Simulation system for learning of social rules using gazebo and openai gym

Abstract

With the increasing development of service robotics and the increasing appearance of social robots, showing respect for social norms becomes crucial in the integration of these robots in environments with human beings. In recent years, research related to social norms in robotics has moved from model-based approaches to learning-based approaches. Such approaches require the realization of several experiments that are often done in a robot and a real environment. This work proposes the development of a simulation system for social robots based on ROS, Gazebo and OpenAI with the implementation of Proxemics and Social Force Model. With this system, it will be possible to accelerate the learning of social rules among other capabilities, as well as to enable research in this area to be carried out even without having a physical robot.

Test Framework for a Virtual Competition Testbed

“…In an effort to facilitate research in the field of social robotics and specifically the learning of social norms (e.g. proxemics), Pimentel and Aquino-Junior [2] have developed a virtual environment with a scoring mechanism using ROS, Gazebo and OpenAI. Most recently, SEAN by Nathan Tsoi et al [3] developed a virtual world with the focus on the development and evaluation of algorithms for social navigation (dynamic conditions such as simulated pedestrians, cars, etc.…”