Navigation robot training with Deep Q-Learning monitored by Digital Twin

Abstract

This paper aims to present adherence to the Deep Q-learning algorithm applied to the movement and execution of tasks in a vehicular navigation robot. The robot's objective was to transport parts through a delimited environment. A decision system was developed according to the Deep Q-learning algorithm with an artificial neural network that received sensor data as input and enabled autonomous navigation in an environment with different obstacles. This article will also present the application process and experiments with the DQN algorithm, finally showing the results of the learning process and the use of the concept of digital twins to monitor the movement and signals sent by the robot to a Cloud service, which allows the visualization of navigation through augmented reality.