Operator Decision Modeling in Cooperative UAV Systems

Abstract

Coupled operator-multiple UAV-tracking systems are modeled in a unifying framework using a Bayesian network with conditional dependencies between the elements. Both the UAV attitude/navigation system and the tracking system are easily integrated into the framework using a nonlinear estimator such as the Extended Kalman Filter. Discrete operator decisions are modeled as a Bayesian network block, with conditional dependencies on the UAV and tracking estimators. The Bayesian network block contains a series of random variables with softmax probability distributions, and a discrete selection random variable. Maximum likelihood optimization of the distribution parameters and block structure are shown to produce optimal estimates. The identified probabilistic network graph can be used in estimation form (update distributions based on new data) or prediction form (predict the probability of a decision based new data). Initial results are shown using a high fidelity multiple UAV simulator at the Australian Centre for Field Robotics.