CO-PRE: A Multi-Sensor Fusion Framework with Visual-Inertial Navigation and Curvature-Optimized Path Planning for UAVs

Abstract

With the widespread application of unmanned systems in various fields, achieving high-precision autonomous navigation of unmanned aerial vehicles has become a research hotspot. Therefore, a navigation system that integrates vision, lidar, inertial sensor and barometer is designed in this study. Based on the factor graph optimization structure, the effective fusion of multi-source information is realized, thereby improving the consistency of positioning and mapping. Meanwhile, a probabilistic path planning method based on curvature constraints is proposed to reduce trajectory discontinuity and support dynamic feasibility requirements. The results show that the positioning error of the complete system is 0.43 m and the mapping completeness is 98.7%. Compared with the traditional A* algorithm, particle swarm optimization algorithm, and probabilistic rapid expansion random tree method, the improved algorithm reduces the path length by about 17.04% and shortens the planning time by about 16.71%. In the simulation test, its average energy consumption is 3.42 J/m, the average path deviation is 0.65 m, and the number of obstacle avoidance re-planning is 2.4 times/task. The research results show that the system has good operating performance in complex environments and has certain advantages over the baseline method in terms of efficiency, stability and obstacle avoidance ability.