SimSLAM 2D: A Simulation Framework for Testing and Benchmarking of two-dimensional Visual-SLAM Methods

摘要

Localization proves still a challenging task in robotics, with Visual-SLAM being a both powerful and popular solution. In Ground-SLAM, a branch of Visual-SLAM that uses groundplane views, there is still a lot to explore but the tools for testing new methods need improvement. In this paper, we present a novel framework for simulating the motion of Visual-SLAM methods over a given groundplane. It can also be used to benchmark different methods on a given dataset. We present the functioning of our system, as well as the requirements for a similar system to work. In order to showcase the capabilities of SimSLAM 2D, we present a test case of different trajectories on a smooth concrete terrain comparing the performance of two methods, StreetMap and ORB-SLAM. We also test StreetMap in localization only mode with the purpose of showing the visualization tools. We thereby show the ease of simulating and testing a Visual-SLAM method on a 2D environment in our framework, which represents a useful tool when developing new Visual-SLAM methods.