LiDAR-Based SLAM in a 2D Simulated Environment

One of the best-known problems in robotics is simultaneous localization and mapping, where the robot tries to build the map of a place it navigates while simultaneously determining its configuration in that space. Different sensor modalities, e.g., LiDAR sensor(s), visual sensor(s), or their combination, can tackle this problem. Configurations of the robot over time are estimated using standard estimation techniques such as the extended Kalman filter or factor graphs. Unfortunately, one of the main issues during the robot’s movement is the drift, which occurs due to imperfect sensors and actuators. This can be diminished by deploying a loop closing algorithm that detects whether an already visited place has been observed. That being said, a SLAM implementation itself has a steep learning curve. Therefore, this paper provides an extensive overview of all the steps needed to implement such a system. Also, we provide the quantitative evaluation of our approach, where we deploy our state-of-the-art loop closing algorithm that diminishes drift. We showed how our loop closing approach improves the absolute trajectory error measure compared to the standard loop closing approach.