Trajectory tracking control of an unmanned aerial vehicle with deep reinforcement learning for tasks inside the EAST

The robotic arms inside the EAST (Experimental Advanced Superconducting Tokamak) are bulky and slow, making them unable to efficiently complete remote handling tasks such as inspection and grasping. Miniature intelligent UAVs have the potential to assist in remote handling tasks. A key challenge is to achieve autonomous flight along a set trajectory within the EAST's vacuum vessel. This paper presents an autonomous UAV system with deep reinforcement learning for this purpose. The autonomous flight of a quadrotor UAV within the EAST was simulated using OpenAI Gym-style environment. To verify that the trained policy is transferable, we experimentally verified the trajectory tracking of UAVs along specific trajectories in real scenarios. The results show that our autonomous UAV system can complete trajectory-tracking flight tasks inside the EAST vacuum vessel.