Assessing the Reality Gap of Robotic Simulations with Educational Purposes

摘要

The use of simulation and visualization tools to validate mechanical designs and programming for robots has become increasingly important. Being used in many scenarios, simulators have collaborated for research in machine learning, deep learning and training of professionals. Nowadays, with the Covid-19 pandemic and social distancing, simulators have great potential as a tool to help in the teaching of robotics. However, research indicates that the user may encounter differences between what happens in the simulator and in real life, this effect is called reality gap. This article aims to investigate the effects of this gap in an educational robotics simulator, named Virtual Robotics Toolkit (VRT), as well as its impacts on the testing of algorithms and to propose strategies to reduce the impact of this effect. Three batch tests were carried out on two different robots, both in the simulation and in the real world. By capturing data from sensors such as tachometer, infrared, gyroscope and a light sensor, it was possible to compare the performance in the two scenarios and their differences. Finally, the entire test field and models were made available along with videos of some tests.