Proposal and Verification of the Algorithm for Judging Driving Stability Based on Driver, Vehicle, Environment Data for the Take-over of Autonomous Vehicle

A lot of research and development for the autonomous vehicle have been progressed recently, and it is about to be commercialized. In specific situations, however, the vehicle needs to be controlled by human and the driving mode must be switched from autonomous to manual. At the moment, it is difficult to have driving stability immediately because drivers were doing other tasks in the autonomous mode. For safety issue, driving stability after the take-over should be recovered quickly, and it is necessary to monitor driving stability in real time. In this study, we propose an algorithm for judging driving stability quantitatively using the data of drivers, vehicles, and the environment. We suggest one class classification approach based on convolutional auto encoder. And also, the numerical experiments are conducted to compare the results between the proposed algorithm and a rule-based algorithm, so we verify the effectiveness and rationality of the algorithm.