DEROA: A Differential Evolution Rolling Optimization Approach for Multi-UAV Trajectory Planning based on Spatial Grid Probability Map

Abstract. With large-scale UAV swarms used in wide-area inspection, multi-UAV cooperative perception faces core challenges like uneven target distribution and inefficient paths, causing blind planning with risks of redundant detection or misses. To address traditional method limitations in real-time target existence probability quantification and dynamic path optimization, this paper proposes the Differential Evolution Rolling Optimization path planning method (DEROA) based on a spatial grid probability map. Using a constructed target grid probability map, DEROA dynamically updates the paths as the probabilities of the grid evolve, in order to maximize the probability of general perception expectation of the multi-UAV system. The main innovations are as follows: (1) A probabilistic quantification grid map with multisource information fusion for inspection targets integrates historical trajectories, geographical obstacles, and real-time perception data. Dynamically updates grid target existence probabilities via Bayesian inference to direct UAVs to high-probability areas, addressing the deficiency of traditional modeling in representing dynamic target distributions. (2) The differential evolution-based rolling optimization cooperative algorithm combines DE’s global search capability with rolling horizon optimization’s real-time adjustment, achieving gridded dynamic path planning through distributed model redictive control. Experiments show that DEROA improves high-probability area coverage by 65.7%–106.6% and 1.0%–10.9% compared to traditional algorithms, with non-faulty UAV task coverage maintaining 0.44-0.97 under failure mechanisms, demonstrating strong robustness. (3) A dynamic reward function incorporating collision avoidance, communication constraints, and energy consumption, coupled with a path inflection point simplification algorithm reducing flight turns by 40%–60%, enables DEROA to achieve a maximum target discovery probability of 0.62 (3.3%–37.8% improvement), significantly enhancing perception efficacy in large-area scenarios.