Learning to fly by combining reinforcement learning with behavioural cloning