Analysis and implementation of 2D slam algorithms for indoor mobile robots

Abstract

In this research a comparative analysis and implementation of Simultaneous Localization and Mapping (SLAM) algorithms for indoor mobile robots has been conducted, the purpose of which was to identify a possible area of improvement for future implementations. Robot are widely being used for daily tasks and as such require a technique to be able to easily navigate indoor environments, specifically small maps for helper robots, however simple methods such as dead reckoning and wireless beacons are unreliable. To resolve this problem a new method is being researched upon which is SLAM. SLAM algorithms which are commonly known, accepted by experts in robotics and are compatible with the Robot Operating System (ROS) have been selected for this study. They are GMapping, Hector SLAM and Karto SLAM. Each method is comparatively analysed with various parameters such as computational complexity which is a measure of how complex the algorithm is with respect to the amount of steps and calculations needed, Central Processing Unit (CPU) usage load which measures how much the processor is being used and scenario-based efficiency which takes into account the performance based on different scenarios such as different speeds, map sizes and obstacles. Additionally, the analysis and selection is based on validation techniques such as Corner Detection & Matching, K-Nearest Neighbours and Map Completeness conducted on simulated data. Based on the comparative analysis performed via MATLAB, Hector SLAM was found to be the most appropriate technique for indoor robots in small maps. This technique was implemented on the robot platform and real-world tests conducted. The chosen method i.e. Hector SLAM managed to successfully generate fully visible maps in small indoor locations with numerous features. However, maps with larger sizes or smoother features were not as efficient.