Date of this Version
This thesis proposes a framework to include human cognitive process in making decision into control loop. Our approach in solving this problem is to add a complementary control which includes a model of human decision to the controller, capable of predicting human command accuracy in the near future. The outcome of decision unit can either be presented to the operator as a directive or may adjust the issued command toward better results. In order to construct human decision model, we combined tree approaches of bio-physical connectionist, mathematical abstraction and behavioral cognitive models in the adaptive gain theory framework. We first extended the classic DDM to include a layer that represent the change of strategy from weighted additive to heuristics and proved that such model is mathematically sound and well defined. Then by the help of adaptive gain theory, we showed that by collecting feedback signals from operator, we are able to predict her decision quality.
We also presented a prototype for a supervisory controller that includes the dynamics of making decisions by humans in the control loop. A two-stage model for strategy selection and decision making was utilized to cover a range of situations, in which the operators are required to make proper decisions. A controller was designed to dispatch the tasks between the system and the operator to keep the operator close to the best performance region. A case study for a simple one-attribute task was simulated to show the effectiveness of the proposed controller.
One major assumption in designing complementary control unit was that the feedback physiological signals can be mapped onto NE-LC gamma plane, hence the quality of decision at each time is known to controller. In the experiment section, we relaxed this assumption and showed that by using commercially available technologies, it is possible to infer the decision strategy and accuracy.
Adviser: Q. Hui