Hybrid Multi-Target 3D Path Planning Algorithm for Unmanned Aerial Vehicle

To address the challenges of multi-target sequence decision-making and the complexity of 3D spatial search in multitarget inspection tasks for unmanned aerial vehicle(UAV), this study proposes a hybrid path planning method combining deep reinforcement learning with terrain-adaptive sampling. First, the multi-target path planning problem for UAV is modeled as a Traveling Salesman Problem (TSP), employing a multi-target sequential planning approach based on the Hybrid PreferenceGuided Local Search Reinforcement (HPLSR). To reduce the computational cost of the path cost matrix, a sparse graph structure is constructed using Delaunay triangulation to achieve efficient edge weight estimation. After obtaining the target visit sequence, an improved Rapidly-exploring Random Tree algorithm, termed Terrain-Adaptive Informed RRT∗ (TAI-RRT*), is proposed to generate collision-free local trajectories between consecutive target points. Through slope-adaptive sampling, elliptic shrinkage sampling, and KDTree nearest neighbor acceleration mechanisms, efficient obstacle avoidance and local path smoothing in complex terrain are achieved. Experimental results show that this method achieves higher computational efficiency and a shorter overall path, and exhibits good scalability.