Vision Augmented State Estimation with Fault Tolerance

Obtaining accurate and reliable measurement data is really crucial for any vehicle system, especially if the system deals with maintaining safe operations of the vehicle. Conventional direct methods of obtaining such measurements by using sensors like GPS, INS, and Wheel Encoders are not reliable. The raw measurement obtained from these sensors is accompanied with bias and noise. Also, these sensors tend to underperform in certain conditions and faults, which adds to the inaccuracy and uncertainty of a system. Many different approaches exist, which intelligently fuse information from multiple sensors in order to produce reliable estimates. However, when used independently, these approaches suffer due to faults like plausibility faults, frequency faults, and error code faults, which can occur at the sensor level. This thesis therefore takes into account the development and implementation of an architecture which merges a planned fault tolerance approach to an intelligent state estimation process. The main contribution of the developed prototype is that it demonstrates the effectiveness of the developed architecture in mitigation of various faults that occur at the sensor level. This prototype lays the ground for the use of VO as an alternate source of measurement to the conventional sensors. Starting from the implementation of sensor data specific integrity checking to the implementation of Extended Kalman Filter, fault tolerance methods have been employed at different levels. A 3 DoF dynamic vehicle model has been used to enable the Extended Kalman Filter predictions. Finally, test cases were devised to validate the effectiveness of the prototype.