Optimización de trayectorias en plataformas robóticas móviles usando técnicas de inteligencia artificial

Abstract

As time progresses, the growth of the robotics industry is exponential and companies that seek to automate their processes do so to optimize resources, such as the time spent.

The use of artificial intelligence is one of the current solutions for the optimization of various processes, by allowing learning in supervised environments, where robotic instrumentation tends to minimize the margin of error in the future thanks to implementations of AI algorithms.

In the present project, a solution for the optimization of trajectories is exposed using as support an AI algorithm of reinforcement learning with neural networks implemented in the omnidirectional robotic platform of the KUKA youBot robot to move from one point to another avoiding obstacles presented in its path.

The AI algorithm used for learning is Deep Q Network (DQN), this algorithm consists of deep neural networks to maximize some notion of rewards in a cumulative way. whereby means of a Hokuyo lidar motion sensor, placed in the front part of the robotic platform, they are acquired. You sample data from an environment, which is processed in the algorithm to be recognized as collisions or rewards. As the rewards learned by the algorithm are greater. the possibility of collision with an obstacle decreases, moving the robotic platform towards an obstacle-free zone.

The programming language of this DON algorithm is based on Python 2, this language works together with ROS (robotic operating system) and allows to know, in an understandable way, how the execution of the movement is carried out through the publication and subscription to the topics corresponding to the robotic platform, thus facilitating the calibration of the parameters used in it.