3D Path Planning for UAV with Improved Double Deep Q-Network

Unmanned aerial vehicle (UAV) has been widely used in civil and military fields due to its advantages such as zero casualties, low cost and strong maneuverability. Path planning in 3D obstacle environment is one of the fundamental capabilities of UAV for mission performing. In this paper, we propose a 3D path planning algorithm to learn a target-driven end-to-end model based on an improved double deep Q-network (DQN), where a greedy exploration strategy is applied to accelerate learning. The model takes target and obstacle message as input, and moving command of UAV as output. It can realize path planning successfully for UAV in 3D complex environment. Besides, the experimental results show that improved double DQN has better convergence speed compared with DQN and double DQN.