Hi!

We developed a C++ class frame work for reinforcement learning. Our Motivation is to animate students, people of research and other interested people to use and play with reinforcement learning algorithm. You can easily generate your own learning problem, and test the different algorithms on them.
The most difficult thing on reinforcement learning is to find the right parameters, state discretizations and reward models for the algorithm. So a tool is needed which provides you a comfortable framework for testing several state discretizations, reward models and algorithms without much effort. By now we haven't found such a tool, so we developed it, because we believe such a tool would be very convenient for research.
We are just 2 students who developed the toolbox as a Project for university. It was quite a lot of effort but we think we did a good work, but we want to know some opinions from experts. It would be great if everybody who is interested in, tests the toolbox or at least looks at out our framework. You can visit our homepage at: www.igi.tu-graz.ac.at/ril-toolbox. Thanks in advance for many replies :)

mfG Gerhard Neumann

Hi,

I already saw your toolbox and it seems very interesting.

I am looking more precisely to an already made control software for humanoid robot that use reinforcement learning. I want it with simulation tool in order to test this robot in a realistic house environment. The model could be : a robot with a RL control and linked to a human operator who gives recompense or punishment.

My idea is to build or to test a system that can learn all possible useful actions in a house (dress up a table, prepare a meal, clean up house or simply take something and put it somewhere appropriatly, ...).

Do you know if something like this exists ?