Trabalho de Conclusão de Curso de Graduação
Comparação de algoritmos de aprendizagem por reforço profundo na navegação do robô móvel e desvio de trajetória
Fecha
2022-09-23Registro en:
MORAES, L. D. de. Comparação de algoritmos de aprendizagem por reforço profundo na navegação do robô móvel e desvio de trajetória. 2022. 69 p. Trabalho de Conclusão de Curso (Graduação em Engenharia de Controle e Automação) - Universidade Federal de Santa Maria, Santa Maria, RS, 2022.
Autor
Moraes, Linda Dotto de
Institución
Resumen
This work presents two Deep Reinforcement Learning (Deep-RL) approaches to enhance the problem of mapless navigation for a terrestrial mobile robot. The methodology focus on comparing a Deep-RL technique based on the Deep Q-Network (DQN) algorithm with a second one based on the Double Deep Q-Network (DDQN) algorithm. As sensor, 24 laser range findings samples were used and the relative position and angle of the agent to the target were used as information for the agent, which provide the actions as velocities for the robot. By using a low-dimensional sensing structure of learning such as the one proposed, it is shown that it is possible to successfully train an agent to perform navigation-related tasks and obstacle avoidance without the use of complex sensing information, like in image-based approaches. The proposed methodology was successfully used in three distinct real and simulated environments. Overall, it was shown that Double Deep structures further enhance the problem for the navigation of mobile robots when compared to the ones with simple Q structures. This work presents two Deep Reinforcement Learning (Deep-RL) approaches to enhance the problem of mapless navigation for a terrestrial mobile robot. The methodology focus on comparing a Deep-RL technique based on the Deep Q-Network (DQN) algorithm with a second one based on the Double Deep Q-Network (DDQN) algorithm. As sensor, 24 laser range findings samples were used and the relative position and angle of the agent to the target were used as information for the agent, which provide the actions as velocities for the robot. By using a low-dimensional sensing structure of learning such as the one proposed, it is shown that it is possible to successfully train an agent to perform navigation-related tasks and obstacle avoidance without the use of complex sensing information, like in image-based approaches. The proposed methodology was successfully used in three distinct real and simulated environments. Overall, it was shown that Double Deep structures further enhance the problem for the navigation of mobile robots when compared to the ones with simple Q structures.