Cross domain constraint clustering guides reinforcement enhanced multi objective particle swarm optimization for UAV path planning under multiple threat constraints

UAV path planning in three-dimensional threat fields requires four conflicting objectives to be satisfied at once: path tortuosity, threat exposure, altitude deviation, and trajectory smoothness. Classical swarm optimizers converge prematurely under dense threats. Structured domain weighting is therefore required. Cross-domain Constraint Clustering with Reinforcement-Guided Multi-Objective Particle Swarm Optimization (XC-RL-MOPSO) is proposed. Q-learning pre-trains a feasible reference trajectory across 2000 episodes. This trajectory biases early swarm search toward low-exposure regions. Four objectives enter Pareto dominance. A fifth, reference adherence, is retained as a convergence diagnostic only. Objectives are clustered into three domains. Safety receives 0.40. Control receives 0.35. Efficiency receives 0.25. Weights remain static. A six-variant ablation is conducted across 15 replicates. Results indicate that domain-clustering removal degrades hypervolume by 21.0% ( p < 0.001). Hence domain clustering is the primary hypervolume driver. RL guidance removal raises hypervolume by 37.6%; guidance narrows Pareto spread while improving convergence and proximity. Static weighting exceeds adaptive scheduling by 11.1% ( p = 0.028). Comparative evaluation spans 30 runs against Standard-MOPSO, NSGA-II, and MOEA/D. Hypervolume reaches 1.114 against 0.944, 0.912, and 0.628. IGD reaches 0.130. Epsilon-indicator reaches 0.167. Lowest objective values are recorded on tortuosity, threat, and smoothness. Robustness is confirmed across three environments of varied threat density and geometry. Phase-1 training requires 66.1 s; Phase-2 optimization requires 16.6 s. This study concludes that domain clustering governs Pareto quality, while reinforcement guidance governs convergence.