Integrated stochastic optimization and statistical experimental design for multi-robot target tracking

This paper presents an integrated approach for enhancing the performance of stochastic optimization processes by incorporating techniques from statistical experimental designs, such as response surface methodology. The two-stage process includes an “exploratory” phase, during which a fraction of the finite time budget is reserved for conducting informative measurements to best approximate the stochastic loss function surface, followed by execution of the optimization process for the remaining time. We formulate a representative stochastic optimization problem for the case of multiple distributed mobile sensors engaged in surveillance for one or more objects of interest. We show via simulation studies that the employment of such an exploratory phase, with the use of screening experimental designs to provide local approximations to the response surface, improves the stochastic optimization process.