Autonomous Navigation of Quadrotors in Dynamic Complex Environments

Abstract

This article introduces a novel framework utilizing velocity obstacles to enhance the autonomous navigation of quadrotors in dynamic complex environments. In this framework, quadrotors rely on onboard sensors to perceive the surrounding environment and construct an occupancy grid map for environmental representation. The density-based spatial clustering of applications with noise (DBSCAN) algorithm is employed to extract the positions and velocities of dynamic obstacles within the environment. Based on these results, we propose a velocity obstacle-based gradient field, called gradient velocity obstacle (GVO), for generating collision-free velocities ensuring safety. Compared with existing methods, GVO preserves the original feasible set while ensuring computational efficiency. Moreover, it exhibits excellent fault tolerance to environmental perception noise. Additionally, we design motion primitives based on B-spline parameterization. By optimizing within position and velocity state spaces, collision-free trajectories are dynamically constructed in real-time. Extensive simulations and experiments validate our framework's effectiveness, showcasing significant improvements in navigation efficiency and safety. The experimental section of the entire work can be found at the following link: https://www.youtube.com/watch?v=TOEeoFO4OxY.