UAV Trajectory Planning in Complex 3-D Environments via Conditional Diffusion and Differentiable Sequential Convex Programming

UAV trajectory planning in complex 3-D environments remains challenging because environmental safety constraints and flight-feasibility requirements are strongly coupled. This paper proposes CADM-DSCP, a generation-optimization framework that integrates a constraint-aware conditional diffusion model with differentiable sequential convex programming. The diffusion model generates multiple environment-conditioned path candidates using terrain, obstacle, no-fly-zone (NFZ), and start-goal information, while DSCP refines them by repairing local constraint violations under safety-distance, pitch-angle, curvature, and smoothness constraints. The final trajectory is selected according to a comprehensive planning cost. Experiments on the CEC2022 benchmark functions and a constrained 3-D UAV planning scenario show that CADM-DSCP achieves competitive optimization behavior and generates trajectories with improved safety margin, smoothness, and planning cost compared with PLO, PSO, GA, DE, and GWO. Ablation results verify the contribution of the diffusion prior, constraint-aware loss, multi-candidate strategy, and DSCP refinement.