Fusion-based localization for personal mobility vehicles using INS, GPS, and RTAB mapping

Precise localization is crucial for autonomous vehicle technology, enabling optimal operation with minimal human input. This study develops and evaluates a cost-effective localization system for personal mobility vehicles (PMVs) using a sensor fusion approach that integrates an inertial navigation system (INS)—derived from wheel odometry and an inertial measurement unit, IMU—a consumer-grade global positioning system (GPS), and real-time appearance-based mapping (RTAB Mapping). The extended Kalman filter (EKF) algorithm is utilized to combine these sensory data, addressing challenges such as GPS signal degradation, lighting variations, and odometry drift. In the proposed EKF framework, wheel odometry and IMU data provide high-rate motion predictions, while GPS and RTAB Mapping observations are used for corrective updates, depending on their availability. The EKF hyperparameters are tuned according to each sensor’s noise characteristics to optimize state estimation. Field experiments using the SEATER (single-passenger electric autonomous transporter) have demonstrated the system’s performance in diverse real-world conditions, including open fields, GPS-obstructed environments, multipath zones, and varying lighting scenarios. The experimental results indicate that GPS alone performs well in open areas but suffers in obstructed or multipath environments. Meanwhile, RTAB Mapping alone is highly sensitive to illumination changes. The proposed fusion of INS, GPS, and RTAB Mapping consistently outperforms individual sensors and other fusion configurations, reducing the root mean squared error (RMSE) to around three meters. This performance is achieved with low-cost sensor suit, showing the trade-off between cost and performance. The proposed fusion method also shows dependable, long-term localization in various environments, indicating its potential for practical PMV applications.