Abstract: Reinforcement learning (RL) is a promising approach for achieving agile flight control of quadrotors. However, the low sample efficiency of reinforcement learning and the challenge of ...