An Application of Reinforcement Learning to Aerobatic Helicopter Flight