Data fusion for autonomous vehicle navigation based on federated filtering

To improve the position estimation accuracy of a four wheeled vehicle, the performance of federated filter structure is investigated. The complete nonlinear model of the vehicle resembles a real car which is used for position estimation. The vehicle is equipped with an accelemeter to measure acceleration, a tachometer for measuring angular velocity of the front wheel, a compass to measure the heading angle and a potentiometer to measure the front wheel steering angle.The federated extended Kalman filter (FEKF) with constant information sharing coefficients has been applied to the vehicle dynamical model and it is shown that with inexpensive sensors, accurate positioning can be achieved. Observability test indicates that vehicle model is not fully observable and centralized Kalman filter implementation leads to nonoptimal result. Simulation results illustrate higher precision and better reliability of FEKF with respect to centralized filter in this case study.