Autonomous-Drifting

Autonomous Drifting using Reinforcement Learning

Installation

sudo ./setup_env.sh

cd fyp_ws

catkin_make

. devel/setup.bash (add source [full path to setup.bash] in your .bashrc)

roscd drift_car_env/scripts/

sudo pip install -e .

roscd drift_car/scripts/rl

sudo pip install -r requirements.txt

The first time you open Gazebo, it will download all models from the Gazebo servers, which may take some time. Run rosrun gazebo_ros gazebo to run Gazebo and install models.

Commands

To run	Command
ROS Core	`roscore`
Gazebo Simulator	`roslaunch drift_car_gazebo drift_car.launch`
Controller	`roslaunch drift_car_gazebo_control drift_car_control.launch`
Keyboard Teleop	`rosrun drift_car_gazebo_control teleop_gazebo.py`
Joystick Gazebo Controller	`rosrun drift_car_gazebo_control joystick_gazebo.py`
Joystick Car Controller	`rosrun drift_car_gazebo_control joystick_car.py`
Double Dueling Deep Q-Network	`rosrun drift_car main.py`

PILCO

Install MATLAB, enabled with Robotics System Toolbox.

Add src/drift_car/scripts/rl/modules and src/drift_car/scripts/rl/pilco to MATLAB path.

Start the bridge library with rosrun drift_car_env matlab_bridge.py.

To train - drift_car_learn.

To apply learned controller - applyController.

Car Model

To run using the Monster Truck, rosed drift_car_gazebo drift_car.launch and toggle the comments to load truck.xacro.urdf.

编辑分享