Abstract
Different control methods for the autonomous navigation of robots have been designed during industry 4.0, several works use Simultaneous Localization And Mapping (SLAM) or path planning systems for trajectory tracking, however, there are different restrictions when it is required to avoid obstacles and reconfigure parameters in real-time. The present work shows an algorithm based on the use of Deep Q-Networks (DQN) and Reinforcement Learning, where the model is in charge of maximizing the reward while executing actions on the robot and in turn extracts information about its position and obstacles within the simulated environment. A series of experiments were carried out for the configuration of the algorithm whose results validate the operation of the network, showing that the robot learns through exploration, exploiting the knowledge learned from previous scenarios. Using a simulated environment allowed the DQN network to compute complex functions due to the randomness, leading to higher autonomous learning performance over other control methods.
Original language | English |
---|---|
Pages (from-to) | 543-551 |
Number of pages | 9 |
Journal | Procedia Computer Science |
Volume | 217 |
DOIs | |
State | Published - 2022 |
Event | 4th International Conference on Industry 4.0 and Smart Manufacturing, ISM 2022 - Linz, Austria Duration: 2 Nov 2022 → 4 Nov 2022 |
Bibliographical note
Funding Information:This work was financed by Universidad Tecnica de Ambato (UTA) and their Research and Development Department (DIDE) under project PFISEI31
Publisher Copyright:
© 2022 The Authors. Published by ELSEVIER B.V.
Keywords
- Navigation
- Obstacle Avoidance
- Reinforcement Learning
- Robotics