Abstract
Collision avoidance in the presence of dynamic obstacles in unknownenvironments is one of the most critical challenges for unmanned systems. Inthis paper, we present a method that identifies obstacles in terms ofellipsoids to estimate linear and angular obstacle velocities. Our proposedmethod is based on the idea of any object can be approximately expressed byellipsoids. To achieve this, we propose a method based on variational Bayesianestimation of Gaussian mixture model, the Kyachiyan algorithm, and a refinementalgorithm. Our proposed method does not require knowledge of the number ofclusters and can operate in real-time, unlike existing optimization-basedmethods. In addition, we define an ellipsoid-based feature vector to matchobstacles given two timely close point frames. Our method can be applied to anyenvironment with static and dynamic obstacles, including the ones with rotatingobstacles. We compare our algorithm with other clustering methods and show thatwhen coupled with a trajectory planner, the overall system can efficientlytraverse unknown environments in the presence of dynamic obstacles.