Neurocomputing-RL-experiments

De Grupo de Inteligencia Computacional (GIC)

Reinforcement Learning has recently gained a lot of interest in the research community as an adaptive alternative to traditional control systems. Nevertheless, few experimental comparisons on realistic control problems can be found in the literature. We have conducted extensive experiments using an Actor-Critic architecture with continuous states and spaces on three different control problems: airplane pitch control, underwater-vehicle control and variable-speed wind-turbine. We specifically compare: -three different conditioning methods -different degrees of coarseness in the representation of the functions being learned -different policy evaluation methods

The complete experimental setup can be found in the following paper:

Borja Fernandez-Gauna;Juan Luis Osa;Manuel Graña

   Experiments of Conditioned Reinforcement Learning in Continuous Space Control Tasks
   Neurocomputing

And the source code used in the experiments can be downloaded from GitHub (https://github.com/borjafdezgauna/RLSimion/tree/b15fe7335daf3279da8f051f5031184e53d1184b).

That version of the software lacks an appropriate documentation and manual of use because it was intended for internal use only. We are currently working hard on improving the user interface for a more intuitive use, but right now, we would like to encourage interested people to contact us for guidance on the code.