Fail safe control for automated driving: design and experimental validation

Nowadays most automotive companies are investing in the field of autonomous driving with the final goal of achieving full independence from human drivers. To accomplish this objective, it is necessary for the ego vehicle controller to handle failure situations. Ensuring the safety of the driver and other agents in the environment indeed requires the main logic to drive the vehicle not only under normal conditions but also in those situations where the controller must ensure avoidance of zones posing safety risks. The proposed algorithm has a pyramidal structure. The high-level component consists of a decision-making algorithm with the objective of computing the optimal state trajectory which satisfies the nominal constraints and ensures the existence of an emergency maneuver capable of avoiding risk zones, regardless of the initial conditions in which the failure occurs. To achieve this objective, a set-based method is adopted to retrieve the Avoidance and Constraints admissible sets in which the state of the vehicle should lie to meet the previously defined requirements. Membership in such regions is guaranteed by imposing, during nominal driving, additional constraints defining the boundaries of these sets. When the failure occurs, the target of the chosen maneuver, computed at the previous iteration, is exploited by the emergency maneuver controller to generate the emergency trajectory. To simplify this procedure, each controller is designed to guarantee the stability of the closed-loop system and unitary output gain with respect to the target. Once the high-level optimal trajectory is computed, the low-level controller solves an optimal control problem to derive the optimal input required by the PID controllers to drive the vehicle. The capabilities of the proposed solution are first tested in a simulation environment and then on a real test bench to check the algorithm's robustness to disturbances and delays that are present in the real world.