A Comparative Study of SLAM Algorithms for Indoor Navigation of Autonomous Wheelchairs

Abstract

Following the rapid research and development in the field of autonomous automobiles robots are now being developed to be human assistants. Here, the focus of autonomous wheelchair systems is for patients with rare diseases, e.g., Motor Neuron Disease (MND); like Amyotrophic Lateral Sclerosis (ALS). The goal is to develop an autonomous wheelchair that helps these patients navigate indoors safely and securely. Such robots follow a common protocol, which was inspired by humans: perception, thinking, and reaction. In the world of the autonomous wheelchair navigation rules are dynamically generated in real-time by means of machine learning or deep learning. Simultaneous Localization and Mapping (SLAM) makes an autonomous robot aware of its position in the environment, while simultaneously building a map of its environment. This paper compares five SLAM algorithms: Gmapping, Hector, RTAB-Map, VINS-Mono, and RGBD-SLAM, for building indoor navigation maps. An electric powered wheelchair was modified to incorporate LiDAR, Kinect camera, IMU, and wheel encoders. The wheelchair was made functional for testing and comparing the defined metrics for localization, mapping accuracy, and its usability for patients with differing medical conditions. RTAB-Map was found to be the most scalable algorithm, one that works with different combinations of sensors. This algorithm produces accurate maps while accurately estimating robot location consistently, and for a range of speeds.