Sliding Mode Observer And Control-Based Approach For Longitudinal Fault-Tolerant Control Algorithm Of Autonomous Vehicles

This paper describes a sliding mode observer and control-based approach for a longitudinal fault-tolerant control algorithm of autonomous vehicles. To detect faults in a longitudinal acceleration sensor, the relative acceleration is reconstructed using a sliding mode observer. The relative acceleration is then used for probabilistic fault detection. When a fault is detected, the virtual acceleration is derived using the reconstructed value as an input for a lower level controller. In an upper level controller, the desired robust acceleration is derived when considering uncertainties such as the acceleration of the preceding vehicle, virtual acceleration, and the control perfounance. Actual driving data and a 3D vehicle model are used for a performance evaluation. It is expected that the algorithm can be applied as a fail-safe for autonomous driving.