Scan Matching and SLAM for Mobile Robot in Indoor Environment

Abstract

Autonomous indoor mobile robots are very promising application of robotics. In order to realize autonomous navigation, a robot that enters an unknown environment needs to reconstruct a consistent map of the environment and estimate its pose with respect to the map, simultaneously. This problem is well known as the Simultaneous Localization And Mapping (SLAM) problem, which has attracted a lot of interest from researchers in past few decades. The most popular approaches towards SLAM problem are usually developed based on the probabilistic methods, such as Extended Kalman Filter (EKF) SLAM, particle filter SLAM, and maximum likelihood SLAM. In recent years, a robust technology named “Scan Matching” plays a very important role in solving the SLAM problem. By matching sensor scans that are taken from different poses, the scan matching method can efficiently estimate the rigid transformation of the robot between two poses. Due to the fact that the exploring sensors are usually very accurate and robust, scan matching is very efficient for mobile robot to localize itself with respect to the given reference scans or maps. Although vision based approaches are getting more and more popular in SLAM research field, vision sensors are sensitive to the unpredictable variations of environment, such as the change of the lighting condition. Besides, most vision based solutions construct sparse feature points based maps which are not sufficient for robot autonomous navigation. Therefore Laser Range Finder (LRF) based scan matching method and fast indoor SLAM framework are still widely desired in consideration of the robustness of LRF towards environment changes. Another widely adopted sensor is Inertial Measurement Unit (IMU) which provides measurements of accelerations and rotating rates at the same time. In consideration of cost efficiency, Micro Electrical Mechanical Systems (MEMS) technology based IMU is preferable in consuming grade applications as well as robotic researches. However, the measurements of low-cost MEMS-IMUs are usually corrupted by various types of noises. Thus, a calibration work to compress noises is necessary before the usage of MEMS-IMU. The main contributions of this study are consisted of three main parts: Fast and Robust Scan Matching Approach: Various scan matching methods have been introduced. And the most widely used methods, Iterative Closest Point (ICP) and its variants, have been deeply investigated and modified to obtain better performance. New preprocessors as well as new distance metric for association process have been proposed to enhance the robustness of ICP. A new framework of incremental scan matching has been developed to fulfill the scan matching task in large indoor loops. Line-segment based EKF-SLAM with Slope and Edge Detection: In this contribution, planar line-segment based EKF-SLAM has been well investigated at the very beginning. Then, the slope and edge in structured indoor environment have been modelled to line-segments and merged into the EKF-SLAM framework. In order to overcome…