ORB-SLAM, IMU and Wheel Odometry Fusion for Indoor Mobile Robot Localization and Navigation

Abstract

In this paper, we propose a light-weight multi-sensor fusion method containing ORB-SLAM, IMU and wheel odometry for localization and navigation of an indoor mobile robot in GPS-denied environment. Known as an accepted generally visual simultaneous localization and mapping (SLAM) system, ORB-SLAM based on feature matching computes real-time camera pose. The Inertial Measurement Unit (IMU) measures the angular velocity of the robot by one of its gyroscopes. The wheel odometry provides linear motion velocity for the robot and records distance the robot has moved. Through leveraging both rotation characteristic of IMU and linear characteristic of wheel odometry, the rough localization estimation for the robot is obtained. During every navigation of the robot, the rough localization estimation provides relatively accurate mapping scale of the real world for ORB-SLAM. And the mapping scale revises the monocular camera pose of ORB-SLAM to obtain global robot pose estimation in the real world. In the experiment, the robot can locate itself with tolerable error and perform great navigation ability in a specific scene.