BahiaRT Setplays Collecting Toolkit and BahiaRT Gym

Abstract

One of the challenges in the research using machine learning is the requirement to build realistic datasets using common sense knowledge from domain experts. The BahiaRT Setplays Collecting Toolkit is a software that supports research where soccer fans or experts can watch robots playing soccer and capture situations where they want to demonstrate a better setplay to that robot team. All demonstrations were gathered in a dataset used to feed a reinforcement learning mechanism that produced a control policy for setplays selection in a robot soccer team. This software impacts many research areas such as autonomous vehicles, unmanned aerial vehicles, dataset organization, reinforcement learning, and learning from demonstration. The BahiaRT Gym is an open-source tool designed to integrate the OpenAI Gym toolkit with the RoboCup Soccer Simulation 3D server(rcssserver3d), in order to make available an easier way to create training environments for the soccer teams while also facilitating the development of other RoboCup leagues’s environments as well.