GCMVF-AGV: Globally Consistent Multiview Visual–Inertial Fusion for AGV Navigation in Digital Workshops

An accurate and globally consistent navigation system is crucial for estimating the positions and attitudes of automated guided vehicles (AGVs) in digital workshops. A promising navigation technology for this purpose is tightly coupled visual–inertial fusion, which offers advantages such as quick response (QR), absolute scale, and accuracy. However, existing visual–inertial fusion systems have limitations, including long-term drift, tracking failures in textureless or poorly illuminated environments, and a lack of absolute references. To create a reliable and consistent AGV navigation framework and correct for long-term drift, we have designed a novel framework, globally consistent multiview visual–inertial fusion for AGV navigation (GCMVF-AGV). This framework uses a downward-looking QR vision sensor and a forward-looking visual–inertial sensor together to estimate AGV poses in real time. The downward camera provides absolute AGV positions and attitudes with reference to the global workshop frame. Furthermore, long-term visual–inertial drift, inertial biases, and velocities are periodically compensated between spatial intervals of QR codes by minimizing visual–inertial residuals with the rigid constraints of absolute poses estimated from the downward visual measurements. We have evaluated the proposed method on the developed AGV navigation platform, and experimental results demonstrate the position and attitude errors of less than 0.05 m and 2°, respectively.