Controllers and agents

RLSimion features the following controllers and agents:

• Regular controllers, such as Proportional-Integrative-Derivative, Linear-Quadratic Regulator, or Variable-Speed Wind-Turbine controllers from Jonkman, Boukhezzar and Vidal[#Fernandez-Gauna2017]. Regular controllers are important as a reference for benchmarking. They can also be used to initialize RL agents[#Fernandez-Gauna2018].

• Linear VFA-based Q-function algorithms

  • SARSA
  • Q-Learning
  • Double Q-Learning

• Actor-Critic algorithms

  • CACLA
  • Regular gradient ascent
  • Incremental Natural Actor-Critic
  • Off-Policy Actor-Critic
  • Off-Policy Deterministic Actor-Critic
  • Besides, actors can be combined with Value function learning critics:
    • Temporal-Difference (lambda)
    • TDC(lambda)
    • True Online Temporal-Difference.

• Deep RL methods

  • DQN
  • Double-DQN
  • DDPG.