Maximum likelihood principle based adaptive extended Kalman filter for tightly coupled INS/UWB localization system

In the indoor Inertial Navigation System/Ultra-Wide Band (INS/UWB) tightly coupled navigation system, the filtering performance of the extended Kalman filter would be degraded when the statistical characteristics of system noises are unknown or inaccurate. To solve this problem, a novel adaptive EKF based on maximum likelihood principle is presented. Firstly, a two-dimensional kinematic model of the test vehicle is established, and a third-order Auto-Regressive model is introduced to model the noise of the low-cost inertial measurement unit. According to the MLP, an objective function consisting of measurement noise matrix and predicted error covariance matrix is constructed. Then, the problem of online estimation of system noise statistic is transformed into optimizing the objective function, which is iteratively computed by the expectation maximization technique. Subsequently, the AEKF with a time-varying noise estimator is presented. Finally, an indoor test vehicle motion platform is built. Experiment results demonstrate that, compared with the classical EKF, the positioning accuracy of AEKF is improved significantly under the two conditions of pedestrian interference and obstacle interference, which shows the promising potential of the proposed AEKF in improving positioning accuracy and robustness of the INS/UWB system.