Skip to content

My attempt to use reinforcement learning in ROS to teach a monoped robot how to balance itself by hopping . Created a custom gym environment and created the Q-learning agent to operate on the gym environment.

Notifications You must be signed in to change notification settings

aihoque2/QLearnHopper

Repository files navigation

IlliniHopper

Behold my IlliniHopper! Hail to the orange, Hail to the blue...

Following the constructsim tutorial, I decided to create my own monoped robot in order to teach myself more about ROS and using OpenAI in a Gazebo Environment

I used the monoped robot from ETH Zurich's xpp package and tried to teach it to hop with by using Q-learning. The algorithm did not converge, but the result was quite interesting:

Alt Text

Requirements:

ROS-melodic Gazebo

to run please use the following commands:

catkin_make

to launch the monoped:

roslaunch my_legged_robots_sims main.launch

to launch the Qlearning:

roslaunch my_hopper_training main.launch

The node graph (or rqt_graph):

About

My attempt to use reinforcement learning in ROS to teach a monoped robot how to balance itself by hopping . Created a custom gym environment and created the Q-learning agent to operate on the gym environment.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published