SLAM in dynamic environments via ML-RANSAC

Simultaneous localization and mapping (SLAM) in dynamic environments is an important problem in robotics navigation, yet it is less studied. In this paper, we present a novel approach to segment and track multiple moving objects in real dynamic environments. Detected objects are classified into stationary and moving objects using a state-of-the-art method referred to as multilevel-RANSAC (ML-RANSAC) algorithm. The algorithm is designed to track moving objects in conflict situations while running SLAM. The ML-RANSAC algorithm is developed to robustly estimate velocity and position of the multiple moving objects in an unknown environment whereas the state of the objects (static or dynamic) is not known a priori. The main characteristic of the algorithm is its ability to address both static and dynamic objects in SLAM and to detect and track moving objects (DATMO) without dividing the problem into two separate parts (SLAM and DATMO). We apply the proposed algorithm on two sets of simulated data to validate its performance in situations where the objects are either occluded or placed in dense dynamic scenario. We have compared our method with the true data via simulation studies. Furthermore, we have implemented the algorithm on a Pioneer P3-DX mobile robot navigating a real dynamic environment. Simulation studies as well as real-time experiments suggest that the algorithm is able to track and classify objects accurately while performing SLAM in dynamic environments.