Advanced Vehicle State Estimation: A Tutorial and Comparative Study
Jay A. Farrell, Paul F. Roysdon · IFAC-PapersOnLine · 2017
Autonomous vehicles require high bandwidth, high sample rate, precision vehicle and world relative state estimation, especially those such as automobiles that involve human safety with a mixed set of vehicles operating in a complex environment containing humans. An additional requirement is high reliability. Such systems will ultimately involve a multiplicity of sensor modalities. Sensor fusion is critical to achieving these application requirements. Several of the sensors (e.g., vision, radar, Lidar, ultrasound, Global Navigation Satellite Systems (GNSS)) have various spurious measurement types. Standard Extended Kalman Filter (EKF) approaches are not sufficiently reliable at removing the effects of such spurious measurements. The EKF approach must decide at the time each measurement arrives whether it is valid. If so, the measurement is used and discarded; otherwise it is not used and discarded. When that decision is wrong, either measurement information is lost or the state and covariance estimates are corrupted. Either situation can result in divergence of the EKF. An alternative is to maintain all recent measurement data within a moving time-horizon. This window of data can be processed within a Bayesian framework to extract the optimal state trajectory estimate over the time-horizon, under various fault scenario assumptions. Because the time window of data is maintained it is straightforward to change the assumptions as to which data are valid and reprocess the data. Therefore, in prior articles, this approach was referred to as a Contemplative Real-Time (CRT) estimator. It is closely related to Moving Horizon Estimation (MHE) and Simultaneous Localization and Mapping (SLAM). This paper has tutorial content explaining the interrelationships between the EKF, Iterated Extended Kalman Filter (IEKF), and CRT within the Bayesian framework; discussion of the fault detection procedures, and comparative experimental results.