Benchmarking RL Agents

Hey reddit,

I am currently learning about reinforcement learning agents and Python. I am at the point where I would like to explore benchmarking between various agents on a set of problems(mostly pertaining to mountain car, pole balancing and mazes) is there a package that lets me do this easily?

I have heard of PyBrain and MMLF as well as RLGlue but digging around is confusing me quite a bit.

What I have so far is: some agents coded in python(Numpy) not using any Machine learning libraries. I have created some quick tests to run the agents on with XOR tasks

but now I want to be able to "plugin" my agents with some other tasks, such as the above. I assume I will need a bridge or an interface but the trick is how I can connect the task coded using the pyBrain library to my agent. I think RL Glue is supposed to take care of this but the documentation is scarce on using PyBrain tasks with it.

Any ideas or suggests for some other packages? They must be compatible with Python.

Thanks