Introduction
Autonomous driving systems (ADS) have been improving their capabilities during the recent years, being able to handle more scenarios in a safe manner. Nevertheless, this kind of systems still require human intervention in some situations where the complexity of the scene is high. By using a traded control architecture, a supervisor module may decide which agent has the authority over the vehicle (ADS or human operator) in a given driving scene. This approach is suitable for situations where the current agent is not able to handle the vehicle safely anymore.
The shift of control (SOC) is a crucial task of any traded control system that may increase the risk of collision if it is not performed correctly, since the agent who is receiving the responsibility has to be ready and aware of the driving context before taking control; this still remains one of the greatest challenges for assisted technologies in automobiles [2]. The interaction between the human operator and the ADS allows to perform a safe SOC and make safer decisions, since the ADS can perceive the driver status and preferences and the human operator can understand the goals and capabilities of the ADS [3].
This work presents a traded control architecture where the ADS estimates the complexity level (CL) of the current scene in real time and computes a proper level of driving automation (LoDA). Then, it decides how involved the human operator must be in the driving task. The system uses a graphical user interface on board of the ego-vehicle to inform the LoDA and the required involvement level (RIL) of the human driver.