Laboratory Experiments of Model-based Reinforcement Learning for Adaptive Optics Control