Visual-inertial-wheel SLAM with high-accuracy localization measurement for wheeled robots on complex terrain

Visual-Inertial SLAM methods have demonstrated remarkable results using low-cost sensors. However, these methods often perform poorly in challenging scenarios like complex 3D and slippery terrain. To address this, a visual-inertial-wheel SLAM for wheeled robot on complex terrain is proposed. To improve localization accuracy on 3D terrain, a model for fusing Inertial Measurement Unit (IMU) and wheel odometry is proposed, which is based on an assumption that the movement of robots on ground with gradient changes can be approximated as motion on an arc-terrain during a very short time. Moreover, to exclude abnormal wheel odometry measurements on complex terrain, a wheel odometry anomaly detection method with adaptive thresholds is proposed. Additionally, the proposed SLAM system tightly couples vision, IMU and non-abnormal wheel odometry measurements in local and global Bundle Adjustment (BA) to estimate the pose of robot with high-accuracy. Corresponding experiments demonstrate the localization accuracy of trajectory is significantly improved.