Learning to Adapt: Meta-Learning for Model-Based Control