Multi-objective trajectory planning for air-ground collaborative logistics UAVs oriented towards dynamic logistics demands

Abstract

To address scheduling failures and collision-avoidance safety issues caused by the coupling of unexpected orders and complex three-dimensional threats in dynamic low-altitude logistics delivery, this study investigates a bi-level multi-objective trajectory planning problem that coordinates cloud-based global scheduling with UAV-side local collision avoidance. A Dual-Track Collaborative Architecture (DTCA) for dynamic trajectory planning is proposed. The architecture establishes a global–local dual-track multi-objective trajectory evaluation model for dynamic demands and develops a three-dimensional spatiotemporal corridor-based collaborative scheduling mechanism supported by a Ground-based Air Traffic Control (G-ATC) platform, thereby enabling closed-loop coordination between cloud-based resource optimization and onboard collision avoidance. A FAST-SPEA2 global dynamic scheduling algorithm incorporating a fuzzy inference mechanism is developed to improve replanning capability in response to unexpected orders, while a multi-subpopulation balanced HMOWOA algorithm is designed to enhance local trajectory optimization and safe collision avoidance in complex built environments. Experimental results demonstrate that the DTCA and its core algorithms achieve significant advantages in dynamic order response, global scheduling optimization, and local collision avoidance. Specifically, compared with the UAV-side HMOWOA-only scheme, the dual-track collaborative architecture reduces the average global fitness value by 12.62%. Compared with MOPSO, the second-best performer among 11 benchmark algorithms, FAST-SPEA2(DTCA) reduces the global mean fitness by 47.89%, with the statistical significance of the performance differences confirmed by Wilcoxon rank-sum tests across the evaluation metrics. The HMOWOA algorithm further achieves an average reduction of 1.46% in the local mean fitness compared with seven benchmark algorithms. In addition, ablation experiments show that the complete FAST-SPEA2 and HMOWOA algorithms achieve superior overall performance compared with their respective baseline algorithms and four single-strategy variants (S1–S4). These results demonstrate the effectiveness of the proposed DTCA in dynamic demand response, global resource scheduling, and safe collision avoidance in complex low-altitude environments.