Evaluation of SLAM Algorithms for Highly Dynamic Environments

Abstract

Simultaneous Localization And Mapping (SLAM) has received considerably attention in the mobile robotics community for more than 25 years. Most SLAM algorithms have been developed for and successfully tested in static environments. Previous studies that investigated the use of SLAM algorithms in dynamic environments only considered partially dynamic environment in which only a few objects are non-static. In this paper, we evaluate several popular SLAM algorithms for use in highly dynamic environments in which all objects are only temporarily static, i.e. all objects will be moved within a short time frame. To this end, we built a static test environment and defined two different scenarios based on a warehouse environment to simulate highly dynamic environments. Four different 2D SLAM algorithms that are available in Robotic Operating System (ROS) are employed and evaluated through visual inspection of produced maps and the difference between the object positions in obtained maps and their real positions in the environment. Based on our conducted evaluation Hector Mapping achieves the best performance in both scenarios.