A feasibility study on Three-dimensional potential field approach based on Gaussian Mixture Model for UAV collision avoidance

Nowadays, unmanned aerial vehicle (UAV) is increasingly popular and being utilized for various applications. This matter is naturally accompanied by the requirement for collision avoidance and path planning functions in the presence of surrounding obstacles to ensure safety, particularly for low-altitude or indoor flights. In this paper, we propose a three-dimensional artificial potential field (APF) approach for collision avoidance and path planning purposes, and its implementation is demonstrated. For these purposes, we develop a probabilistic method using multiple Gaussian distribution functions to mimic the shape of three-dimensional obstacles, an extension of the existing two-dimensional Gaussian mixture model (GMM) approach. The parameters of the GMM are updated by the Expectation-Maximization (EM) algorithm using obstacle data that can be measured using some sensors such as 3D Lidar. To demonstrate the performance of the developed algorithms, we use a quadcopter simulation model in Matlab/Simulink environment. Future research directions of the current method are also discussed.