Gaussian Mixture Model-Based Vector Approach to Real-Time Three-Dimensional Path Planning in Cluttered Environment

This work presents an obstacle-free three-dimensional (3D) path planning algorithm for unmanned aerial vehicles (UAV) navigating in cluttered environments. Gaussian mixture model (GMM), a class of unsupervised machine learning, is employed for environment perception based on a proposed vector approach to obstacle-free path planning. GMM circumscribes an obstacle by an ellipsoidal surface defined by its eigenvectors and eigenvalues. These eigens specify a region occupied by the obstacle so that the path planner re-routes the path whenever the UAV comes closer to this region. The proposed path planner is verified to bypass the challenges and limitations of computationally lightweight path planners such as artificial potential fields. The path planner is also compared with rapidly exploring random tree and shows better performance in optimal path and low computational time. The performance of the proposed path planner is validated in a simulated environment filled with obstacles of various sizes, shapes, and orientations.