Enhancing GPS Position Estimation Using Multi-Sensor Fusion and Error-State Extended Kalman Filter

An optimally designed GPS receiver can typically achieve 3-5 meters accuracy. This accuracy is insufficient for autonomous navigation where the margin of error must be low. Some existing sensor fusion approaches (IMU and GPS fusion using EKF) can achieve 2-3 meters (average error of 2.66 meters) accuracy, but the existing systems cannot function if the GPS receiver is not functional. Other modern techniques like Differential GPS are expensive to implement. Hence an accurate, cost-effective, and available alternative is required. As an alternative to Conventional GPS which typically has an accuracy of 3-5m which is not sufficient for autonomous navigation, a system that consists of data from LiDAR, IMU, and GPS has been implemented using sensor fusion with the help of error state extended Kalman filter. The implemented system was put to test using data synthesized from Carla Simulator. The proposed system was shown to have improved accuracy by reducing the position error by nearly 95%.