Just a quick post, but finished up logging QPi data for Acrobot. Experimental settings were essentially the same as in hillcar (50 trajectories for pi, and pi_e, same distribution of time for qPi, all variables perturbed by N~(0, 0.01) noise).
A visualization from one of the logs:
I have to say, visualizing the data with Weka is very valuable; it gives a much better understanding of what is going on than you could ever get from just looking at log data. From the figure above, for example, its easy to tell that blue and green correspond to loading torque in one direction or the other and that red is the no-torque action.
So the question from here is what next? I could keep logging more data, or perhaps more of the math/theory part should be next?