UAV-Enabled Inspection System With No-Fly Zones: DRL-Based Joint Mobile Nest Scheduling and UAV Trajectory Design

Abstract

Unmanned aerial vehicle (UAV)-enabled inspection is regarded as a promising technology in the electricity system. This paper investigates a UAV-enabled inspection system in an urban environment with no-fly zones (NFZs), where the UAV flies to inspection points to capture images while constrained by limited onboard energy. The aim of this paper is to minimize the whole inspection time via joint optimization of the mobile nest’s scheduling and UAV trajectory while satisfying constraints related to energy maximization and avoiding NFZs. The formulated problem is a multivariate mixed non-convex optimization problem (due to the presence of NFZs), which makes it challenging to address with conventional techniques such as successive convex approximation and graph theory. To solve these issues, we first model the proposed problem as a Markov decision process (MDP). Then we propose an efficient two-step optimization method that involves optimizing the UAV trajectory with a modified multi-step dueling DDQN (MSD-DDQN) algorithm and planning the mobile nest’s path using the A* algorithm. Finally, the effectiveness of the proposed algorithm is verified by simulation experiments. In particular, the proposed method not only accelerates algorithm convergence but also reduces inspection time by 35 % compared to baseline methods.

Keywords

• UAV inspection
• UAV trajectory design
• mobile nest path planning
• deep reinforcement learning