Abstract
Real-time global localization of a mobile robot is a prerequisite to accomplishing the planning and navigation tasks. Adaptive Monte Carlo Localization (AMCL) algorithm is broadly applied for mobile robots' localization in indoor environments. The localization accuracy of the AMCL is limited due to the irregularity of the laser sensor, the random nature of particle sampling, and the problem of final pose inference. This paper addresses the issues of global localization in planar indoor environments by presenting a feature-based technique that synergizes the capabilities of an RGB-D camera and 2D LiDAR. We proposed a novel method by customizing the Extended Kalman Filter (EKF) using Encoders and Inertial Measurement Unit (IMU) data for initial pose prediction and subsequent correction with RGB-D and LiDAR observations. Our approach incorporates robust state vector fusion, considering sensor malfunctions for improved accuracy. The developed system is evaluated using a mobile service robot, and the results demonstrate enhanced accuracy, computational efficiency and robustness over state-of-the-art techniques.
DOI: 10.61416/ceai.v26i4.9123