A Probabilistic Octree Fusion Model for Analytical-Based Observer Fault Detection in LSAVs: 2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring)

Recently, there has been a considerable improvement in low-speed autonomous vehicles (LSAVs), which will function key roles in future intelligent transportation systems. To be successfully distributed on a real road, these vehicles must have the ability to drive autonomously along collision-free paths whilst flowing traffic laws. LSAVs use Lidar sensors to avoid obstacles in its path. However, Lidar sensors have unreliability limitation, which consequently any decision made by sensors alone is insufficient and has let to serious accidents. This is because the difficulties to determine in the sensor fusion system, how wrong information can affect the decision made by the vehicle. In this paper, an observer system is present for fault detection of automated sensor fusion system for a LSAV, which functions based on octree fusion. Through this study, an analytical observer processing the information obtained by physical redundancy and an octree fusion process based on a probabilistic model of occupation of the voxels. This method shows that the decision made by the vehicle is more accurate than the existing system especially when a sensor sends incorrect information to the sensor fusion system.