Aprendizagem por reforço como técnica de controle para o problema do pêndulo invertido
This work aims to utilize some Machine Learning algorithms to solve the inverted pendulum problem with one degree of freedom and compare the outcomes with the pole placement method. In this way, three reinforcement learning algorithms were implemented in python: HillClimbing with adaptive noise scal...
| Autor principal: | Krul, Alexandre Mendonça |
|---|---|
| Grau: | Trabalho de Conclusão de Curso |
| Idioma: | por |
| Publicado em: |
Brasil
2021
|
| Assuntos: | |
| Acesso em linha: |
http://riu.ufam.edu.br/handle/prefix/5849 |
| Resumo: |
|---|
|
This work aims to utilize some Machine Learning algorithms to solve the inverted pendulum problem with one degree of freedom and compare the outcomes with the pole placement method. In this way, three reinforcement learning algorithms were implemented in python: HillClimbing with adaptive noise scaling, REINFORCE and DeepQNetworks and their results were compared with the state space pole placement method, also implemented in this work. The results showed that all the methodwere able to balance the pendulum. The ITAE errors with relation to the vertical angular position for the methods HillClimbing, REINFORCE, DeepQNetworks and Pole Placement were 410, 55, 50 and 52, respectively. |