Decision-Theoretic Control of Unmanned Systems

Toyon Research Corporation proposes to develop a system that allows a user to control multiple unmanned systems (UMS) of varying levels of autonomy. We will use Dynamic Decision Networks (DDNs) to create decision-theoretic agents that can make rational, optimal decisions on their own. DDNs are an extension of the highly-successful Dynamic Bayesian Networks (DBNs) used for developing a world view by fusing sensor measurements with a priori intelligence and contextual information. DBNs and DDNs are graphical models which provide the ability for the UMS to explain its inference and reasoning to the warfighter supervising them. These cognitive architectures naturally provide the ability to update their internal world state as they receive new information about a dynamic environment and change their plans accordingly. In Phase I, we will design a DDN-based system that can interact with a human supervisor. This will involve constructing the networks, specifying the necessary probabilities, and developing utility functions for varying levels of autonomy. We would focus the Phase I option on identifying a graphical means by which UMS explain their actions to users and use Phase II to implement the Phase I design in a software prototype for the purposes of testing, demonstration, and evaluation.