Modelling of novices control skills with machine learning

We report an empirical study on the application of machine learning to the modelling of novice controllers’ skills in balancing a pole (inverted pendulum) on top of a cart. Results are presented on the predictive power of the models, and the extent to which they were tailored to each controller. The behaviour of the participants in the study and the behaviour of an interpreter executing their models are compared with respect to the amount of time they were able to keep the pole and cart under control, the degree of stability achieved, and the conditions of failure. We discuss the results of the study, the limitations of the methodology in relation to learner modelling, and we point out future directions of research.