A hybrid trajectory optimization solution applied to UAVs based on point cloud information and bio-inspired evolutionary algorithm

This paper addresses the challenges of trajectory optimization and flight safety for unmanned aerial vehicles (UAVs) in unknown environments caused by limited computational power of edge devices and restricted perception range. A trajectory optimization solution based on point cloud information and bio-inspired evolution optimization methods is proposed. The approach constructs a mathematical model through time parameterization and constraint modeling, then solves the trajectory using evolution optimization to enhance the safety and reliability of autonomous navigation. Furthermore, we develop an autonomous navigation system specifically designed for quadrotor UAVs to handle flight challenges in complex environments. Focusing on three key issues in UAV trajectory optimization for unknown environments, i.e., high computational cost of gradient information acquisition, elevated risks beyond perception ranges, and insufficient performance in complex scenarios, this work presents several innovative solutions. First, by integrating with the Ego-Planner framework, we design a gradient generation algorithm based on point cloud maps, which reduces replanning iterations by 33 % while improving success rate by 2.27 % and decreasing time consumption by 3.24 %. Second, we develop a bio-inspired trajectory optimization method by incorporating sensor perception range constraints and an improved beetle antennae search algorithm, significantly enhancing navigation safety in dense obstacle environments. Finally, we integrate these technologies to create a high-performance, lightweight autonomous navigation platform, whose feasibility and robustness are verified through multi-environment experiments. • This paper addresses the challenges of trajectory optimization and flight safety for unmanned aerial vehicles (UAVs). • A trajectory optimization solution based on point cloud information and bio-inspired evolution optimization methods is proposed. • A gradient generation algorithm is developed based on point cloud maps. • A bio-inspired trajectory optimization method by incorporating sensor perception range constraints. • A high-performance, lightweight autonomous navigation platform is developed and verified through multi-environment experiments.