Deep reinforcement learning implementation for collision avoidance of mobile robot. In this project, DDPG, TD3 and SAC are adopted to realize short-distance navigation for turtlebot3. The user can easily train a practical path planner for real mobile robot without any prior knowledge.
| interface | version |
|---|---|
| ubuntu | 18.04 |
| ros | melodic |
| pytorch | 1.4.0 |
ROS melodic
refer to here
Gazebo 9
sudo apt install ros-melodic-gazebo-*
download models.zip and unpack under ~/.gazebo/.
dependancy
sudo apt install ros-melodic-xacro
Anaconda
refer to here
virtual environment
create env
conda create -n my_env python=2.7
conda activate my_env
update pip
pip install --upgrade pip
install dependence
pip install requirements.txt
Building pkg
create workspace
mkdir catkin_ws
cd catkin_ws
mkdir src
cd src
catkin_init_workspace
Download this project and put it into src.
cd ..
catkin_make
source devel/setup.bash
Run Samples
# open gazebo
roslaunch turbot_rl setup.launch
# start training
rosrun turbot_rl training_node.py