A Lightweight Hybrid Optimization Framework With DQN‐Based Local Correction for 3D UAV Path Planning

ABSTRACT This work addresses the problem of UAV path planning in complex three‐dimensional urban environments using a computationally efficient hybrid optimization framework. The proposed method combines a PSO‐like swarm motion update with exploratory characteristics inherited selectively from the Chimp Optimization Algorithm, creating a simple yet effective global search mechanism. For enhancing its local adaptability, it incorporates a lightweight Deep Q‐Network as a corrective component that offers limited directional corrections only in unsafe, ambiguous, or locally suboptimal situations rather than acting as a primary control policy. Under this architecture, the swarm‐based mechanism contributes to stable convergence toward the goal, while the DQN helps resolve local navigation challenges such as collision‐prone and congested areas. The proposed method is tested in several dense three‐dimensional environments with narrow passages and complex obstacle distributions. Experimental results demonstrate that the hybrid model consistently outperforms the benchmark metaheuristic methods in terms of collision avoidance, path smoothness, and trajectory quality, while maintaining low computational overhead suitable for UAV platforms with limited onboard resources.