Exploration in Model-based Reinforcement Learning by Empirically Estimating Learning Progress

Formal exploration approaches in model-based reinforcement learning estimate the accuracy of the currently learned model without consideration of the empirical prediction error.