Robustness assessment of automated sensor fusion system in LSAV

The emergence and the future deployment of automated vehicles (AVs) on the UK roads has given a boost to the research in the automotive industry to reach the milestones of the roadmap 2030 vision. The roadmap covered different types of vehicles on road, which will become with the full automation, the future vehicles on the UK road by 2030. A key gateway to the adoption of automation is related to quantifying robustness of the different categories of sensors integrated in the vehicles, and to the automated sensor fusion system that fuse the information of these sensors. Without this value, a less complete, reliable, accurate picture of the environment around them will be derived. In this context, this thesis investigated the assessment of robustness of the automated sensor fusion systems in low-speed automated vehicle (LSAV). Automated sensor fusion systems are the brain of the vehicles, and play an important role for the vehicle to drive safely around the obstacles. On light detection and ranging (LiDAR) solution-based, errors or inaccurate information may affect the normal behaviour of the vehicle and can cause hazardous situations on road. For this reason and due to the lack of methodologies regarding the propagation of faults on automated Sensor fusion system (ASFS), this thesis focused on the detection of faults from LiDAR inputs and recovery of the information. Two strategies of fusion, the first homogeneous with LiDARs only and the second heterogeneous with LiDARs and a stereo camera, have been designed using the probabilistic model of octree and enhanced with a voting scheme for fault detection and recovery of the information based on the confidence of the information estimated from the LiDARs. These strategies of fusion have been integrated in an observer-based solution to keep the independence of the fault detection system to the rest of the ASFS. Four observer-based solutions were proposed and compared, then assessed in scenario with fault injection. A fault injection methodology has been designed and both the observer-based and the testing methodology have been integrated in a study of functional safety to evaluate the decrease of threat in the presence of faults when a sufficient control action were taken. The results shown that a heterogeneous fusion with only three LiDARs helped with a stereo camera provided the performance in term of robustness and safety, and an optimum solution in term of cost