The algorithm is also known as DQPID. From the article:
"Double Q-learning algorithm for mobile robot control"
As appears in Expert Systems with Applications.
Authors: Ignacio Carlucho - Mariano De Paula - Gerardo Acosta
- Ros indigo
- numpy
- python 2.7
- Pionner 3at
- Husky from clear path
- Ictiobot
- drone (using hector_quadrotor)
- gym inverted pendulum
python main.py
In main.py there is a variable called platform. By assigning to this variable the available robots in the robot dictionary, the algorithm will run it accordingly,
with the parameters configured in the dictionary.
By default it is set in 'pioneer_pi' wich is configured for running the pioneer robot in a simulated environment using gazebo.
Into the catkin workspace clone
git clone https://github.com/IgnacioCarlucho/amr-ros-config
then you can launch an example scenery as:
roslaunch amr_robots_gazebo empty_world.launch
Once the gazebo simulation is running you can then execute the algorithm by running:
python main.py
This simulation is speed up for doing faster trials, they can be slowed down using gazebo.
follow tutorial on: http://www.clearpathrobotics.com/assets/guides/husky/HuskyMove.html
and then launch with
roslaunch husky_gazebo husky_empty_world.launch
set
platform = 'husky_pi_random'
on main.py and then you can run the algorithm by doing:
python main.py
Clone to your catkin workspace the hector quadrotor
git clone https://github.com/tu-darmstadt-ros-pkg/hector_quadrotor
then you can launch an example:
roslaunch hector_quadrotor_gazebo quadrotor_empty_world.launch
Sorry this model is not available for the general domain.
- Incremental Q-learning strategy for adaptive PID control of mobile robots Carlucho et al. Link