Shared control architectures for vehicle steering

Various schemes for sharing control between a human driver and automation system have been proposed, each with the aim of freeing attention while supporting smooth transitions of authority when the control challenge exceeds driver or automation capabilities. When sharing control of a vehicle, it can be expected that the driver develops internal models of the automation and its capabilities and assumes or assigns intent to its actions. In this paper, we develop system models of input mixing and haptic shared control to describe the communication channels open to the driver for monitoring automation behavior and determining automation intent. We pay particular attention to haptic (torque) feedback in the axis of steering that functions both to carry information and to couple the dynamics of the driver's backdrivable arms and hands with the dynamics of the automation system and vehicle. We assess the various feedback loops present for their promise to reduce cognitive load while maintaining situation awareness. An interpretation of the backdrivable biomechanics and automation impedance in terms of potential wells produces insight into the structured information available through haptic feedback and the internal models that the driver uses to predict automation behavior. The constructed models make explicit the additional communication channels open between human and automation in haptic shared control relative to input mixing.