In this paper, a solution is presented to address the sensor fault detection and isolation (FDI) problem in state estimation for autonomous vehicles (AVs). The primary impetus for autonomous driving lies in its potential to ensure vehicle safety, a goal that requires an accurate determination of location, heading, and speed. Although sensors can directly obtain these measurements, they are often affected by noise and disturbances with unknown but bounded (UBB) distributions. To mitigate these effects, state estimation techniques are commonly employed, leveraging sensor fusion. This work aims to design an FDI methodology that continuously evaluates the accuracy of the state estimation algorithm in an AV. In order to achieve this goal, various observation techniques for robust FDI are compared, including a novel approach of EKF formulated within the LPV framework, named LPV-EKF. A zonotopic LPV-EKF observer is implemented to perform FDI on both state estimation inputs and outputs, considering an UBB noise distribution. The proposed methodology for the identification of anomalies is optimised to minimise the detection time in real world scenarios. The experimental results for FDI, collected from an autonomous Renault Zoe (SAE Level 3), are analysed and discussed.

Peer Reviewed