Value-Based Tasking Controllers for Sensing Assets

Sensing in support of combat teams will be increasingly performed by UAVs. With the accelerating pace of modern combat operations, and the developing networked-management of such, the problem of how to task sensor assets is entering the realm where automated decision-support tools should be applied. The objective by which the relative value of possible future sensor tasks should be compared is the value that they bring to the combat operation, with adjustment by the possible costs of these sensing operations. A mathematical theory allowing for the determination of the control-problem value as a function of information state (where this information state is described by a probability distribution) has recently been obtained. Possible future sensing tasks are mapped into the (stochastic) observation outcomes, and these are further mapped into potential a posteriori probability distributions. The value to the combat operation is the expectation of the minimax value as a function of these potential future probability distributions. This is used as an objective function, according to which optimal sensing-platform tasking is computed. Both openloop and observation-feedback sensing-platform tasking controllers are developed. Importantly, a new, and quite general, representation of the form of such a value function is obtained. The value takes the form of a pointwise minimum of linear functionals. Further, this form is retained through dynamic programming backward propagation for Research partially supported by AFOSR grant FA9550-06-1-0238 and NorthropGrumman. All authors are with the Dept. of Mechanical and Aerospace Engineering, University of California San Diego, San Diego, CA 92093-0411, USA. [email protected], [email protected], [email protected] solution of the problem. This representation will allow for the solution of high-dimensional problems in tasking of sensing UAVs. A simple, illustrative example is shown.