ROSDevCon2018 Day2: Using OpenAI with ROS
The Construct Robotics Institute 59:45
1,540 views · 18 likes Watch on YouTube ↗
ROS Developers Conference 2019 Registration is Now Open: http://rosdevcon.com
Speaker: Ricardo Téllez | CEO of The Construct
http://www.theconstructsim.com/ros-developers-online-conference-2018-rdc-worldwide/ros-developers-conference-speaker-the-construct/
ABSTRACT:
A FRAMEWORK FOR USING OPENAI WITH ROS
We are going to present the openai_ros package and show a live demo on the screen of how it works using Gazebo simulations.
The openai_ros package integrates the OpenAI gym environment into ROS, allowing to use all its algorithms and classes inside ROS environments, in a very simple and easy way, and removing the burden to integrate the OpenAI environment into the Gazebo robots.
With the openai_ros package we have achieved to compartmentalize the use of reinforcement learning (RL) algorithms for training ROS robots to three degrees of implication, depending on which level the roboticist needs to take part.
The definition of those levels allows to simplify the use of OpenAI infrastructure with ROS robots and provide a simple structure for modifying, depending on the needs of the roboticist.
The spectators are going to learn:
0- How the whole structure of the openai_ros package works.
1- Level 0 example: use openai_ros package to use Qlearn algorithm to train a Cartpole Gazebo simulation, and then change to train it again with Sarsa algorithm.
#Robotics #Conference #ROSDevCon18
Speaker: Ricardo Téllez | CEO of The Construct
http://www.theconstructsim.com/ros-developers-online-conference-2018-rdc-worldwide/ros-developers-conference-speaker-the-construct/
ABSTRACT:
A FRAMEWORK FOR USING OPENAI WITH ROS
We are going to present the openai_ros package and show a live demo on the screen of how it works using Gazebo simulations.
The openai_ros package integrates the OpenAI gym environment into ROS, allowing to use all its algorithms and classes inside ROS environments, in a very simple and easy way, and removing the burden to integrate the OpenAI environment into the Gazebo robots.
With the openai_ros package we have achieved to compartmentalize the use of reinforcement learning (RL) algorithms for training ROS robots to three degrees of implication, depending on which level the roboticist needs to take part.
The definition of those levels allows to simplify the use of OpenAI infrastructure with ROS robots and provide a simple structure for modifying, depending on the needs of the roboticist.
The spectators are going to learn:
0- How the whole structure of the openai_ros package works.
1- Level 0 example: use openai_ros package to use Qlearn algorithm to train a Cartpole Gazebo simulation, and then change to train it again with Sarsa algorithm.
#Robotics #Conference #ROSDevCon18
Category (YouTube): Science & Technology
Playback is via YouTube's official embedded player. Data from YouTube; Exumo is not affiliated with YouTube.