Título |
Regulation of a Van der Pol oscillator using reinforcement learning |
Tipo |
Congreso |
Sub-tipo |
Memoria |
Descripción |
9th International Congress on Telematics and Computing, WITCOM 2020 |
Resumen |
In this work, we propose a reinforcement learning-based methodology for the regulation problem of a Van der Pol oscillator with an actuator subject to constraints. We use two neural networks, one who learns an approximation of the cost given a state, and one that learns the controller output. We employ a classic PID controller with compensation as base policy in a rollout scheme. This policy is further improved by a neural network trained on trajectories from random initial states. The results show that the resulting control policy reduces the cost for a minimal energy trajectory given an initial state. |
Observaciones |
DOI 978-303062553-5
Communications in Computer and Information Science |
Lugar |
Puerto Vallarta |
País |
Mexico |
No. de páginas |
281-296 |
Vol. / Cap. |
v. 1280 |
Inicio |
2020-11-02 |
Fin |
2020-11-06 |
ISBN/ISSN |
|