Table 1.
Core features and experimental applications of the AirSim and UAVDT datasets.
Fig 1.
The overall architecture of the PGD model: The raw input data (such as 3D terrain point clouds, drone states, and inspection point coordinates) undergoes feature compression, path generation, and policy optimization to output the globally optimal path that satisfies safety, efficiency, and task requirements.
Fig 2.
Comparison of UAV path planning models under the UAVDT and AirSim datasets in terms of Pl, Cr, Tc, Fr, Se, and Ge metrics.
Table 2.
Experimental environment configuration, model hyperparameters, and physical constraints.
Table 3.
Comparison of experimental results: PGD model vs. baseline models on UAVDT and AirSim datasets.
Table 4.
Ablation study: Performance comparison between PGD and models with removed modules on UAVDT and AirSim datasets.
Table 5.
Performance comparison between transformer and common feature compression methods on UAVDT and AirSim datasets.
Fig 3.
Visualization of the performance degradation in the 6 core metrics (Pl, Cr, Tc, Fr, Se, Ge) of the PGD model when removing core modules (transformer/GAN/DDPG).
Table 6.
PGD model cross-dataset generalization test results.
(Training set: AirSim canyon scene; test set: UAVDT suburban scene.)
Fig 4.
Comparison of reward value changes with the number of episodes during training for different path planning algorithms.
Fig 5.
PGD model generalization verification: Comparison of elevation - obstacle proximity distributions between AirSim and UAVDT.