One of the main limitations of current exoplanet imaging instruments is the starlight that leaks through because our adaptive optics systems can not keep up with the quickly changing atmosphere. This results in the "wind-driven halo", a butterfly-like pattern of residual starlight in the direction of the wind. Another issue is telescope vibrations, which leads to additional leakage of starlight. These problems can be mitigated with predictive control. During my first master project, we investigated if it is possible to use Reinforcement Learning for this. Reinforcement Learning is a subfield of Artificial Intelligence which takes inspiration from the way humans and animals learn. We make use of the Recurrent Deterministic Policy Gradient algorithm and test this in simulations and in the lab. We show that this can lead up to more than an order of magnitude improvement in contrast.