In human-machine cooperation, machines may assist operators in a variety of ways. This paper discusses the coordination of various assistances on cognitive basis through a PID-based control approach. Cognitive assistance can be viewed as a two-dimensional problem. The question of when to provide assistance can be viewed as a control problem, and the question of what assistance to provide can be viewed as an interface problem. This research proposes pairing cognitive engagement level and performance relevant situation criticality with a PID control approach to determine the appropriate moment to provide proper assistance. Based on the stage of human cognitive processing, the interfaces of cognitive assistance are grouped into three levels: soft aid, soft intervention, and hard intervention. This paper took driving assistance as an exemplary application to validate the approach of cognitive assistances coordination. In the experiment, an intelligent machine driver monitored drivers’ real-time performance by measuring the time headway to front obstacles and the lateral deviation to lane center. Simultaneously, it monitored drivers’ cognitive state by measuring the eye movement with an eye tracker. With five sessions of driving, coordinated cognitive assistance was compared with no aid, soft aid, soft intervention, and hard intervention, respectively. The experimental results confirmed that coordinated cognitive assistance is the most effective approach to assist both primary and secondary tasks. It also proves to be a more enjoyable and less obtrusive assistance system when compared to other individual types of assistance. In addition, coordinated cognitive assistance can be extended to other real-time control relevant tasks.