Stagnation-Aware responsive differential evolution for high-dimensional and constrained UAV path planning

Unmanned aerial vehicles (UAVs) require flyable paths through constrained static two- and three-dimensional environments, while high-dimensional black-box optimization challenges population-based search on multimodal landscapes. Differential evolution (DE) is suitable for both settings, but success-history adaptation records which parameters worked without using population-level progress as a common timing signal for mutation and crossover. This paper proposes Stagnation-Aware Responsive Differential Evolution (SARDE), a training-free method built around population-progress state feedback. A smoothed median-improvement signal jointly selects the mutation mode and crossover rate, retaining explorative rand/1 during adequate progress and activating exploitative current-to-pbest/1 when progress falls below a threshold. A replaced-parent archive supplies historical donors to both modes, and greedy selection generates the next state, forming an archive-supported donor loop. These operations reuse objective values already produced by DE and require no additional function evaluations. The measured median within-problem wall-time ratios to DE are 1.125 on CEC2017 at 30D and 1.021 on eight UAV scenarios. Under a controlled common-population protocol, SARDE was compared with eight algorithms on CEC2017 at 10D, 30D, 50D, and 100D, CEC2022 at 10D and 20D, and eight static UAV scenarios comprising four 2D and four 3D cases at D = 14 –45. SARDE achieves the leading Friedman average rank at 50D and 100D on CEC2017 and across the UAV set.