor_map envs/cfg/sensor_nap.yaml 
roslaunch img_env sensor_map.launch

# open a new  terminal
cd DDRL4NAV/sh
bash start_redis.sh
bash start.sh config/config_sensor_map.sh
bash tfboard.sh config/config_sensor_map.sh

# show gui
rosparam set /sensor_map0/show_gui true

yao2021crowd IROS-21

git clone [email protected]:DRL-Navigation/DDRL4NAV.git --recurse-submodules

cd DDRL4NAV/USTC_lab/env/img_env
catkin_make --only-pkg-with-deps img_env
echo "source `pwd`/devel/setup.bash" >> ~/.bashrc
python create_launch.py ped_map envs/cfg/ped_nap.yaml 
roslaunch img_env ped_map.launch

# open a new  terminal
cd DDRL4NAV/sh
bash start_redis.sh
bash start.sh config/config_ped_map.sh
bash tfboard.sh config/config_ped_map.sh

# show gui
rosparam set /ped_map0/show_gui true