An Easy to Use Deep Reinforcement Learning Library for AI Mobile Robots in Isaac Sim
Maximiliano Rojas, Gabriel Hermosilla, Daniel Yunge, Gonzalo Farías
Abstract
The use of mobile robots for personal and industrial uses is becoming popular. Currently, many robot simulators with high-graphical capabilities can be used by engineering to develop and test these robots such as Isaac Sim. However, using that simulator to train mobile robots with the deep reinforcement learning paradigm can be very difficult and time-consuming if one wants to develop a custom experiment, requiring an understanding of several libraries and APIs to use them together correctly. The proposed work aims to create a library that conceals configuration problems in creating robots, environments, and training scenarios, reducing the time dedicated to code. Every developed method is equivalent to sixty-five lines of code at maximum and five at minimum. That brings time saving in simulated experiments and data collection, thus reducing the time to produce and test viable algorithms for robots in the industry or academy.