S. N. Balakrishnan, University of Missouri-Rola, USA

Abstract: From the time when Newton was challenged to solve the Brachistochrone problem in 1696 to this day, mankind has been fascinated with the field of optimization and more recently with optimal control. Although problems in many disciplines could easily be posed in an optimal framework, finding tractable and implementable solutions is not an easy task. Approximate Dynamic Programming(ADP) offers a computationally feasible framework and an'adaptive critic' neural architecture presents a feedback controller synthesis.

This talk will start with the dual network adaptive critic structure and its evolution to the single network adaptive critic(SNAC) formulation. Solutions to infinite-time optimal control problems(regulators, tracking problems) and finite-time optimization problems(path planning, minimum time problems) and advantages of using the adaptive critics will be discussed. Applications will include vibration control, agile missile trajectories, helicopter control, temperature profile control at high speeds and inventory control. The results shown are drawn from simulations and implementations and the underlying math models are ordinary, partial differential equations and Markov processes.

Current research on extensions of the adaptive critic concepts to systems driven by impulse control will also be discussed at the end.