Adaptive Fault Tolerant Control for Safe Autonomous Driving using Learning-based Model Predictive Control

This paper presents an adaptive fault tolerant control approach for autonomous vehicles (AV) under actuator or sensor faults to improve driving safety. A learning-based stochastic model predictive control (SMPC) strategy incorporating vehicle real dynamics characteristics is developed to realize accurate autonomous trajectory tracking. First, a vehicle dynamics model integrating typical actuator and sensor faults is established. Then, a model online learning strategy is designed to update the vehicle dynamics in real-time. Gaussian process (GP) is applied to identify and learn the real dynamic changes caused by faults which is hard to describe by standard models. Finally, the online learning vehicle dynamics is integrated into SMPC to optimize motion control for accurate trajectory tracking. Extensive simulations are studied to evaluate the model online learning performance and the safe tracking performance with adaptive fault tolerant control under various fault conditions.