In this project we aim to adap Openai Reinforcement Learning platform to the Moveo robot, a low-cost, 3D printed robotic manipulator. This research will widen the field of application of Moveo, concretely for robotics surgeries and also will make the research in robotics available and affordable for a large portion of the research community. Using only open-source technologies we experiment with the different required modules and pose the global architecture to train robots focused on manipulation ...
In this project we aim to adap Openai Reinforcement Learning platform to the Moveo robot, a low-cost, 3D printed robotic manipulator. This research will widen the field of application of Moveo, concretely for robotics surgeries and also will make the research in robotics available and affordable for a large portion of the research community. Using only open-source technologies we experiment with the different required modules and pose the global architecture to train robots focused on manipulation with Reinforcement Learning techniques. This project is constructed by the combination of three main components: A real robot, a simulation platform and the Reinforcement Learning module. These components are interconnected using the robotic middleware ROS which will provide the corresponding communication interface. As a realistic simulation platform the Gazebo software is used. The Reinforcement Learning module is the OpenAI Gym which interfaces with ROS based robots through the package Openai-ROS. In this document we provide a detailed analysis of the functioning of the package Openai-ROS besides a robotic model in Gazebo with the corresponding functionalities for it to be used along with Openai-ROS and perform the training.
+
