Safe Transitions to Manual Driving From Faulty Automated Driving System
This thesis presents a method to assess the safety of transitions from automated to manual driving when vehicle automation fails. The method is based on contributions to the understanding of human driving behavior, also presented in this thesis. Interviews with drivers and driving simulator studies of driving with automation, and particularly analyzes of situations where automation failed provided the base for the proposed method. Among the results of the studies, it was found that drivers were more likely to control an automation failure if automation was only replacing the driver in longitudinal control of the vehicle, i.e., steering still managed by the driver. Moreover, the studies found that drivers responded to the failures with varying success. For the most critical failures, almost half of the drivers collided, while for a less critical failure, about two thirds of the drivers managed to control the situation and avoid a collision.
Individual differences between drivers were considered to have contributed to the varying success to control automation failures. The proposed method for assessing the safety of transitions therefore adapts online to the individual driver. While the vehicle is driven manually, the driver's capability to control the vehicle is estimated and described as a subset of the vehicle's state-space. In the event of an automation failure, the proposed method assesses whether vehicle states are within this subset or not. If vehicle states are within the subset, the driver is deemed capable of taking over, and the transition to manual control is classified as safe.
The method has been evaluated on data from real vehicles, with human drivers, to demonstrate its performance. Results indicate that the proposed method correctly classifies transitions as safe or unsafe.
HC1, Hörsalsvägen 14, Chalmers University of Technology
Opponent: Prof. Mauro Da Lio, Department of Industrial Engineering, University of Trento, Italy