One of the prime science cases of the next generations of high contrast imaging instruments on ground-based telescopes is the direct imaging of potentially habitable exoplanets. In order to reach the high contrast needed, the instruments are equipped an extreme adaptive optics (XAO) systems. The XAO systems will control thousands of actuators from kilohertz to several kilohertz at a framerate. Most of the habitable exoplanets are located at small angular separation from the host star, where the XAO control might be the limiting factor in the contrast. An up-and-coming field of research aimed at improving AO control methods is the application of fully data-driven control methods such as Reinforcement learning (RL). RL is an active branch of machine learning that aims to learn a control task via interaction with the environment and hence can be seen as an automated approach for AO control. In particular, model-based reinforcement learning (MBRL) has been shown to cope with both temporal and misregistration errors. Moreover, MBRL is shown to learn continuously on timescales of some seconds and automatically adjust to changing conditions. This work shows advances in the practical implementation of MBRL in a lab setup and simulation environments.